We’re entering a new era of AI-powered business applications, and today we’re excited to publish the 2026 release wave 1 plans for Microsoft Dynamics 365, Microsoft Power Platform, and role-based agents in Microsoft 365 Copilot, outlining a broad set of capabilities slated for release between April 2026 and September 2026. These updates reflect our ongoing commitment to making AI an essential partner in how organizations operate, innovate, and grow.

Dynamics 365 leads this wave with AI-powered, agentic innovations across sales, service, finance, supply chain, human resources (HR), and commerce—helping organizations unify data, automate processes, and elevate customer and employee experiences. Microsoft Power Platform continues to expand modern app development, intelligent automation, and enterprise-grade governance to empower makers and developers to innovate with confidence. Role-based agents in Microsoft 365 Copilot further evolve into intelligent daily command centers, helping to deliver richer, data-grounded insights and extensibility that help teams work smarter across every role.

To help you stay current on the most important and innovative capabilities, we’re moving beyond bi-annual launch events to lighter, more frequent business applications updates, featuring expert insights and demonstrations from Microsoft product leaders and engineers.

• Watch on demand: Dynamics 365 Business Applications Update
• Register for the Power Platform and Copilot Studio update April 15 at 9 AM PDT

Be sure to stay updated on the latest features and create your personalized release plan using the release planner.

Highlights from Dynamics 365

2026 release wave 1 updates for Dynamics 365 deliver AI-powered, agentic experiences across sales, service, finance, supply chain, commerce, HR, projects, sustainability, and enterprise resource planning (ERP)—bringing deeper Copilot integration, intelligent automation, unified customer and operational data, and enhanced cross-app capabilities to help organizations drive efficiency, elevate customer and employee experiences, and operate with greater agility and confidence.

Dynamics 365 Sales

Dynamics 365 Sales brings the power of AI to help sellers build their pipeline, enrich opportunities, and accelerate deal closure, while helping sellers easily access accurate, up-to-date information and recommending high-impact actions that sellers can take. Copilot experiences in Dynamics 365 Sales can draw on data spanning customer relationship management (CRM) and Microsoft 365 signals, like email and meeting recaps, to deliver actionable insights across Dynamics 365 and Microsoft 365 experiences.

Dynamics 365 Customer Service

Dynamics 365 Customer Service will continue to enhance agentic capabilities across case management, email, customer intent, quality evaluation, and knowledge management. AI-infused admin and supervisor help to provide more transparency and quicker time-to-value. These investments strengthen end-to-end service orchestration, from helping identify customer intent to driving autonomous workflows that elevate service quality and responsiveness.

Dynamics 365 Contact Center

Dynamics 365 Contact Center advances the agentic contact center in 2026 release wave 1 with new AI-powered capabilities that improve self-service, support accelerate assisted service, and help organizations run contact center operations more intelligently in 2026 release wave 1. It expands to include emerging channels, supervisor insights, and extensibility, giving organizations a unified, AI-powered system to elevate the customer experience.

Dynamics 365 Field Service

Dynamics 365 Field Service strengthens service execution across technician productivity, resource scheduling, and work order management. Investments focus on mobile usability and reliability, intelligent scheduling through the Scheduling Operations Agent, and end‑to‑end execution across assets, projects, and financial operations in this release wave. Together, these updates help organizations manage service complexity and deliver consistent service outcomes.

Dynamics 365 Sustainability

Dynamics 365 Sustainability introduces more intuitive reporting navigation, advanced calculation versioning, and granular data‑locking to reinforce governance and regulatory confidence in this wave. Expanded finance integration, streamlined workflows, and updated templates and factor libraries will further empower organizations to make informed decisions and support progress toward their sustainability goals.

Dynamics 365 Finance

Dynamics 365 Finance delivers continued global scale enhancements that drive greater financial automation, strengthen global regulatory compliance posture, and enhance financial planning and analytics—helping organizations operate more efficiently and achieve their financial and operational goals with confidence.

Dynamics 365 Supply Chain Management

Dynamics 365 Supply Chain Management’s 2026 wave 1 enhances supply and demand planning with price-demand correlation and capacity-to-promise (CTP) date protection. Supplier communication and engagement are streamlined, while warehousing gains AI-powered picking, inventory rebalancing, and hands-free scanning—driving supply chain efficiency.

Dynamics 365 Project Operations

Dynamics 365 Project Operations brings rich capabilities in 2026 release wave 1—from change order support and smarter project planning to smoother quoting, budgeting, and contract workflows. New enhancements streamline item consumption, mobile expense management, subscription billing, and modern-architecture migration—delivering connected project experience.

Dynamics 365 Commerce

Dynamics 365 Commerce strengthens business-to-business (B2B) with multi-outlet ordering, unified sign-in, outlet-specific catalogs, and built-in credit management to help reduce friction and protect cash flow. It modernizes order management and assisted-selling workflows in retail stores, helping to improve associate productivity, and customer experiences across channels. It also enables cross-legal-entity inventory lookup and flexible, attribute-based pricing to help accelerate mass updates and help drive higher sales.

Dynamics 365 Human Resources

Dynamics 365 Human Resources continues to advance in areas such as recruitment, onboarding, reporting, and integrated workforce management. By merging enhanced user experiences with broader ecosystem integration and expanding regional payroll collaborations, the platform enables organizations to optimize employee engagement, support operational accuracy, and confidently achieve their workforce objectives.

Finance and operations cross-app capabilities

Finance and operations cross-app capabilities will introduce new enhancements that strengthen the foundation for AI experiences across Dynamics 365. These updates include improvements to Model Context Protocol (MCP) servers, as well as the general availability of immersive home, which is an AI-powered workspace designed to help users stay focused and prioritize what matters most.

Dynamics 365 Customer Insights – Data

Dynamics 365 Customer Insights – Data acts as the grounding layer for CRM copilots and AI agents, delivering real‑time, unified customer profiles that help power accurate decisions. With enriched data, teams can act on insights directly in their workflow to deliver timely, personalized experiences that deepen engagement and drive better outcomes. The result is an AI-ready data core that elevates agents and helps deliver more connected, intelligent CRM experiences.

Dynamics 365 Customer Insights – Journeys

Dynamics 365 Customer Insights – Journeys empowers end-to-end, agentic customer engagements across sales, marketing, and service, allowing businesses to proactively react to customer behavior using Copilot and AI agents. With smarter orchestration tools, teams can deliver impactful campaigns at scale to drive stronger relationships, higher efficiency, and revenue growth. Part of Dynamics 365, every interaction within your organization benefits from shared data and consistent intelligence across Microsoft CRM applications.

Dynamics 365 Business Central

Dynamics 365 Business Central accelerates the move to agentic ERP with enhancements to our AI‑powered agents that automate sales and purchase scenarios in 2026 release wave 1. Alongside new business capabilities, we invest heavily in developer productivity to support extensibility—improving advanced language (AL) testing, debugging, Copilot extensibility, and agent design.

Highlights from Microsoft Power Platform and Microsoft Copilot Studio

2026 release wave 1 updates for Microsoft Power Platform deliver modernized app experiences across Power Apps and Power Pages, AI-powered automation and agent innovation in Power Automate and Copilot Studio, enhanced Dataverse intelligence and programmability, and strengthened governance, security, and cost management capabilities to help organizations build, scale, and manage intelligent solutions with confidence.

Power Apps

Power Apps continues to modernize app experiences with a refreshed model-driven user interface (UI), improved mobile and offline capabilities, streamlined search, and expanded AI features. This release brings standardized modern theming to everyone, real-time Dataverse access for offline-first canvas apps, enhanced search in grids and lookups, and broader availability and extensibility of generative pages to help teams build and scale intelligent apps faster.

Power Pages

Power Pages will further empower pro-developers and low-code makers to build intelligent business portals for your employees, customers, citizens, and partners through better integration with market leading AI tools. Additionally, enhanced security agent features will further support low-code makers, pro-developers, and admins with actionable insights and abilities for securing their websites.

Power Automate

Power Automate is Microsoft’s comprehensive automation platform for cloud flows, desktop flows, and process mining. This release introduces AI agent authoring, optimization, and self-healing capabilities for desktop flows, Copilot Studio-powered actions in cloud flows, enhanced maker and collaboration tools across both, general availability of object-centric process mining, and consolidated governance reporting.

Microsoft Copilot Studio

Microsoft Copilot Studio continues its journey to make agent and agentic workflows even easier to build and more powerful. Now you can further customize agents built with Agent Builder in Microsoft 365 Copilot, and power your automation with high value AI actions. Deeper governance, multi-agent orchestration, and evaluations enable further scaling. With connections to Microsoft Foundry and Work IQ, your agents can use the latest AI technology in coordination with your organizational data.

Microsoft Dataverse

Microsoft Dataverse continues to invest in enterprise-ready agentic and low-code data platform capabilities. The spotlight is on Work IQ and Copilot integration, delivering organization-specific decisions with adaptive learning and full auditability. We’re also enhancing agent programmability with Dataverse APIs, MCP servers, and Python SDK, plus new storage management tools for enterprise-grade compliance at scale.

Microsoft Power Platform governance and administration

Microsoft Power Platform governance and administration introduces admin controls for agent security, real-time risk assessment in Copilot Studio, and AI-powered governance agents that automate tenant monitoring and remediation in this release. Enhanced visibility into usage patterns, granular Copilot credit consumption with pay-as-you-go (PAYG) caps, and connector dependencies help you optimize costs, demonstrate return on investment (ROI), and enforce compliance with organizational policies using features within the Power Platform Admin Center. GitHub integration and deploy from Git mature your application lifecycle management (ALM) practices with full audit trails.

Business Applications Update

Power Platform &

Copilot Studio edition

Register now

Updates to role-based agents in Microsoft 365 Copilot

2026 release wave 1 updates for Microsoft role-based agents transform Sales Agent and Finance Agent in Microsoft 365 Copilot into intelligent daily command centers, helping to deliver richer, data-grounded insights, enhanced chat and mobile experiences, contextual support across Outlook and Teams, and strengthened governance and extensibility to help organizations drive productivity and scale AI responsibly.

Sales Agent

Sales Agent becomes the seller’s daily command center with richer Sales Chat and Sales Home experiences across desktop and mobile in 2026 release wave 1. Sellers will gain streamlined access to deal and account insights through configurable record summaries, contextual support in Outlook and Teams, and improved email and meeting intelligence. New governance and extensibility controls will also help organizations scale AI responsibly.

Finance Agent

Finance Agent helps finance professionals and their stakeholders interact with financial information from their ERP within the flow of work. In 2026 release wave 1, we continue expanding how this financial assistant supports common finance tasks such as reconciliation, variance analysis, and data preparation in Excel, as well as customer communications in Outlook. By bringing financial insights and assistance directly into familiar productivity tools, the Finance Agent helps teams investigate issues faster, respond to stakeholders more efficiently, and spend less time manually preparing or reconciling data so they can focus more on financial analysis and decision support.

For a complete list of new capabilities, please refer to the Dynamics 365 2026 release wave 1 plan, the Microsoft Power Platform 2026 release wave 1 plan, and role-based agents 2026 release wave 1. We also encourage you to share your feedback in the community forums for Dynamics 365 and Microsoft Power Platform.

Business Applications Update

The Business Applications Update offers an early preview of new capabilities coming in the months ahead. This refreshed structure is designed to reflect the reality of our time: innovation does not happen twice a year; it is constant. Whether you are a strategic leader or a hands-on practitioner, this new cadence is built to get you quickly up to speed.

