That may sound pretty handy, but all those tools can also be a burden, especially for the engineers who have to keep them sharp. It can also be confusing for employees to have to switch between different tools and user interfaces for filing expenses.

Microsoft has built up its collection of corporate expense tools over the years, onboarding various internal and third-party expense management platforms through acquisitions and because of different business needs. As the weight grew heavier, so did the need to address their maintenance issues.

“We had multiple tools that needed streamlining,” says Amruta Anawalikar. She is a senior program manager for Microsoft Commerce Financial Services (CFS) in Finance Engineering, the team under Azure Cloud + AI that manages expenses at Microsoft. “We wanted to reduce employee productivity costs and improve operations, engineering capacity, and business capacity.”

[Find out how Microsoft is creating efficiencies in finance with Microsoft Dynamics 365 and machine learning. Learn more about migrating critical financial systems to Microsoft Azure.]

Flipping the design

These disconnected systems were working across 110 countries and regions with no synchronization, each with its own vertical infrastructure. The system to manage them all had been built up piece by piece.

Because of varying configurations specific to local environments, deployments were no small task. Expense categories, policies, and payment rules differed across tools. None of the systems even spoke the same language—“report name” in one tool might be called “report description” or “expense purpose” in another.

Solving for these issues through tool customizations wasn’t an option, either, because one of the primary tools (a third-party product) didn’t support the API integrations needed for that kind of flexibility.

Standardization and unification were badly needed.

We’ve flipped the whole design. By generalizing the entire ecosystem and focusing on a modern Azure-based design, it allows multiple tools to exist.

– Sumeet Deshpande, Microsoft CFS Finance Engineering

“Historically, the efforts were really focused locally within each tool and over the years, each ended up having its own over-customized, tool-centric designs built around them,” says Sumeet Deshpande. He is a principal software engineering manager for CFS Finance Engineering. “Even though the systems were very similar, we couldn’t leverage the same components.”

Deshpande’s team set out to create a unified expense management experience that put engineering before process.

“We’ve flipped the whole design,” Deshpande says. “By generalizing the entire ecosystem and focusing on a modern Azure-based design, it allows multiple tools to exist, and the problems we were having in over-aligning to the specific tools have gone away.”

The modernization journey

Microsoft’s journey to build a modern backend pipeline to unify its corporate expense tools began two years ago, with a milestone that became a catalyst for change.

In 2019, the legacy tool MSExpense 1.0 was being retired. Faced with 110 countries and regions to migrate (and myriad tax and statutory regulations to go with them), the team expected the migration to MSExpense 2.0 to take two to three years and cost an estimated $2 million. But then they decided to try a new strategy.

“That’s the point where we started thinking differently in how our tools need to be either onboarded or retired,” says Mohit Jain, a senior software engineer who led the retirement for CFS Finance Engineering.

They started breaking big problems down into smaller pieces, dividing countries and regions into buckets based on their level of complexity and tackling the migration one at a time.

The entire migration took just six months, and cost nothing.

“This was a really important part of our journey and how we approached problems going forward,” Jain says.

Building on that momentum, the team implemented a major overhaul of the user experience in 2020 and introduced OneExpense, automating much of the process with built-in machine learning to essentially eliminate the need for employees to file expense reports at all.

According to Deshpande, that’s what set the stage for remaking the back end.

“MSExpense 1.0 was retired at rocket speed,” he says. “We built on that and delivered end-to-end automation. That was a powerful story for leadership—people started to listen to us after that.”

Meet the hero: Microsoft Azure’s architecture

With a firm engineering mindset and funding to move forward, the mission to modernize was on.

Microsoft Azure provided the cloud base that would help the team achieve their internet-first, top-down goal. Whereas the structures of the third-party expense tools were locked, Microsoft Dynamics 365, the company’s powerful suite of business solutions software, swooped in to stretch their flexibility.

Key to that flexibility was building a disconnected architecture that allowed the team to create a plug-and-play modular design that enabled any individual system to be swapped in or out.