• Watch on demand: Dynamics 365 Business Applications Update
• Watch on demand: Power Platform and Copilot Studio Business Applications Update

The post 2026 release wave 1 plans for Microsoft Dynamics 365, Microsoft Power Platform, and Copilot Studio offerings appeared first on Microsoft Dynamics 365 Blog.

Work is changing at a structural level.

Three forces are converging. The interface layer is shifting to AI assistants. Agents handle workflow orchestration. And an intelligence layer is consolidating information across structured and unstructured sources. Together, these forces mark Frontier Transformation, where AI moves beyond basic efficiency to open new opportunities for creativity, innovation and growth.

This transformation also creates a new kind of business application: one that is integrated with the AI assistant people use every day, accessible to agents and grounded in the unique intelligence of each organization.

We call these agentic business applications. The applications themselves still reflect real business processes. But how people interact with them, how work moves through them and how they connect to the rest of the business is fundamentally different.

Today, we’re taking a significant step forward in bringing agentic business applications to life across Microsoft 365, Dynamics 365, and Microsoft Power Platform.

Interact with business applications inside Copilot

Microsoft 365 Copilot is becoming an interactive workspace for business applications. Dynamics 365 Sales, Dynamics 365 Customer Service, and custom apps built with Power Apps will surface directly as agents with rich UX inside chat in Microsoft 365 Copilot. Agents using Apps SDK and MCP Apps can also bring Microsoft partner apps into the conversation, including tools teams already use, like Adobe Express, Figma, and Wix. This is the interface layer shift in practice. Instead of switching between applications, users state what they need in Copilot and the system responds. You can review data and take action without leaving the conversation. Copilot becomes the place where work gets done.

As an example, a human resources (HR) employee can now call on their custom HR app, built with Power Apps, right within Copilot to compile a list of office locations with the highest new hire counts this quarter, viewing the results in an organized table with filter options. Additionally, they can prompt the application to show the results in a map view, all without leaving their Copilot interface.

Or a customer service representative can begin their day in Microsoft 365 Copilot by reviewing a summary of priority cases they need to focus on, easily viewing and updating their data from Dynamics 365 Customer Service.

Public preview for this capability will be available later this month for Power Apps, with availability for Dynamics 365 Sales and Dynamics 365 Customer Service launching in early April 2026. Throughout the next month, we’ll also introduce support for this capability across a handful of Microsoft partner apps, including Adobe Express, Adobe Acrobat, Base44, Box, Canva, Coursera, Figma, Miro, Monday.com, Optimizely, and Wix. All pre-built partner app experiences will be accessible via the Microsoft 365 Agent Store for users with Microsoft 365 Copilot.

Bring Copilot and agents into Dynamics 365 and Power Apps

The experience works in both directions. Microsoft 365 Copilot and agents like Researcher and Analyst will be accessible directly within Dynamics 365 Sales, Dynamics 365 Customer Service, and custom apps built with Power Apps. Employees get the same Copilot capabilities they trust across Microsoft 365 while staying grounded in their operational systems.

Customers can continue to benefit from pre-built agents in Dynamics 365, including Sales Qualification Agent, Case Management Agent, and Account Reconciliation Agent, which help teams automate routine work and focus on higher value decisions.

Consider a seller working in Dynamics 365 Sales who asks Researcher to generate a full account overview: customer relationship management (CRM) context, internal knowledge, and external research combined in one response, surfaced in place. The unit of value shifts from “find the right screen” to “get the answer and act.” This creates a more consistent experience across productivity tools and business applications. Work moves from insight to execution with less friction between systems.

Microsoft 365 Copilot in Dynamics 365 Sales, Dynamics 365 Customer Service, and canvas apps in Power Apps will be available in public preview by early April 2026. Microsoft 365 Copilot in model-driven apps built with Power Apps will reach general availability by early April 2026. A Microsoft 365 Copilot license is required. This experience with Power Apps also requires a Power Apps premium license.

Microsoft 365 Copilot in Power Apps allows us to ask questions and make decisions directly against our Dataverse data, while also combining insights from Microsoft 365 when needed. The experience now feels truly unified, allowing our users to summarize complex operational data, trigger actions, and seamlessly access insights. We’ve seen significant increases in the value provided to both our internal solutions and customer-facing products.

Peter Kestenholz, Founder & Head of Innovation, Context&

Grounded in your organization’s intelligence with Work IQ

Underpinning all of this is Work IQ. Work IQ connects signals from Microsoft 365 with operational data from Dynamics 365 and Power Apps. It follows work as it happens across documents, meetings, chats, and business processes. This is the intelligence layer: the thing that resolves entities and relationships across structured and unstructured sources, so agents and Copilot share a common understanding of what is happening across the business.

Decisions discussed in a meeting or email can connect to live data in a business application. Changes in one place surface where attention is needed elsewhere. And because this intelligence is grounded in Dataverse and your organization’s own data, actions stay aligned to real processes and real context.

For example, when a pricing change is discussed in a meeting, Work IQ understands how that decision impacts active opportunities in Dynamics 365 Sales, surfacing the affected opportunities within Copilot for review.

Work IQ plays an important role in making business applications agentic. Without it, agents operate on partial information. With it, they act on the full context of the business.

Users with a Microsoft 365 Copilot license can experience Work IQ with Dataverse integration directly inside Power Apps, Dynamics 365 Sales, and Dynamics 365 Customer Service in public preview by early April 2026.

See how it all comes together

Copilot, agents, and Work IQ come together as a system of work. Within that system lies a new generation of business applications: applications that understand context, respond to intent, and support execution where work actually happens. The business application stack is entering a significant architectural shift. What we’re announcing today is one step in that larger transition. We are building a platform where applications, intelligence and execution converge so teams operate with more clarity and less overhead.

You’ll see this foundation expand across Dynamics 365, Microsoft Power Platform, and Microsoft 365 as we bring more agentic capabilities into the flow of work. Agentic business applications are already taking shape.

• Tune in to the Microsoft Frontier Transformation digital event on March 9, 2026 at 8:00 AM PT.
• Register for the Business Applications Update event streaming on March 18, 2026 at 9:00 AM PT.

The post A new way of working is taking shape: Frontier Transformation appeared first on Microsoft Dynamics 365 Blog.

Every seller knows how much time gets lost between selling moments. Information arrives in many forms—emails, screenshots, documents, handwritten notes—and turning that into structured CRM data often means manual copying, rework, or skipped fields altogether. At the same time, answering everyday questions like “Which leads should I follow up on?” or “How is my pipeline shaping up right now?” can require complex filters, multiple views, or exporting data just to get a clear answer.

Dynamics 365 Sales is evolving to address these challenges with agentic assistance. Instead of sellers adapting to rigid forms, grids, and filters, agentic AI in Dynamics 365 Sales now adapts to how sellers naturally work—by understanding unstructured inputs, interpreting intent, and assisting directly at the point of action. Two purpose-built agents bring this to life:

• A Data Entry Agent that uses LLMs to understand pasted content and uploaded files, extract relevant details, and quickly populate CRM forms for faster lead and contact creation.
• A Data Exploration Agent helps sellers quickly understand trends across opportunities, leads, or accounts by turning natural language questions into filtered views and visual insights.

Together, these agents reduce two of the biggest productivity drains in sales—manual data entry and cumbersome data exploration—so sellers can spend less time managing CRM and more time engaging customers.

Let’s look at how these experiences use agentic AI in Dynamics 365 in real sales scenarios:

Capture sales data faster with the Data Entry Agent
Accurate customer data is critical, but sellers encounter information in many forms—emails, websites, documents, and business cards. The Data Entry Agent uses large language models to understand unstructured text and files, infer intent, and map extracted details to the right CRM fields, without requiring sellers to manually interpret or retype information.

Capture Lead and Contact details instantly with Smart Paste

When a seller receives an inbound email from a prospect, creating a lead often means manually copying names, email addresses, phone numbers, and company details into CRM. For example, a prospect may write:

You want to respond quickly, but first you need to log the lead.

With Smart Paste (Preview), sellers can copy the email content, navigate to the lead or contact form. The system analyzes the copied text, extracts key details such as name, company, email, and phone number, and suggests values inline for the relevant fields. Each suggestion includes inline citation from the email, so sellers can clearly see the source of the information.

Sellers can review AI-generated field suggestions, view citations, accept what looks right, and save—enabling faster lead capture with greater confidence in data accuracy.

Similarly, a seller may be reviewing a prospect’s website or LinkedIn profile in separate tabs. Instead of manually re-entering details later, they can copy text from the company’s About Us page or the prospect’s LinkedIn profile and paste it directly into a CRM form. The agent analyzes the content and suggests values such as industry, company name, location, and job title, allowing the seller to review and apply the information immediately while the context is still fresh.

Convert Physical Documents into CRM Records with Files (Preview)

After trade shows, conferences, or in-person meetings, sellers often return with a stack of business cards or documents from dozens of conversations. Manually transcribing this information delays follow-up and increases the chance of errors.

With Files (Preview), sellers can upload images of business cards or documents such as .txt, .docx, .csv, .pdf, .png, .jpg, .jpeg, or .bmp, directly into the form. The system analyses the uploaded files and suggests values for relevant fields, including names, titles, company details, email addresses, and phone numbers. Sellers simply review and confirm the suggestions, turning what once took hours into minutes.

This enables faster post-event follow-up and more complete lead and contact records.

Find and understand sales data faster with the Data Exploration agent

Finding the right records and understanding trends is essential for sellers, but navigating views and filters can be time-consuming. Powered by natural language understanding, the Data Exploration Agent (Preview) translates seller questions into structured filters, allowing users to interact with CRM data using plain language instead of complex query logic, making it easier to plan, prioritize, and understand pipeline health directly within their views.

Find the right records faster using Natural Language in Views

Filtering records in CRM can be time-consuming, especially when multiple criteria are involved. Imagine planning your day and opening My Open Leads to focus on recent campaign responses. Instead of building complex filters, you simply type: “Leads from the Summer Campaign created last month.”

Or, when preparing for a forecast call, you search: “Opportunities from Technology accounts closing next quarter.”

The system interprets the request and automatically applies the appropriate filters to the view. Sellers can review and modify the filters if needed, giving them both speed and control. This simplifies daily planning, follow-ups, and pipeline reviews.

Visualize data in Views to spot trends quickly

Understanding trends often requires more than scanning rows of data, but building dashboards or exporting reports isn’t practical for day-to-day sales work. With Visualize (Preview), sellers can turn the filtered data they’re already viewing into interactive charts with a single click—directly within the view and without breaking their flow.

Because the visualization is generated from the current view and visible columns, it automatically reflects the exact filters, segments, and scope the seller is working with. Sellers can hover to see detailed values, drill into specific segments, and switch chart types on the fly as new questions come up. This makes it easy to answer questions like “Where are most of my open opportunities concentrated?”, “Which lead sources are driving volume right now?”, or “How is my pipeline distributed across stages?”

Visualize is designed for quick, in-the-moment understanding, not deep reporting. It complements Power BI by giving sellers immediate visual insight at the point of work—without creating reports, navigating dashboards, or leaving CRM—so they can recognize patterns and act faster while staying in flow.