“We can literally replace any component, with Dynamics 365 as a unified mechanism for any expense tool that exists,” Deshpande says. “Our primary hero is the modern Azure-based design, which really synergizes reusability across expense tools and allows multiple tools to exist because they have been there for a valid reason. The most important piece within it is Dynamics 365, but that’s just one part of the puzzle.”

Standardization was also a fundamental piece.

As part of the earlier automation project, Anawalikar had gathered extensive market research data to standardize policies. By reaching out to different markets as well as other large tech companies to investigate the sources of governance, they were able to trim a lot of fat and create a universal set of standards and definitions.

This enabled Jain to create what they called the OneExpense contract: a common data language that would allow the siloed systems to understand each other.

“Whatever expense tool you run your infrastructure on, eventually we converted that tool-specific data to a OneExpense contract,” Jain says. “It was a very important piece to make a contract that’s tool agnostic—that’s what we need for our downstream needs.”

Anawalikar calls it something else: “The holy grail.”

“Everything in this ecosystem, right from automation to expense reporting, is based on the expense contract,” she says. “We need to protect it at all costs.”

Those vertical structures in the architecture that made management of the corporate expense tools so labor-intensive were up for disruption as well, and were restructured horizontally.

This is where Dynamics 365 really has an edge. It gives us that flexibility of customization, which will make it extendable for any future needs as well.

– Mohit Jain, senior software engineer, Microsoft CFS Finance Engineering

“When they’re horizontal, they’re the same across different systems,” Jain says. “Even with different tools, the overall experience remains the same because the tool-specific data was converted to a generic OneExpense contract.”

At the core of the system, the team built its Data Integration Service: an orchestrator that controls various microsystems for calculating taxes, sending statutory forms, and sending emails to approvers. Functioning much like an orchestra conductor, the orchestrator queues each system at the right time to perform the right function, sometimes each playing solo, and other times in harmony with other instruments.

“Adding up all of those microservices would have been really difficult in any other system,” Jain says. “This is where Dynamics 365 really has an edge. It gives us that flexibility of customization, which will make it extendable for any future needs as well.”

The tool belt of the future

The Finance Digital team is already enjoying life in the modern expense management world.

Deployments are easy, having gone from 100 percent manual (with a lot of typos) to standardized configurations deployed with 90 percent automation.

Modular plug-and-play components include a tax calculation system and a centralized automated auditing system that’s more efficient at flagging errors. Plus, self-monitoring and self-healing tools detect and fix issues behind the scenes before they’re flagged by employees filing help tickets.

With all of this comes greatly reduced operational costs, higher productivity, and employees who no longer have to view expense filing as a laborious, confusing task that interrupts the main focus of their jobs.

As Anawalikar says, one of the primary goals achieved is to “make expenses less expensive.”

“This expense story today is a showcase of everything Azure Cloud + AI has to offer, namely the power of Azure Cloud, the use of AI and ML, and the use of Dynamics 365,” Deshpande says.

What gets Deshpande and his team really excited as they iterate further is the architecture’s modular ability to work with any corporate expense tool.

“If Dynamics 365 evolves and rolls out multiple advanced versions, it will still work,” Deshpande says.

Jain is also looking forward to a future state in which the architecture is even lighter-weight and Microsoft Dynamics 365 is truly functioning as the Swiss Army knife of all corporate expense tools.

Along the way, the team has experienced what it feels like to be part of a team that’s empowered by research, data, and its own supportive leadership.

“We were encouraged to challenge the status quo and ask questions,” Jain says. “We used disruption for a positive outcome.”

• Find out how Microsoft is creating efficiencies in finance with Microsoft Dynamics 365 and machine learning.
• Learn more about migrating critical financial systems to Microsoft Azure.

The post Transforming Microsoft’s corporate expense tools with Microsoft Azure and Microsoft Dynamics 365 appeared first on Inside Track Blog.

Most people don’t give much thought to printing.

In the best-case scenario, you select a button and your paper comes out. Other times, you might have to fiddle with locating printers, driver installations, and of course, the occasional paper jam. There are good reasons why this most humble of office essentials is also a common symbol of office frustrations.

IT administrators like Kathren Korsky think about printers a lot more than most.

As a senior service engineering manager for End User Services at Microsoft, Korsky oversees their organization’s printing strategy and infrastructure. That means maintaining print servers, ensuring connectivity, managing security permissions, and staying on top of compatibility issues with a broad network of third-party hardware partners.

It also means dealing with the security risk printer servers create.

How do printers create such challenges?

Before, anyone who wanted to print in a Microsoft office had to connect to Microsoft’s corporate network. That meant giving them VPN access just so they could print something.

“Corpnet is a very precious corporate asset, and VPN access ends up being a security liability,” Korsky says. “We must eliminate our print service dependency on VPN to achieve our strategic Zero Trust goals.”

Adding to these acute pains were the everyday aches of Microsoft branch offices without corpnet connections at all, where employees were severely constrained when attempting to print to a shared printer, not to mention the maintenance and high energy costs that physical servers consume.

Then about four years ago, Microsoft Digital began migrating all of its internal servers to the cloud, a project that transitioned 95 percent of its physical servers to Microsoft Azure virtual machines (VMs).

[Learn how Microsoft used Azure to retire hundreds of physical branch-office servers. Find out how Microsoft enabled secure and compliant engineering with Azure DevOps. Unpack seamless and secure cloud printing with Universal Print.]

Connecting printers to the cloud

Korsky’s team joined that cloud migration, and over four years they reduced the company’s 320 on-premises print servers around the world to around 80 Microsoft Azure print server VMs. The team benefited from Microsoft Azure’s security and management capabilities while achieving a print server uptime improvement to nearly 100 percent.

Korsky says the 70 hours per month their team formerly spent patching servers has been reduced to seven.

While the move to Infrastructure as a Service (IaaS) delivered great benefits for the print service, that was not enough. The team needed a solution that could work completely in the public internet space and draw on the advantages of becoming a Platform as a Service (PaaS) approach, which was going to be the next step in the print service transformation.

Working together with Microsoft’s Azure + Edge Computing team, they experimented with a previous offering, Hybrid Cloud Print, but felt that more was needed to simplify the administrator’s experience.

Seeing an opportunity, Korsky and their team knew the moment was ripe for a major transformation that would not only greatly reduce their administrative overhead, but also eliminate those pesky corpnet dependencies while enabling public internet connectivity in a safe and secure way.

Working together, Microsoft Digital and Azure + Edge Computing teams built in robust management capabilities and easily accessible data insights and reporting, and a new printing experience called Universal Print was born.

As Universal Print began to roll out to groups across Microsoft, beginning with the Azure + Edge Computing team, one of the challenges was the wide variety of different brands, makes, and models of printers that would need to integrate with the service.

“We as a product group wanted to support a broad set of currently available printers in market, and some of them are quite old,” says Jimmy Wu, a senior program manager for Azure + Edge Computing who worked with Korsky’s team to deploy Universal Print into the Microsoft infrastructure. “The challenge was how do we do that when our service isn’t even publicly available at the time.”

As a solution, they created a piece of connector software that served as a communication proxy between the physical printer and the cloud service. It’s now available to customers as part of their Universal Print subscription.

With the migration and product rollout complete, Universal Print was validated in private preview by Microsoft customers who also saw a need for a cloud print service. It then moved into public preview in July.

Printers are now being published in Microsoft Azure Active Directory through a centralized portal, with little need for on-premises infrastructure or maintenance.

What’s more, the elimination of on-premises servers and all the physical space, energy consumption and cooling systems that go with it help support Microsoft’s commitment to achieve carbon negativity by 2050.

For branch office managers grappling with whether to invest in costly corporate network setups, Korsky says, “it solves for some real business decisions that companies have to make about branch office locations.”

And the employee who just needs to print? They can think about it even less.

“What’s really great is that our users benefit from a seamless, familiar print experience,” Korsky says. Users click a button and their paper comes out—without all the interference of printer discovery, network permissions and driver installations standing in their way.