Enable these agentic capabilities in Power Platform Admin Center

• To enable Data Entry agent capabilities, go to Power Platform Admin Center → Settings → Product → Features.
  Under AI form fill assistance, turn On
  • Automatic suggestions
  • Smart paste and file suggestions and
  • Form fill assist toolbar. Changes apply to model-driven apps once saved.

• To enable Data Exploration agent capabilities, go to Power Platform Admin Center → Settings → Product → Features.
  • 
    Under Natural language grid and view search, set Enable this feature for to All users immediately
  • Turn On Allow AI to generate chartsto visualize the data in a view and enable AI-generated chart styling for a consistent visual experience.

Focus More on Selling, Less on Administration

With agentic AI in Dynamics 365 Sales, the platform evolves from a system of record into a system that understands, assists, and adapts—helping sellers spend more time selling and less time managing CRM.

The post From manual work to meaningful selling: How Agentic AI is transforming Dynamics 365 Sales  appeared first on Microsoft Dynamics 365 Blog.

In October 2025, we announced the general availability of the Sales Qualification Agent (SQA) in Dynamics 365 Sales—a breakthrough in autonomous lead qualification. Sales Qualification Agent empowers sellers by helping build higher quality opportunity while eliminating tedious, repetitive work. Sales Qualification Agent autonomously researches every lead, initiates personalized outreach, and engages prospects to understand purchase intent, ensuring that sellers spend their time meeting prospects who are ready to take the next step. With modes enabling both seller-driven and fully autonomous qualification, the agent supports a key goal for sales organizations—increasing revenue per seller.

Customers are using Sales Qualification Agent in two ways: 

1. Helping boost revenue beyond current sales capacity
   • Responding to inbound leads within minutes instead of days, increasing response rates and in turn, qualified opportunities.
   • Engaging leads that sellers are unable to follow up on due to capacity constraints, or those deemed economically unviable to pursue.
   • Increasing pipeline quality by focusing the seller’s time on a handful of high intent, engaged leads recommended by the agent.
2. Helping reduce sales costs
   • Reducing back-office costs related to lead research and validation, using Sales Qualification Agent in “Research only” mode to hand-off only the leads that meet the ideal customer profile criteria.
   • Automatically disqualifying low-quality leads, saving hours of seller time during the week.

Continuing benchmarking the quality of sales AI agents

Microsoft is building the future of agentic Sales technology with prebuilt AI agents, such as Sales Qualification Agent, the Sales Research Agent, and the Sales Close Agent available in Dynamics 365.

At Microsoft, we’re committed to delivering quality, trust, and transparency with our agents, and that requires rigorous evaluation. As we continue to build new agents and improve existing ones for critical sales workflows, evaluation benchmarks provide a structured and transparent way for our customers to measure quality for the jobs the agent does.

Today, we’re announcing the Microsoft Sales Bench—a new collection of evaluation benchmarks designed to assess the performance of AI-powered sales agents across real-world scenarios. This framework brings together purpose-built metrics, hundreds of sales-specific scenarios, and composite scoring validated by both human and AI judges.

The Sales Bench isn’t starting from scratch. It now formalizes and expands what began with the Sales Research Bench, published on October 21, 2025, which evaluates how AI solutions answer business research questions for sales leaders.

Today, we’re extending the Microsoft Sales Bench with a second benchmark: the Microsoft Sales Qualification Bench, focused on measuring how effectively AI agents qualify leads and generate high-quality pipeline.

Introducing the Sales Qualification Bench for lead qualification

This Microsoft Sales Qualification Bench evolved from rigorous evaluations we conducted since the Sales Qualification Agent’s public preview in April, with the goal of objectively measuring quality as we further developed the agent, partnering with customers from a diverse set of industries. Since the preview, we measured every update against these standards, ensuring improvements are real and repeatable.

We generated a synthetic dataset modeled after companies from three different industries, with 300 leads, with attributes such as name, company, and email ID—representative of what sales teams typically work with before any enrichment or hygiene is performed. In addition to these typical attributes, we also added key knowledge inputs such as value proposition of the products being sold, customer case studies, and documentation for answering customer questions.

In addition to Sales Qualification Agent, we used the evaluation framework to measure ChatGPT by OpenAI on the same dataset. Since we didn’t have access to an autonomous agent from OpenAI, we mimicked how a human seller would use ChatGPT to recreate the three key jobs SQA performs. We provided each system—Sales Qualification Agent and ChatGPT—the exact same lead inputs, knowledge sources, and contextual signals under controlled evaluation configurations. We used a ChatGPT Pro license with GPT-4.1. This model is the closest match (and slightly better) to Sales Qualification Agent’s GPT-4.1 mini, which we intentionally chose to deliver optimal quality at lower cost per lead than newer models. Additionally, Pro license was chosen to optimize for quality: ChatGPT’s pricing page describes Pro as “full access to the best of ChatGPT.”¹

The framework evaluates outputs from the three jobs across Sales Qualification Agent and ChatGPT:

• Research: Company research for the lead—background, strategic priorities, financial health, and latest news.
• Outreach: A personalized email generated based on research, to make initial contact with the lead.
• Engagement: The agent’s conversation with a lead until it’s qualified or dispositioned.

Our scoring metrics span core quality (accuracy, relevance, completeness), trustworthiness (grounding and citations), and business-specific success criteria (e.g., relevancy of company research to highlight interest in the seller’s offerings, personalization of the initial outreach emails sent to catch the lead’s attention, accuracy of responses to the lead’s questions to drive purchase intent, and the timing of handoff to a seller when the lead is ready to engage).

Outputs were scored independently by both human reviewers and an LLM judge built with GPT-5.1, using a 1–10 scale for each metric. These metric-specific scores were then rolled up using a simple average to produce a composite quality score. The result is a rigorous benchmark presenting a composite score and dimension-specific scores to reveal where agents excel or need improvement. Our methodology, metrics, and their definitions are described in this technical blog.

Results

In evaluations completed on December 4, 2025, using the Sales Qualification Bench, Sales Qualification Agent outperformed ChatGPT on each of the three jobs required for sales qualification:

1. Research: The Sales Qualification Agent outperformed ChatGPT with 6% higher aggregate scores, leading on relevancy and completeness in research results that highlighted the lead company’s interest in the seller’s offerings.
2. Outreach: Sales Qualification Agent demonstrated 20% better results compared to ChatGPT, generating email drafts with accurate personalization and mentions of relevant recent events that will resonate with the lead.
3. Engagement: Sales Qualification Agent’s email responses to engage a lead over a multi-turn conversation scored 16% higher than ChatGPT’s. SQA generated emails that responded to the lead’s questions with accurate answers that develop their purchase interest and with precise discovery questions that qualify the lead before handing off to a seller.

In addition to performing better on these metrics, Sales Qualification Agent has the ability to run autonomously, which can help significantly reduce the time spent generating pipeline while helping sales teams build better quality pipeline.

Sales Qualification Agent scores well on these three jobs as its optimized for sales-specific scenarios and uses the following techniques to get great results:

1. It uses agentic Retrieval Augmented Generation (RAG) to relentlessly research each lead, ensuring greater completeness. More on this in the following section.
2. With knowledge of what the company sells, it can contextualize every workflow to increase relevancy for both the seller and the lead.
3. It can retrieve organizational knowledge from attached documents and internal repositories like SharePoint with greater precision, boosting accuracy of its responses when engaging with the lead.

The technical blog details which metrics SQA excels at relative to ChatGPT, where it falls short, and why.

Translating evals to real-world impact

Running evals led to major Sales Qualification Agent improvements during its six-month preview. Early results prompted us to try agentic AI design patterns, especially agentic RAG, which improved our company research by allowing iterative web searches and real-time reasoning. They also led us to enhance data coverage by auto-linking existing CRM records to each lead and inferring company names from lead emails. These updates provided sellers with deeper insights, revealing strategic opportunities and risks beyond basic facts.

For instance, when researching leads for a security company, Sales Qualification Agent can link news on recent cyberattacks to increased demand for its software. As highlighted in the technical blog, research synthesized by the agent makes such inferences more consistently than ChatGPT. Enhancing the agent’s research also improved the relevance and personalization of outreach emails, helping agents better engage leads and clarify their ability and intent to purchase before handing them off to sellers.

Sandvik Coromant, a leader in precision cutting tools, partnered with us to pilot Sales Qualification Agent for their Digital Commerce program. After the updates, Pia Cedendahl, Global Sales Manager for Strategic Channels/Partners and Online Sales, noted, “Sales Qualification Agent’s answers became far more on-point to our business—it’s like having a research assistant that already understands what we care about.” Sandvik Coromant saw improved lead conversion and higher engagement from their Digital Account Managers, validating the impact of our evaluation-driven approach. Pia joined Microsoft leaders at the Microsoft Ignite 2025 session, “Accelerate revenue and seller productivity with agentic CRM,” where she shared how the team saved more than 120 hours and $19,000 in just the first three weeks since launching a pilot, and forecasted a 5% increase in revenue with full rollout.

Better insights, more personalization, proven value

Equipped with agentic AI design and backed by data-driven evaluation, customers can confidently use the Sales Qualification Agents so that:

• Sellers receive comprehensive company overviews, timely news highlights, and actionable recommendations that are consistently delivered with high quality—drawing a clear line from insight to action.
• Sales leaders can expand their qualified pipeline cost efficiently, with the agent ensuring high lead quality.
• Prospects benefit from more personalized outreach, enhancing their experience and supporting increased conversion rates.

What’s next

We’ll continue to refine Sales Qualification Agent using agentic design patterns, aiming to make every improvement measurable and meaningful. Stay tuned for the full evaluation results and methodology for the Sales Qualification Bench, which will be published for transparency and reproducibility. We also intend to add more evaluation frameworks and benchmarks to the Microsoft Sales Bench collection including benchmarks that cover future sales agent capabilities.

¹ChatGPT pricing page, accessed November 24, 2025

The post Dynamics 365 sets the bar for agentic sales qualification on new benchmark appeared first on Microsoft Dynamics 365 Blog.

The Sales Qualification Agent (SQA) in Dynamics 365 Sales introduces a new class of autonomous sales AI, one that does far more than assist with drafting or summarization. SQA performs multi-step reasoning, conducts live web research, generates personalized outreach, and engages prospects in multi-turn qualification conversations. These capabilities directly shape pipeline quality, seller productivity, and customer relationships. 

As agentic AI becomes deeply embedded in revenue-critical workflows, trust must be earned through transparent, repeatable, and rigorous evaluation—not anecdotal wins or point demos.

Today, we’re announcing the Microsoft Sales Bench—a collection of evaluation benchmarks designed to assess the performance of AI-powered sales agents across real-world scenarios. Adding to the Sales Research Bench already published as part of this collection to evaluate Sales Research Agent, today we are also publishing the Sales Qualification Bench to evaluate Sales Qualification Agent in Dynamics 365 Sales.

This post presents the detailed evaluation methodology and results for the agent, including a head-to-head comparison against chatGPT using identical data, tasks, and scoring rubrics. These efforts establish the first benchmark purpose-built to measure end-to-end sales agent workflows, from research to outreach to live qualification. 

SQA Architecture  

The Dynamics 365 Sales Qualification Agent (SQA) architecture is designed as an end-to-end, enterprise-grade AI system that autonomously researches leads, synthesizes insights, and generates seller-ready outreach. It combines an intelligence engine powered by large language models with iterative web and enterprise data research, tightly integrated with Dynamics 365 Sales and Microsoft Copilot Studio for orchestration. Built on secure enterprise foundations, the architecture enforces governance, compliance, and data protection while enabling scalable, trustworthy AI-driven sales workflows. 

Evaluation Metrics and Methodology 

To understand how well the Sales Qualification Agent (SQA) performs in real-world sales qualification workflows, we designed the Sales Qualification Bench, a comprehensive evaluation that mirrors how sellers actually research leads, personalize outreach, and engage with prospects. Our goal was straightforward: measure whether SQA can help reps qualify faster, personalize more effectively, and carry higher-quality customer conversations—using the same signals and information they rely on every day. 

To ensure that the evaluations accurately represent real-world conditions, we developed a testbed that closely mirrors the complexity and ambiguity found in contemporary sales environments. This allowed us to evaluate SQA end to end, from autonomous research and reasoning to grounded, actionable research briefs, outreach messages, and multi-turn qualification conversations. 

Evaluation Setup

To ensure real-world fidelity, we constructed a production-like lead evaluation environment that mirrors how SQA operates in Dynamics 365 Sales. 

Lead and Data Corpus 

• Three synthetic but realistic seller companies (C1) across distinct industries, with unique: 
  • Product offerings 
  • Knowledge sources 
  • Ideal customer profiles 

• 300+ lead dataset (C2) expanded into a scenario-rich corpus: 
  • Companies across 6 global regions (North America, Europe, Asia, South America, Australia, Africa) 
  • 33 industries 
  • Mixed clarity (well-known brands and long-tail companies) 
  • Structured attributes (name, role, email) 

• CRM roles represented: 
  • Sales representatives 
  • Digital specialists 
  • Customer success managers 
  • Each linked to relevant accounts, opportunities, and cases 

• Company segment coverage: 
  • Enterprise 
  • Mid-Market 
  • Small Business 
  • Government 
  • Education 

• 500+ email exchanges simulating real sales interactions: 
  • Technical product questions 
  • Meeting requests 
  • Ambiguous or low-intent inquiries 

Simulated Agent Workflows 

All evaluations reflected real SQA behavior: 

• Autonomous web-based research 

• Role-aware outreach generation 

• Multi-turn qualification conversation handling 

Tasks Evaluated and Evaluation Metrics 

1. Company Research 

For each lead, the agent generates a structured research brief including: 

• Business overview, strategy and priorities 

• Financial signals 

• Recent news relevant to the seller 

2. Lead Outreach 

Based on its research, the agent generated a personalized email aligned to: 

• The lead’s role 

• The seller’s value proposition 

• The company’s business context 

• Value-based positioning 
   

3. Qualification Conversations (Engage) 

The agent then autonomously engages back and forth with the lead, progressively asking them questions for customer-configured qualification criteria such as budget, need, and timeline and answering the lead’s questions such as: 

• “What does your solution do?” 

• “How are you priced?” 

• “How do you compare to competitors?” 

• “Who else uses this?” 

Each metric is scored independently on a 0–10 scale, where higher scores indicate stronger performance. We used an LLM-as-a-judge approach to score outputs against the ground truth and rubric and manually reviewed a sampled subset of evaluations to calibrate the judges and validate scoring consistency. To reduce judge variance and mitigate hallucination risk, each sample was evaluated five times, and the mean across runs was recorded as the final score. 

Benchmarking Strategy with ChatGPT 

To ensure an objective and fair comparison, we replicated a standard seller workflow in ChatGPT UI using GPT-4.1 with Pro license, a more advanced model than the GPT-4.1-mini variant currently used by SQA. 

Standard Prompting 

This setup simulates how a seller naturally interacts with a general-purpose LLM: 

• High-level contextual instructions only 
• Mirrors SQA’s autonomous research-to-outreach flow 

This ensures: 

• Workflows remain representative and unbiased 
• Comparisons reflect real-world usability, not prompt-engineering skill 

Identical Knowledge Sources and Context 

ChatGPT was given the exact same knowledge sources as SQA, including: 

• Full lead information and seller value proposition 
• Seller Q&A documentation via the SharePoint connector 
• Historical conversation context for reply generation 

This isolates differences in agent reasoning and orchestration, not data access. 

Evaluation Results  

Microsoft evaluated the Sales Qualification Agent (SQA) and ChatGPT with over 300 leads, covering research, outreach, and qualification tasks with identical knowledge sources. Evaluations completed on December 4, 2025, showed that SQA consistently outperformed ChatGPT-4.

• Research: SQA was 6% more effective at relevant, thorough company research. 
• Outreach: SQA was 20% better at personalized communication and timely event references. 
• Engagement: SQA scored 16% higher for precise responses and targeted qualifying questions. 

SQA also operates autonomously, reducing overhead and boosting pipeline quality for sales teams. 

Results by Task Category 

1. Company Research 

SQA was 6% better than ChatGPT, winning in its ability to perform more relevant and complete research that highlighted the lead company’s interest in the sellers offerings: 

• SQA provided more relevant results: To ensure sellers spend their time on the most important leads, they need to determine whether a lead is good fit for their offerings. While both SQA and ChatGPT were given the same context (seller company and value proposition of the offerings), SQA consistently did better at tying its research back to this context, helping sellers determine fit. Appendix A shows an example where SQA was able to tie the company’s strategic priorities to its need for a collaboration platform and infer strong purchase ability from its robust operational health and minimal leverage burden.

• SQA synthesized results with higher level of fidelity and completeness: The agent’s value is directly correlated to its ability to eliminate tedious work for the seller. SQA produced more detailed research synthesis (as demonstrated in Appendix A), giving a single, trusted source for the seller to get equipped with any insights they may need.  

These results stem from numerous experiments aimed at optimizing web research for the best outcomes at minimal cost, rather than relying on costly advanced models. Sellers get deeper insights with SQA’s agentic RAG for real-time reasoning with iterative web search results, combined with unique capabilities that increase data coverage, for example, auto-linking CRM records and extraction of company name from lead emails. 

2. Personalized Outreach 

SQA was 20% better than ChatGPT, notably ahead in the level of personalization and mentions of relevant recent events that will resonate with the lead. 

• More personalized and customer-centricity: A lead is more likely to respond to a cold outreach email that directly explains how the seller’s offering can address their needs. SQA did so effectively by starting with the lead’s situation and recent events, while ChatGPT often focused on the seller and uses heavier technical jargon. A clear, actionable call to action bookends the email and guides the conversation forward. Appendix B shows an example of how SQA was able to tie a recent acquisition the lead’s company made to the value proposition of the seller’s offering. 

These results are based on direct engagement with sellers – every sales team that deploys SQA gives us precious feedback that all other customers benefit from.   

3. Qualification Conversations (Engage) 

SQA was 16% better than ChatGPT. It responded with greater precision to the lead’s questions to develop purchase interest and asked pointed discovery questions to better qualify the lead before handing off to a seller. 

• Answers accurately by correctly understanding the lead’s intent and maintaining conversation context effectively. To drive deeper buyer consideration, SQA independently answered even the most technical questions that leads had about the seller’s offerings while maintaining the context from earlier messages in the simulated conversation, delivering clear, direct, and well-structured responses. Appendix C demonstrates SQA’s ability to pull the most relevant information from provided knowledge sources (in this case, files with technical specifications) during an ongoing conversation with a lead. 

• Handles uncertainty responsibly, handing off to a supervisor/seller when appropriate. Both SQA and ChatGPT were instructed to handoff a lead to a supervising seller when a suitable response cannot be generated or when the lead is considered qualified as per pre-defined criteria. SQA handed off accurately and at the right moment in more tests than ChatGPT.  

• Demonstrates strong discovery coverage. To maximize the value exchange from each follow-up conversation with the lead, SQA and ChatGPT were instructed to include discovery questions in their response to assess pre-configured qualification criteria (covering lead’s need, budget, buying authority and purchase timeline). SQA was able to ask pointed discovery questions to cover more of these criteria than ChatGPT in our simulated conversations. This resulted in SQA identifying and handing off better qualified leads through its engagement.

These gains are attributable to the hard lessons we have learnt through close collaboration with customers to understand the diversity of needs on intent detection and knowledge retrieval across AI agents for Dynamics 365 Sales, Service and Contact Center applications.  

Conclusion 

Sales Qualification Agent (SQA) is not a simple productivity tool—it is a complex multi-step agent directly influencing revenue outcomes. The Sales Qualification Bench represents a foundational step toward enterprise-grade trust, transparency, and continuous quality improvement for agentic AI in sales. 

Through rigorous, scenario-driven benchmarking, SQA demonstrated clear leadership across research, outreach, and qualification—while also surfacing concrete architectural and rule-quality investments for the next phase of growth. 

Upcoming plans for the Sales Qualification Bench include using the benchmark for continuous improvement of the Sales Qualification Agent, running further comparisons against a wider range of competitive offerings, and publishing the eval package so customers can reproduce it on their own data. Evaluation is not a one-time event. It is now embedded as a first-class engineering discipline in the SQA development lifecycle—ensuring the agent evolves safely, measurably, and in lockstep with seller needs. 

Appendix

Sample inputs and responses for each category  

Appendix A. Sample of research synthesis generated by SQA

Appendix B. Sample of an initial outreach email generated by SQA 

Appendix C. Sample of an engagement email generated by SQA 

The post Sales Qualification Agent: How we evaluated and improved AI quality with benchmarks appeared first on Microsoft Dynamics 365 Blog.

Imagine being able to ask your CRM system a question like, “Which opportunities are likely to close this week?” or “Who has met with Ernie Kerrigan at Contoso recently?” and getting an instant, accurate answer without writing a single query or navigating through multiple Views in Dynamics 365.

Whether you’re using Copilot in Dynamics 365 Sales, Power Apps customized through Microsoft Copilot Studio or Microsoft 365 Copilot for Sales, under the hood, these experiences leverage one common engine: AI-powered Dataverse (DV) Search, which seamlessly connects business users to the underlying database schema, translating intent into action without requiring technical expertise. Thousands of enterprise customers already rely on this capability to power their business workflows.  

Figure 1: How AI-powered Dataverse Search Connects Copilot Experiences Across Dynamics 365 

We’ve reimagined the technology behind Dataverse Search from the ground up. Leveraging recent breakthroughs in agentic AI, the new system delivers answers that are more relevant, contextual, and accurate to your specific business data. Think of it as an intelligent assistant that not only understands your question but figures out the best way to answer it using an adaptive reasoning process.  

In this blog, we’ll explore why this agentic approach was necessary, how it works under the hood, and how it scales to enterprise needs supporting complex schemas, massive datasets, and domain specific terminologies while adhering to Microsoft Responsible AI principles. In particular, the agentic approach is model-agnostic, and while different models or fine-tuned models can influence the quality of results, the choice of model is orthogonal to the architecture. For this post, our emphasis remains on the agentic loop and its role in delivering dynamic, context-aware answers. Further, we will demonstrate our success via evaluation results and show you ways to customize it for your business. 

Queries to Conversations: Unlocking Your Live Business Data 

Every organization’s Dynamics 365 environment is unique, and most customers customize it extensively. Over time, these customizations lead to complex schemas, ambiguous relationships, and massive datasets spanning millions of records and terabytes of data. Our original Dataverse Search system was pioneering, but it relied on a fixed-plan natural language to SQL pipeline. A user’s question was converted to SQL through sequential stages: parsing, schema mapping, data linking, and SQL generation. This design was prone to cascading failure in a sequential pipeline. Each stage operated in isolation without shared context, so a single error could invalidate the entire query. Every question followed the same fixed flow, even when certain steps were unnecessary. This resulted in brittle behavior and suboptimal answers for complex or ambiguous queries that spanned multiple tables. 

We recognized the need for a more adaptable, resilient approach to tackle the complexities of enterprise data. This upgrade shifts DV Search beyond simple Search into intelligent, interactive conversations with your business data. For you, this translates into immediate, actionable value by providing:  

• Real-Time, Actionable Answers: Ask, “Which of my open opportunities in New York are scheduled to close this month?” and get an instant answer from the live Dataverse data. This isn’t a report from last night’s data refresh; it’s the current state of your business. 

• Democratized Data Access: A service manager can ask, “Show me active, high-priority cases that haven’t been updated in 3 days” without needing to understand the underlying table structure of incidents and case/activities. 

• Deeper Contextual Conversations: The agent supports multi-turn conversations. After asking about opportunities in New York, you can follow up with, “Of those, which ones are for our ‘Pro’ license?” The agent remembers the context, providing a progressively refined answer. 

Under the Hood: Agentic Architecture 

To overcome some of the limitations of the earlier system and to meet the complex customer scenarios, the new DV Search architecture introduces an Agentic Orchestrator powered by GPT4.1. It transforms query handling from a static pipeline into a dynamic reasoning loop: plan → execute → refine. Instead of blindly converting text to SQL, the orchestrator treats each question as a goal, intelligently deciding the best steps to reach it. 

Figure 2: Agentic Architecture for AI-powered Dataverse Search 

Context Awareness and Conversations: When a user submits a new or follow-up question, a dedicated preprocessing component reviews prior conversation history and rewrites the query as a single, self-contained question, enabling coherent multi-turn conversations. For example, if you ask, “Show my top opportunities in Q4” and then follow up with “How about in Europe only?”. the component understands the second question is a refinement of the first rather than starting from scratch or losing track of prior context. This conversational capability makes interactions feel natural and efficient. The refined question is then enriched with the business’s domain knowledge (glossary) to fully reflect the user’s intent within the specific business context. 

Dynamic Planning and Execution: When the self-contained question comes in, the orchestrator doesn’t simply translate it into SQL. Instead, it breaks the query into logical steps and decides which tools to use and in what order, while also utilizing the domain knowledge encapsulated with the supplied glossaries. These tools include:  

• schema_linking_tool: Schema lookup for understanding tables and relationships 

• data_linking_tool: Semantic Search for finding relevant data values and resolving data ambiguities 

• sql_execution_tool: SQL execution tool for retrieving results 

• submit_plan_update_tool: Captures both the original plan and any course corrections made during execution 

The orchestrator adapts on the fly if the first attempt fails or returns incomplete results. It analyzes the issue, revises the plan, and retries. This self-correcting loop is a major improvement over older systems that suffered from cascading failures. 

Handling Relational Complexity: One of the most powerful aspects of this approach is its ability to handle relational complexity. Operational business application schemas often require multi-hop joins across multiple tables, including custom entities. The orchestrator understands these relationships and can navigate them intelligently, ensuring accurate joins and filters even in highly customized environments. For example, if a question involves linking Accounts to Opportunities and then to a custom Product table, the agent plans the steps and executes them seamlessly. 

Personalization and Learning: Personalization further enhances the experience. Over time, the system learns from usage patterns within your organization. If you frequently work with the Accounts table or use certain custom fields, the agent prioritizes those interpretations in future queries. This learning is based on interaction signals, not external data, and is carefully scoped to respect privacy and organizational boundaries. The result is a system that becomes more aligned with your business logic the more you use it. 

Real-World Example 

Imagine you run Fourth Coffee Machines, a business selling premium espresso and grinder units to commercial and residential customers. It’s managed through a Power App built on Dataverse. A seller begins with a simple keyword search in top-bar search in Power Apps for “Fourth Coffee” to confirm the account record. Thanks to fuzzy matching and relevance re-ranking, even typos like forth coffee or 4thcoffee surface the right entity instantly. 

From there, the seller asks Copilot: “Show me my open opportunity at risk with Fourth Coffee.” The agent rewrites the query, scopes it to the current user, interprets at risk as a cold rating, and joins Account → Opportunity. It executes SQL, returns the results, and summarizes them with citations—no manual filtering, no report building. 

Finally, the seller pivots to a KPI question: “What is the HRR for Coffee Grinder 02?” Here, the agent consults the business glossary, which defines HRR as Happy Response Rate (positive sentiment ÷ total reviews in the Product Review table). It computes the metric, explains the formula, and cites the source records. The user now understands exactly how the number was derived. 

Under the hood, this seamless experience is powered by an Agentic Orchestrator that plans, executes, and refines dynamically. It chooses the right tools, adapts when errors occur, and injects domain knowledge from glossaries. By combining dynamic planning, iterative refinement, relational understanding, and personalization, it represents a significant leap forward from static query pipelines. It’s not just about generating SQL it’s about orchestrating an intelligent, context-aware process that feels conversational and delivers real business value.  

Evaluation Results 

To measure how well our agentic system performs in practical enterprise scenarios, we evaluated it against curated datasets of user prompts each representing or assisting with a real job to be done. These prompts reflect the everyday questions and tasks that drive productivity for CRM users— from quick record lookups and aggregation analytics using keyword search or simple filters and joins, to complex multi-join queries requiring domain expertise. By categorizing prompts into levels of complexity, we ensure the evaluations capture the full spectrum of enterprise challenges.  

For each complexity level, we report two practical metrics: Relaxed Execution Accuracy (EX Accuracy) and P80 Latency. Relaxed Execution Accuracy measures how often the generated SQL returns the same rows as the reference SQL when both are executed on the same data—extra columns in the predicted query are allowed, but extra or missing rows are not; order is ignored unless ORDER BY is specified. P80 Latency is the 80th percentile end to end response time, from request receipt through retrieval, model inference, and verification to the final SQL result. Together, these metrics give a transparent, action-oriented view of correctness and responsiveness as task complexity increases. It highlights where the Agentic framework delivers reliable, efficient answers that empowers users to get more done with natural language.  

† Metrics averaged over multiple runs 

In practice, higher accuracy often comes at the cost of increased latency. Conversely, pushing for low latency can reduce end to end quality. This Agentic system is designed to navigate that tradeoff, delivering strong accuracy while keeping latency within practical bounds. This achieves a practical balance for production use. 

Tuning for Your Business: Glossaries and Enriched Schema 

No AI system knows your business out-of-the-box. We’ve added tuning mechanisms that let makers refine how the Q&A agent understands your data: 

• Glossaries: You can define a glossary to teach the agent your company’s unique vocabulary and acronyms. For example, if “QoQ” is common slang on your team for “quarter-over-quarter” or “CTX” refers to a particular set of products, you can add those to the glossary. The next time someone asks “What’s the QoQ growth for CTX?”, the agent will know exactly what that means. This helps align the AI with the lingo of your organization so it interprets queries the same way a knowledgeable employee would. 

• Schema Descriptions: Dataverse allows adding custom descriptions to tables and columns. By populating these descriptions with meaningful info, you give the agent extra context. For instance, two fields might both be called “Status” – one on a custom entity and one on a standard entity. If you add descriptions like “(Order Status – custom)” vs “(System Status code)”, the agent can use that to pick the right field during SQL generation. Essentially, you’re able to clarify the semantics of your data model for the AI. 

Using the inherent metadata in Dataverse (like relationships and display names) plus these maker-driven additions, the agentic system can be tailored to use the correct terms and relationships in your domain, boosting accuracy even further. And because you control these glossaries and descriptions, you can continuously refine the AI’s understanding as your business evolves. 

Conclusion 

By reinventing Dataverse Search with an agentic architecture, we’ve moved from a rigid query engine to an adaptive, intelligent assistant for your business. The system understands nuance, handles ambiguity through reasoning, and even lets you inject your domain knowledge. Early adopters are seeing more questions answered correctly and faster than before, turning previously buried data into actionable insights. One leading global financial services company saw an Execution Accuracy surge from 22% to 97% on their marquee set of scenarios. This marks a significant step toward making enterprise data truly conversational. It empowers everyone from business users to power makers to tap into complex data and get the answers they need instantly and accurately, simply by asking. 

The post Beyond Retrieval: How an Agentic Approach Transforms Microsoft Dataverse Search  appeared first on Microsoft Dynamics 365 Blog.

Frontier Firms put Copilot, agents, and agentic business applications at the core of their operating model to enrich employee experiences, reinvent customer engagement, reshape business processes, and bend the curve on innovation. Today, we’re announcing several new agentic capabilities to help customers move to the Frontier—read on to learn more.

Transforming sales with Sales Development Agent

Earlier in 2025, we introduced our vision for how AI agents will transform critical sales processes like building pipeline and qualifying leads. Today marks the next milestone in that journey with the Microsoft Sales Development Agent, available through the Frontier Program in December 2025. Many sales organizations are under pressure to deliver more revenue with limited resources, and the Sales Development Agent helps sales teams scale their impact. This allows sellers to focus on nurturing customer relationships and closing deals.

Features include:

• Revenue and pipeline growth: The agent continuously researches prospects, crafts personalized outreach, and automatically follows up to ensure no lead is left behind.
• Scalability: Fully independent, yet collaborative, the agent acts as a teammate, with the ability to hand off leads to human sellers when needed.
• Security and governance: Built on Microsoft’s trusted security and compliance foundation and when enabled with Agent 365, the agent adheres to robust policies and access controls to ensure user data and workflows are protected.

Sales Development Agent connects with leading CRM systems like Salesforce and Microsoft Dynamics 365, and the Microsoft 365 apps your teams already use like Microsoft Outlook and Microsoft Teams.

The Microsoft sales team is among the first to use Sales Development Agent to reinvent the sales engagement process. With the use of Sales Development Agent, there was a 15.1% increase in the lead-to-opportunity conversation rate. ¹

Sales leaders want to help sellers act on more leads, reach more customers, grow faster, and improve revenue per seller. Microsoft Sales Development Agent can make that possible by creating an infinitely scalable sales organization, so no lead is left behind. Accenture plans to pilot Sales Development Agent across our global inside sales-as-a-service business—which helps clients sell to customers around the world—to boost their reach and revenue while maintaining cost to serve. We’ll use what we learn to help clients leverage Sales Development Agent, scale their teams, and unlock new growth.

—Chris Hergesell, Sales Reinvention Lead, Accenture Song

Learn more about Sales Development Agent with Accenture

From System of Record to System of Action

In October 2025, we shared our vision for agentic business applications—built on agents, Copilot, and unified data. These components are what define Dynamics 365 as a system of action.

Today, we’re taking that vision further with updates to Model Context Protocol (MCP) servers across Dynamics 365 and Microsoft Power Platform, strengthening the foundation for agentic capabilities across your entire business. MCP servers are configurable bridges between the business data within your line-of-business (LOB) apps, and the agents you build using tools like Microsoft Copilot Studio. It serves as a universal intermediary, unlocking a unified platform agnostic access to app data, modernizing how AI agents are interoperable with your apps.

For customers of Dynamics 365 Sales and Customer Service, we’ve used MCP to simplify integration between agents in Dynamics 365 and the platforms used by sellers and service reps to execute complementary workflows, like lead research, engagement, and qualification, as well as case management and case resolution, available in public preview on November 21, 2025.

For customers of Dynamics 365 ERP, we are announcing the public preview of the MCP server that unlocks hundreds of thousands of ERP functions for real-time use. We are also introducing a new analytics MCP server in public preview starting in December 2025. These two servers provide a secure, standardized foundation to connect ERP data with AI-powered analytics, helping customers make faster, more accurate decisions and innovate without sacrificing governance.

We are also announcing the Power Apps MCP server in public preview that enables agents to seamlessly trigger app capabilities such as approvals, form submissions, and data retrieval. This makes every Power App a composable, reusable building block in your organization’s AI ecosystem empowering both citizen and professional developers to expose app functionality to agents with confidence and control.

Lastly, the Dataverse MCP server, now generally available, allows people to benefit from natural language interactions, receiving real-time answers grounded in Dataverse data, while makers and admins gain powerful, built-in tools for data operations, search, and prompt execution.

We see tremendous excitement from customers and partners for agentic Dynamics 365 applications. Take Ramp, a financial operations platform designed to save companies time and money. Ramp built an agentic solution, currently in preview, using Microsoft Foundry that integrates with Dynamics 365 Business Central and Teams to streamline employee expense management.

Join the movement to the Frontier with Copilot, agents, and agentic business applications

We know that moving to the Frontier isn’t just about technology. That’s why we’re partnering with Harvard Business School to collaborate on research and executive education to help you put this into practice at your own company. We’re also sharing new resources for leaders on their journey to the Frontier Firm with Frontier Function Guides for Sales, HR, and IT and a look inside our learnings at Microsoft–including three ways to turn insight into action. We’re committed to helping you transform—we’ll see you at the Frontier!

If you’re interested in learning more:

• Explore the Microsoft Ignite 2025 sessions to see agents and agentic business applications in action.
• Register for our year-end webinar to learn about more agents in Dynamics 365 (December 2, 2025)
• Join us at the Convergence ERP Conference (December 9-11, 2025).

¹ Internal Microsoft sales team data based on time period January 1 to November 7, 2025. Total customers outreach by the agent: 61,734. Lead-to-opportunity ratio (sales qualification): 15.1%.

The post Microsoft Ignite 2025: Powering Frontier Firms with agentic business applications appeared first on Microsoft Dynamics 365 Blog.

In today’s hyper-competitive business landscape, sales leaders face a relentless challenge: how to drive growth, outpace competitors, and make smarter decisions faster in a resource constrained environment. Thankfully, the promise of AI in sales is no longer theoretical. With the advent of agentic solutions embedded in Microsoft Dynamics 365 Sales, including the Sales Research Agent, organizations are witnessing a transformation in how business decisions are made, and teams are empowered. But how do you know if these breakthrough technologies have reached a level of quality where you can trust them to support business-critical decisions?

Today, I’m excited to share an update on the Sales Research Agent, in public preview as of October 1, as well as a new evaluation benchmark, the Microsoft Sales Research Bench, created to assess how AI solutions respond to the strategic, multi-faceted questions that sales leaders have about their business and operational performance. We intend to publish the full evaluation package behind the Sales Research Bench in the coming months so that others can run these evals on different AI solutions themselves.

The New Frontier: AI Research Agents in Sales

Sales Research Agent in Dynamics 365 Sales empowers business leaders to explore complex business questions through natural language conversations with their data. It leverages a multi-modal, multi-model, and multi-agent architecture to reason over intricate, customized schemas with deep sales domain expertise. The agent delivers novel, decision-ready insights through narrative explanations and rich visualizations tailored to the specific business context.

For sales leaders, this means the ability to self-serve on real-time trustworthy analysis, spanning CRM and other domains, which previously took many people days or weeks to compile, with access to deeper insights enabled by the power of AI on pipeline, revenue attainment, and other critical topics.

Image: Screenshot of the Sales Research Agent in Dynamics 365 Sales

“As a product manager in the sales domain, balancing deep data analysis with timely insights is a constant challenge. The pace of changing market dynamics demands a new way to think about go-to-market tactics. With the Sales Research Agent, we’re excited to bridge the gap between traditional and time-intensive reporting and real-time, AI-assisted analysis — complementing our existing tools and setting a new standard for understanding sales data.“

Kris Kuty, EY LLP
Clients & Industries — Digital Engagement, Account, and Sales Excellence Lead

What makes the Sales Research Agent so unique? 

• Its turnkey experience goes beyond the standard AI chat interface to provide a complete user experience with text narratives and data visualizations tailored for business research and compatible with a sales leader’s natural business language.  
• As part of Dynamics 365 Sales, it automatically connects to your CRM data and applies schema intelligence to your customizations, with the deep understanding of your business logic and the sales domain that you’d expect a business application to have. 
• Its multi-agent, multi-model architecture enables the Sales Research Agent to build out a dedicated research plan and to delegate each task to specialized agents, using the model best suited for the task at hand.   
• Before the agent shares its business assessment and analysis, it critiques its work for quality. If the output does not meet the agent’s own quality bar, it will revise its work. 
• The agent explains how it arrived at its answers using simple language for business users and showing SQL queries for technical users, enabling customers to quickly verify its accuracy. 

Why Verifiable Quality Matters

Seemingly every day a new AI tool shows up. The market is crowded with offers that may or may not deliver acceptable levels of quality to support business decisions. How do you know what’s truly enterprise ready? To help make sure business leaders do not have to rely on anecdotal evidence or “gut feel”, any vendor providing AI solutions needs to earn trust through clear, repeatable metrics that demonstrate quality, showing where the AI excels, where it needs improvement, and how it stacks up against alternatives.

While there is a wide range of pioneering work on AI evaluation, enterprises deserve benchmarks that are purpose-built for their needs. Existing benchmarks don’t reflect 1) the strategic, multi-faceted questions of sales leaders using their natural business language; 2) the importance of schema accuracy; or 3) the value of quality across text and visualizations. That is why we are introducing the Sales Research Bench.

Introducing Sales Research Bench: The Benchmark for AI-powered Sales Research

Inspired by groundbreaking work in AI Benchmarks such as TBFact and RadFact, Microsoft developed the Sales Research Bench to assess how AI solutions respond to the business research questions that sales leaders have about their business data.¹

Read this blog post for a detailed explanation of the Sales Research Bench methodology as well as the Sales Research Agent’s architecture.

This benchmark is based on our customers’ real-life experiences and priorities. From engagements with customer sales teams across industries and around the world, Microsoft created 200 real-world business questions in the language sales leaders use and identified 8 critical dimensions of quality spanning accuracy, relevance, clarity, and explainability. The data schema on which the evaluations take place is customized to reflect the complexities of our customers’ enterprise environments, with their layered business logic and nuanced operational realities.

To illustrate, here are 3 of our 200 evaluation questions informed by real sales leader questions:

1. Looking at closed opportunities, which sellers have the largest gap between Total Actual Sales and Est Value First Year in the ‘Corporate Offices’ Business Segment?
2. Are our sales efforts concentrated on specific industries or spread evenly across industries?
3. Compared to my headcount on paper (30), how many people are actually in seat and generating pipeline?

Judging is handled by LLM evaluators that rate an AI solution’s responses (text and data visualizations) against each quality dimension on a 100-point scale based on specific guidelines (e.g., give score of 100 for chart clarity if the chart is crisp and well labeled, score of 20 if the chart is unreadable, misleading). These dimension-specific scores are then weighted to produce a composite quality score, with the weights defined based on qualitative input from customers, what we have heard customers say they value most. The result is a rigorous benchmark presenting a composite score and dimension-specific scores to reveal where agents excel or need improvement.[2]

[1] For more on TBFact: Towards Robust Evaluation of Multi-Agent Systems in Clinical Settings | Microsoft Community Hub and for more on RadFact: [2406.04449] MAIRA-2: Grounded Radiology Report Generation

[2] Sales Research Bench uses Azure Foundry’s out-of-box LLM evaluators for the dimensions of Text Groundedness and Text Relevance. The other 6 dimensions each have a custom LLM evaluator that leverages Open AI’s GPT 4.1 model. 100-pt scale has 100 as the highest score with 20 as the lowest. More details on the benchmark methodology are provided here

Running Sales Research Bench on AI solutions

Here’s how we applied the Sales Research Bench to run evaluations on the Sales Research Agent, ChatGPT by OpenAI, and Claude by Anthropic.  

• License: Microsoft evaluated ChatGPT by OpenAI using a Pro license with GPT-5 in Auto mode and Claude Sonnet 4.5 by Anthropic using a Max license. The licenses were chosen to optimize for quality: ChatGPT’s pricing page describes Pro as “full access to the best of ChatGPT,” while Claude’s pricing page recommends Max to “get the most out of Claude.”³ Similarly, ChatGPT’s evaluation was run using Auto mode, a setting that allows ChatGPT’s system to determine the best-suited model variant for each prompt.  
• Questions: All agents were given the same 200 business questions.  
• Instructions: ChatGPT and Claude were given explicit instructions to create charts and to explain how they got to their answers. (Equivalent instructions are included in the Sales Research Agent out of box.) 
• Data: ChatGPT and Claude accessed the sample dataset in an Azure SQL instance exposed through the MCP SQL connector. The Sales Research Agent connects to the sample dataset in Dynamics 365 Sales out of box.  

³ChatGPT Pricing and Pricing | Claude, both accessed on October 19, 2025

Results are in: Sales Research Agent vs. alternative offerings

In head-to-head evaluations completed on October 19, 2025 using the Sales Research Bench framework, the Sales Research Agent outperformed Claude Sonnet 4.5 by 13 points and ChatGPT-5 by 24.1 points on a 100-point scale.

Image: Sales Research Agent – Evaluation Results

¹Results: Results reflect testing completed on October 19, 2025, applying the Sales Research Bench methodology to evaluate Microsoft’s Sales Research Agent (part of Dynamics 365 Sales), ChatGPT by OpenAI using a ChatGPT Pro license with GPT-5 in Auto mode, and Claude Sonnet 4.5 by Anthropic using a Claude Max license.

Methodology and Evaluation dimensions: Sales Research Bench includes 200 business research questions relevant to sales leaders that were run on a sample customized data schema. Each AI solution was given access to the sample dataset using different access mechanisms that aligned with their architecture. Each AI solution was judged by LLM judges for the responses the solution generated to each business question, including text and data visualizations.

We evaluated quality based on 8 dimensions, weighting each according to qualitative input from customers, what we have heard customers say they value most in AI tools for sales research: Text Groundedness (25%), Chart Groundedness (25%), Text Relevance (13%), Explainability (12%), Schema Accuracy (10%), Chart Relevance (5%), Chart Fit (5%), and Chart Clarity (5%). Each of these dimensions received a score from an LLM judge from 20 as the worst rating to 100 as the best. For example, the LLM judge would give a score of 100 for chart clarity if the chart is crisp and well labeled, score of 20 if the chart is unreadable or misleading. Text Groundedness and Text Relevance used Azure Foundry’s out-of-box LLM evaluators, while judging for the other six dimensions leveraged Open AI’s GPT 4.1 model with specific guidance. A total composite score was calculated as a weighted average from the 8 dimension-specific scores. More details on the methodology can be found in this blog.

The Sales Research Agent outperformed these solutions on each of the 8 quality dimensions. 

Image: Evaluation Scores for Each of the Eight Dimensions

The Road Ahead: Investing in Benchmarks

Upcoming plans for the Sales Research Bench include using the benchmark for continuous improvement of the Sales Research Agent, running comparisons against a wider range of competitive offerings, and publishing the full evaluation package including all 200 questions and the sample dataset in the coming months, so that others can run it themselves to verify the published results and benchmark the agents they use. Evaluation is not a one-time event. Scores can be tracked across releases, domains, and datasets, driving targeted quality improvements and ensuring the AI evolves with your business.

Sales Research Bench is just the beginning. Microsoft plans to develop eval frameworks and benchmarks for more business functions and agentic solutions—in customer service, finance, and beyond. The goal is to set a new standard for trust and transparency in enterprise AI.

Why This Matters for Sales Leaders

For business decision makers, the implications are profound:

• Accelerated Decision-Making: AI-driven insights you can trust, when delivered in real time, enable faster, more confident decisions
• Continuous Improvement: Thanks to evals, developers can quickly identify areas for highest measurable impact and focus improvement efforts there
• Trust and Transparency: Rigorous evaluation means you can rely on the outputs, knowing they’ve been tested against the scenarios that matter most to your business.

The future of sales is agentic, data-driven, and relentlessly focused on quality. With Microsoft’s Sales Research Agent and the Sales Research Bench evaluation framework, sales leaders can move beyond hype and make decisions grounded in demonstration of quality. It’s not just about having the smartest AI—it’s about having a trustworthy partner for your business transformation.

The post Elevating Sales Performance with Microsoft’s Sales Research Agent: How Rigorous Evaluation Unlocks Trust and Transformation appeared first on Microsoft Dynamics 365 Blog.

Raising the bar for Enterprise AI

The Sales Research Agent in Dynamics 365 Sales automatically connects to live CRM data and can connect to additional data stored elsewhere, such as budgets and targets. It reasons over complex, customized schemas with deep domain expertise, and presents novel, decision-ready insights through text-based narratives and rich data visualizations tailored to the business question at hand.

For sales leaders, this means the ability to self-serve building out rich research journeys, spanning CRM and other domains, that previously took many people days or weeks to compile, with access to deeper insights enabled by the power of AI on pipeline, revenue attainment, and other critical topics.

But the market is crowded with offers that may or may not deliver acceptable levels of quality to support business decisions. How can business leaders know what’s truly enterprise ready? To help make sure customers do not have to rely on anecdotal evidence or “gut feel”, any vendor providing AI solutions must earn trust through clear, repeatable metrics that demonstrate quality, showing where the AI excels, where it needs improvement, and how it stacks up against alternatives.

Figure 1. The Sales Research Agent in the Dynamics 365 Sales Hub.

This post introduces the architecture and evaluation methodology and results behind Microsoft’s Sales Research Agent. Its technical innovations distinguish the Sales Research Agent from other available offerings, from multi-agent orchestration and multi-model support to advanced techniques for schema intelligence, self-correction and validation. In determining how best to evaluate the Sales Research Agent, Microsoft reviewed existing AI benchmarks and ultimately decided to create the Sales Research Bench, a new benchmark purpose-built to measure the quality of AI-powered Sales Research on business data, in alignment with the business questions, needs, and priorities of sales leaders. In head-to-head evaluations completed on October 19, 2025, the Sales Research Agent outperformed Claude Sonnet 4.5 by 13 points and ChatGPT-5 by 24.1 points on a 100-point scale.

Figure 2. Sales Research Bench Composite Score Results.

¹Results: Results reflect testing completed on October 19, 2025, applying the Sales Research Bench methodology to evaluate Microsoft’s Sales Research Agent (part of Dynamics 365 Sales), ChatGPT by OpenAI using a ChatGPT Pro license with GPT-5 in Auto mode, and Claude Sonnet 4.5 by Anthropic using a Claude Max license.

Methodology and Evaluation dimensions: Sales Research Bench includes 200 business research questions relevant to sales leaders that were run on a sample customized data schema. Each AI solution was given access to the sample dataset using different access mechanisms that aligned with their architecture. Each AI solution was judged by LLM judges for the responses the solution generated to each business question, including text and data visualizations. We evaluated quality based on 8 dimensions, weighting each according to qualitative input from customers, what we have heard customers say they value most in AI tools for sales research: Text Groundedness (25%), Chart Groundedness (25%), Text Relevance (13%), Explainability (12%), Schema Accuracy (10%), Chart Relevance (5%), Chart Fit (5%), and Chart Clarity (5%). Each of these dimensions received a score from an LLM judge from 20 as the worst rating to 100 as the best. For example, the LLM judge would give a score of 100 for chart clarity if the chart is crisp and well labeled, score of 20 if the chart is unreadable or misleading. Text Groundedness and Text Relevance used Azure Foundry’s out-of-box LLM evaluators, while judging for the other six dimensions leveraged Open AI’s GPT 4.1 model with specific guidance. A total composite score was calculated as a weighted average from the 8 dimension-specific scores. More details on the methodology can be found in the rest of this blog.

Microsoft will continue to use the evals in Sales Research Bench to drive continuous improvement of the Sales Research Agent, and Microsoft intends to publish the full evaluation package in the coming months, so others can run it to verify published results or benchmark the agents they use (example evals from the benchmark are included in this paper).

Sales Research Agent architecture

The architecture of the Sales Research Agent sets it apart from other offerings, delivering both technical innovation and business value.

1. Multi-Agent Orchestration: The Sales Research Agent uses a dynamic multi-agent infrastructure that orchestrates the development of the research blueprints, the text-based narratives and data visualizations accompanied by an explanation of the agent’s work. Specialized agents are invoked at each step in the journey to deliver domain-optimized insights for user questions, taking organizational data as well as business and user context into account.
2. Multi-Model Support: This multi-agent infrastructure enables each specialized agent to use the model that is best suited to the task at hand. Microsoft tests how each specialized agent performs with different models. Models are easily swapped out to continue optimizing the Sales Research Agent’s quality as the models available evolve over time.
3. Support for Business Language: There is a difference between business language (how business users naturally communicate) and natural language (any language that is not code). The Sales Research Agent can give quality answers to prompts in business language, because it breaks down the prompt into multiple sub-questions, building a research plan and using multi-step reasoning over connected data sources. Additionally, the Sales Research Agent is infused with knowledge of the Sales domain, so it can correctly interpret terminology and context that is only implicit to the user’s prompt.
4. Schema Intelligence: The Sales Research Agent can handle both out-of-the-box and customized enterprise schemas, adapting to complex, real-world environments. It has sophisticated techniques and heuristics built in to recognize the tables and columns that are relevant to the user query.
5. Self-Correction and Validation: The Sales Research Agent incorporates advanced auto-correction mechanisms for its generated responses. Whether producing SQL or Python code, the agent leverages sophisticated code correctors capable of iterative refinement—reviewing, validating, and amending outputs as needed. The correction loop begins with a fast, non-reasoning model to identify and fix straightforward issues. If errors persist, the system escalates to a reasoning model and, if required, a more powerful model to ensure deeper contextual understanding and precise correction. This dynamic, multi-model process helps to ensure that the final code is both accurate and reliable, enhancing the overall quality and trustworthiness of the agent’s insights and recommendations.
6. Explainability: The system tracks every agent interaction and decision, as well as the SQL query and Python code generated to produce the research blueprint. The Sales Research Agent uses this information to help users quickly verify its accuracy and trace its reasoning. Each blueprint includes Show Work, an explanation in simple language for business users, with an advanced view of SQL queries and more details for technical users.

Figure 3. A high-level diagram of Sales Research Agent’s architecture and how it connects to business workflows

Why Enterprise Sales Requires a New Evaluation Framework

In traditional software, unit tests give repeatable proof that core behaviors work and keep working. For AI solutions, evaluations (evals) are needed to demonstrate quality and track continuous improvement over time.

Enterprises deserve evaluations that are purpose-built for their needs. While there is a wide range of pioneering work on AI evaluation, existing benchmarks miss key attributes that are needed for an AI solution to guide critical business decisions:

• The benchmark must reflect the strategic, multi-faceted business questions of sales leaders using their business language.
• The benchmark must measure schema accuracy: whether the system correctly handles tables, columns, and joins on system of record schemas that can be highly customized.
• The benchmark should assess insights across both text-based narratives and data visualizations, reflecting the outputs with which leaders make decisions.

Introducing Sales Research Bench for AI-powered Sales Research

To meet these demands, Microsoft developed the Sales Research Bench, a composite quality score built to evaluate AI-powered Sales Research solutions in close alignment with customers’ actual questions, environments, and priorities. From engagements with customer sales teams across industries and geographies, Microsoft identified the critical dimensions of quality and created real-world business questions in the language sales leaders use. The data schema on which the evaluations take place is customized to reflect the complexities of customers’ enterprise environments, with their layered business logic and nuanced operational realities. The result is a rigorous benchmark presenting a composite score based on 8 weighted dimensions, as well as dimension-specific scores to reveal where agents excel or need improvement.

Benchmark Methodology

The evaluation infrastructure for Sales Research Bench includes:

• Eval Datasets: 200 business questions in the language of sales leaders, each associated with its own set of ground-truth answers for validation.
• Sample enterprise dataset: Eval questions run on a customized schema, reflecting the complexities of enterprise environments.
• Evaluators: LLM-judge-based evaluation, tailored for each of the 8 quality dimensions described below. Azure Foundry out-of-box evaluators are used for Text Groundedness and Text Relevance. For the other 6 dimensions, OpenAI’s GPT 4.1 model is used with specific guidelines on how to score answers, which are provided in the appendix.

Here are 3 of the 200 evaluation questions informed by real sales leader questions: 

• Looking at closed opportunities, which sellers have the largest gap between Total Actual Sales and Est Value First Year in the ‘Corporate Offices’ Business Segment? 
• Are our sales efforts concentrated on specific industries or spread evenly across industries? 
• Compared to my headcount on paper (30), how many people are actually in seat and generating pipeline?  

Dimensions of Quality

The Sales Research Bench aggregates eight dimensions of quality, weighting them as shown in the parentheses below to reflect what we have heard customers say they value most in AI tools for sales research during their engagements with Microsoft.

• Text Groundedness (25%): Ensures narratives are accurate, faithful to the sample enterprise data, and applying correct business definitions.
• Chart Groundedness (25%): Validates that charts accurately represent the underlying data from the same enterprise dataset.
• Text Relevance (13%): Measures how relevant the insights in the text-based narrative are to the business question.
• Explainability (12%): Ensures the AI solution accurately and clearly explains how it arrived at its responses.
• Schema Accuracy (10%): Verifies the correct selection of tables and columns by evaluating whether the generated SQL query is consistent with the tables, joins, and columns in the ground-truth answers. (Business applications typically consist of approximately 1,000 tables, many featuring around 200 columns, all of which can be highly customized by customers.)
• Chart Relevance (5%): Validates whether the data and analysis shown in the chart are relevant to the business question.
• Chart Fit (5%): Evaluates if the chosen visualization matches the analytical need (e.g., line for trends, bar for comparisons).
• Chart Clarity (5%): Assesses readability, labeling, accessibility, and chart hygiene.

Each of these dimensions received a score from an LLM judge from 20 as the worst rating to 100 as the best. For example, the LLM judge would give a score of 100 for chart clarity if the chart is crisp and well labeled, score of 20 if the chart is unreadable or misleading. 

Sample Enterprise Dataset

Evaluation needs representative conditions to be useful. Through customer engagements, Microsoft identified numerous edge cases from highly customized schemas, complex joins and filters, and nuanced business logic (like pipeline coverage and attainment calculations).

For instance, most customers customize their schemas with custom tables and columns, such as replacing an industry column with an industry table, and linking it to the customer object, or adding market and business segment instead of using an existing segment field. As a result, their environments often contain both the out-of-box tables and columns as well as customized tables and fields, all with similar names. By systematically incorporating these edge cases into the sample custom schema, Sales Research Bench evaluates how agents perform outside of the “happy path” to assess enterprise readiness.

Figure 4. Example evaluation case (see the Appendix for more examples)

Evaluating Sales Research Agent and Other Solutions

In addition to the Sales Research Agent, Microsoft evaluated ChatGPT by OpenAI using a Pro license with GPT-5 in Auto mode and Claude Sonnet 4.5 by Anthropic using a Max license. The licenses were chosen to optimize for quality: ChatGPT’s pricing page describes Pro as “full access to the best of ChatGPT,” while Claude’s pricing page recommends Max to “get the most out of Claude.”[1] Similarly, ChatGPT’s evaluation was run using Auto mode, a setting that allows ChatGPT’s system to determine the best-suited model variant for each prompt.

Microsoft implemented a controlled evaluation environment where all systems – Sales Research Agent, ChatGPT-5, and Claude Sonnet 4.5 worked with identical questions and data, but through different access mechanisms aligned with their respective architectures.

The Sales Research Agent has a native multi-agent orchestration layer that connects directly to Dynamics 365 Sales data. This allows it to autonomously discover schema relationships and entity dependencies, and to perform natural-language-to-query reasoning natively within its own orchestration stack.

Since ChatGPT and Claude do not support relational line-of-business source systems out of box, Microsoft enabled access to the same dataset by mirroring it into an Azure SQL instance. Mirroring was done to preserve all the data types, primary keys, foreign keys, and relationships between tables from Dataverse to Azure SQL. This Azure SQL copy was exposed through the MCP SQL connector, ensuring that ChatGPT and Claude retrieved the exact same data but through a standardized external interface. Once responses were captured, they were evaluated using the same evaluators against the exact same evaluation rubrics.

Finally, prompts to ChatGPT and Claude included instructions to create charts and to explain how they got to their answers (Sales Research Agent has this functionality out of box.)

[1] ChatGPT Pricing and Pricing | Claude, both accessed on October 19, 2025

Evaluation Results

In a test of 200 evals on the customized schema, Sales Research Agent earned a composite score of 78.2 on a 100-point scale, while Claude Sonnet 4.5 earned 65.2 and ChatGPT-5 earned 54.1.

The chart below presents the Sales Research Bench composite scores, with scores for each dimension overlaid on the bars within the stacked bar chart.

Figure 5. Sales Research Bench Composite Scores with Dimension-specific Scores.

Breaking this down, the Sales Research Agent outperformed other solutions on all 8 dimensions, with the biggest deltas in chart-related dimensions (groundedness, fit, clarity, and relevance), and the smallest deltas in schema accuracy and text groundedness. Claude Sonnet 4.5 outperformed ChatGPT-5 on all 8 dimensions, with the biggest delta in chart clarity and the smallest delta in chart relevance.

Figure 6. Sales Research Bench Scores by Dimension.

Looking Ahead

Sales Research Agent introduces a new generation AI-first business application that transforms how sales leaders can approach and solve complex business questions. The Sales Research Bench was created in parallel to represent a new standard for enterprise AI evaluation: Rigorous, comprehensive, and aligned with the needs and priorities of sales leaders.

Upcoming plans for the Sales Research Bench include using the benchmark for continuous improvement of the Sales Research Agent, running further comparisons against a wider range of competitive offerings, and publishing the eval package so customers can run it themselves to verify the published results and benchmark the agents they use. Evaluation is not a one-time event. Scores can be tracked across releases, ensuring that AI solutions evolve to meet customer needs.

Looking beyond Sales Research Bench, Microsoft plans to develop eval frameworks and benchmarks for more business functions and agentic solutions— in customer service, finance, and beyond. The goal is to set a new standard for trust and transparency in enterprise AI.

Appendix:

Scoring Guidelines provided to LLM Judges 

Text Groundedness and Text Relevance used Azure Foundry’s out-of-box LLM evaluators. Below are the guidelines provided to the LLM judges for the other six quality dimensions. These judges leverage Open AI’s GPT 4.1 model. 

Schema accuracy: 

• 100: Perfect match – all golden tables and columns are present (extra columns OK, Dynamics equivalents OK) 
• 80: Very good – minor missing columns or one missing table 
• 60: Good – some important columns or tables missing but core schema is there 
• 40: Fair – significant schema differences but some overlap 
• 20: Poor – major schema mismatch or completely different tables 

Explainability: 

• 100 (Excellent): Explanation is highly detailed, perfectly describes what the generated SQL does, technically accurate, and provides clear business context 
• 80 (Good): Explanation is sufficiently detailed and mostly accurate with minor gaps in describing the SQL operations 
• 60 (Fair): Explanation provides adequate detail but misses some important SQL operations or has minor inaccuracies 
• 40 (Poor): Explanation lacks sufficient detail to understand the SQL operations or has significant inaccuracies 
• 20 (Very Poor): Explanation is too vague, mostly incorrect, or provides insufficient detail about the generated SQL 

Chart Groundedness: 

• 100: Data accurately matches ground truth OR both ground truth & chart empty 
• 80: Minor data inaccuracies 
• 60: Some data inaccuracies 
• 40: Major data inaccuracies 
• 20: data completely mismatches ground truth 

Chart Relevance: 

• 100: Question and chart strongly reinforce each other OR both ground truth & chart empty
• 60: Question and chart loosely align but with some disconnect 
• 20: Question and chart do not align at all 

Chart Fit: 

• 100: Optimal chart choice for the task OR both ground truth & chart empty (appropriate emptiness) 
• 60: Acceptable chart choice but not optimal for the task 
• 20: inappropriate/confusing chart type 

Chart Clarity: 

• 100: Chart is crisp and well-labeled OR both ground truth & chart empty
• 60: Chart readable but missing labels/clarity elements 
• 20: Chart unreadable, misleading 

Examples of Evaluation dataset:

Below are some of the evaluation datasets that we have used to benchmark the performance of Sales Research Agent against all the evaluation rubrics mentioned above. These same questions were also evaluated against the competitive offerings.

Click on the + to see the full datasets.

The post Sales Research Agent & Sales Research Bench appeared first on Microsoft Dynamics 365 Blog.

As innovation speeds up, staying agile is essential. To keep your business ahead of the curve with innovation across Microsoft Dynamics 365, Microsoft Power Platform, and Copilot Studio, join us for the Business Applications Launch Event, debuting live on the Dynamics 365 YouTube channel on Thursday, October 23, 2025 at 9 AM Pacific Daylight Time. Subscribe to our YouTube channel to get notified when the update is live.

The Business Applications Launch Event offers an exclusive first look at new capabilities launching over the next few months.

With a newly streamlined presentation format, you can quickly get up to speed on the most important and innovative capabilities—with expert insights and demonstrations from Microsoft product leaders and engineers. It’s our way of helping you stay current, make informed decisions, and move faster in the era of Copilot and AI agents.

The update is your opportunity to:

• Get insights about the latest low-code and AI innovation transforming business from Charles Lamanna, President, Business and Industry Copilot.
• Get a sneak preview of upcoming capabilities across Dynamics 365, Microsoft Power Platform, and Copilot Studio with live demonstrations from Microsoft product leaders and engineers.
• Discover where to access materials to learn about and plan for new and upcoming capabilities.
• All in a new presentation format designed to quickly get you up to speed on the latest updates, so you can get the most from them.

During this update, you’ll hear from the product leaders and engineers behind the technology, including new Copilot and AI agent innovation for Dynamics 365 and Microsoft Power Platform. Demo highlights will include:

Microsoft Dynamics 365 Sales

Learn about updates to the Sales qualification agent. It autonomously researches and engages with leads, helping sales teams quickly identify those with real purchase intent. In this wave, the agent goes further—moving the lead closer to full qualification and boosting the team’s opportunity pipeline with greater precision and impact.

Microsoft Dynamics 365 Customer Service and Dynamics 365 Contact Center

The latest release wave of Dynamics 365 Contact Center helps service reps better understand customer needs and deliver what they need—quickly, efficiently, and with a human touch. Dynamics 365 Customer Service will continue to enhance agentic and Copilot capabilities for case and knowledge management, as well as AI-based routing. Dynamics 365 Contact Center will also focus on expanding agentic and Copilot capabilities to automate service journeys across digital and voice channels, along with introducing new omnichannel and supervisor features in the 2025 release wave 2.

Dynamics 365 ERP products and solution in Microsoft 365 Copilot

Dynamics 365 Finance expands the capabilities of the Account Reconciliation Agent. Today, it supports your team in effortlessly resolving voucher amount mismatches. In this wave, it extends support to include ledger not in subledger and subledger not in ledger exceptions. Instead of relying on manual exception handling and static reports, the solution reviews all transactions on an ongoing basis, services exceptions, and presents them to you. The agent then suggests the most appropriate action for resolution, and you have the freedom to accept it or choose another path. Core updates to Dynamics 365 Finance also include the automation of remittance advice processing.

Dynamics 365 Supply Chain Management introduces capabilities that make AI-led demand planning more flexible. You can now bring in multiple external signals like inflation, weather, and industry indexes right into your forecast.

And new autonomous and intelligent productivity capabilities for finance solution in Microsoft 365 Copilot will reshape the finance process, from reconciliation to collections to advanced analytics help reduce repetitive work and surface actionable insights.

Explore our updates in Dynamics 365 Supply Chain Management and Finance agents for Microsoft 365:

• Dynamics 365 Supply Chain Management—Improve customer impacting predictions with the Data Quality Agent
• Finance agents for Microsoft 365—Automate finance tasks to improve accuracy and efficiency

Microsoft Power Platform

Microsoft Power Platform is getting a major boost with AI and collaboration features. Power Apps now lets people and agents work together—agents can help with tasks like data entry, visualization, and app creation just by describing what you need or sharing an image. Power Automate is evolving with smarter automation tools, including generative AI actions, intelligent document processing, and new human-in-the-loop experiences like advanced approvals. It’s also, it’s adding stronger governance and security controls to help manage automation at scale. Power Pages is making it easier than ever to build secure, data-driven websites, with new tools for low-code makers and developers, and enhanced security insights to keep everything protected.

See wave two highlights for Power Automate—Proactively spot and resolve automation issues

Copilot Studio

Copilot Studio continues to make agent creation and operation even easier and more powerful with autonomous agents in Microsoft 365 Copilot, the ability to build complete teams of agents that work seamlessly together, and improved governance for enterprise scalability. Copilot Studio will offer even deeper integration with Azure AI Foundry and Microsoft Graph, helping ensure your agents can use the latest AI technology alongside your data in the Microsoft Graph.

Catch the wave—Mark your calendar for BALE

The Business Applications Launch Event will be live on the Dynamics 365 YouTube channel on Thursday, October 23, 2025, starting at 9 AM Pacific Daylight Time. We’ll see you there!

The post Explore new AI innovation for Dynamics 365, Microsoft Power Platform, and Copilot Studio at the Business Applications Launch Event appeared first on Microsoft Dynamics 365 Blog.