Universal Print in a remote world

The ability to print via the cloud has proven to be an unexpected boon to businesses and organizations who have had to quickly adapt to operating remotely.

Alan Meeus, a product marketing manager for Microsoft 365 Modern Work, says that of the more than 2,000 external customers currently testing Universal Print, many have accelerated their adoption amid COVID-19.

“Even with people working remotely, there are many use cases for why print is still important,” Meeus says. “There’s a lot of printing going on in critical industries like healthcare, manufacturing, distribution and education. In schools, some kids don’t have access to computers and they still rely a lot on printed materials.”

Universal Print has also helped enable Microsoft 365 users to perform work functions at home that they previously couldn’t.

“If our HR or payroll department needs to run checks, they can do that from home,” says Scott Hetherington, a senior systems analyst for the Wild Rose School Division in Alberta, Canada. “Being able to give them Universal Print right now has been a lifesaver. And it’s been able to help keep people safe in the face of a pandemic by keeping them home as much as possible.”

As more organizations ramp up adoption, the Universal Print team and their partners are looking forward to cultivating a circular feedback loop where they’re gathering feedback from the community and delivering the kinds of improvements customers want. They’re also working towards a longer-term vision of evolving from the IaaS cloud service model for the connector software to going completely serverless, requiring no infrastructure management at all.

For Korsky, it’s all about the growth mindset.

“This has been an amazing journey of experimentation to learn what works well and where changes are required. And we’re partnering in a more collaborative way,” Korsky says. “We took our learnings from Hybrid Cloud Print and came up with this whole new approach that is even better than we originally envisioned, and we’re having great success.”

The printing transformation is making a difference with Korsky’s peers across Microsoft.

“My team’s amazing partnerships with engineering teams across Microsoft allow us to develop impactful internal solutions that also benefit our customers,” says Dan Perkins, a principal service engineering manager in Microsoft Digital’s End User Services. “Universal Print simplifies how we manage our work and reduces the time we spend maintaining our infrastructure. It also improves the security of our print service. We are excited about what the future holds for this transformational offering.”

• Unpack seamless and secure cloud printing with Universal Print.
• Learn how Microsoft used Azure to retire hundreds of physical branch-office servers.
• Find out how Microsoft enabled secure and compliant engineering with Azure DevOps.

The post Jamming to a new tune: Transforming Microsoft’s printing infrastructure with Universal Print appeared first on Inside Track Blog.

Imagine sifting through hundreds of photos of cupcakes to find two images of the same cupcake taken in the same place at the same time, within minutes. That’s one of the ways that Microsoft’s Audit, Risk, and Compliance (ARC) team made sure that the invoices the company is paying are accurate and legitimate—and it’s now able to do that better and at larger scale thanks to a new audit digitization project powered by machine learning (ML).

When an external company submits a payment invoice to Microsoft, they must provide what’s called proof of execution (POE) to Microsoft invoice approvers proving that the service was indeed performed. Hence the virtual mountain of visual evidence of not just cupcakes, but entire lunch buffets, swag orders, promotion campaigns, and more, contained in files of Microsoft PowerPoint decks, PDFs, and Microsoft Word documents.

Microsoft’s Audit team in Microsoft Finance periodically reviews subsidiaries to, among other things, make sure that invoice approvers, vendors, and suppliers are following company procurement policies and processes to protect company assets and interests.

Matching cupcake photos are a flag of possible recycled POE—signaling that a vendor may have reused an existing image, which is not considered legitimate proof of the service provided.

Fei Guo is a senior data solutions manager for Microsoft Strategy and Solutions Technology, which is dedicated to analyzing business needs for the ARC team and connecting them with solutions. She was tasked by her business users to explore solutions for detecting recycled POEs using machine learning so they could easily find those needles in the haystack.

“Finding reused POE documents is extremely difficult for a human to do,” Guo says. Because manually comparing millions of images isn’t possible, auditors would typically test POs based on random samples. “We needed a way to detect similar images at scale.”

[Find out how Microsoft applied Azure Cognitive Services to automate partner claim validation. Learn about automating revenue processing at Microsoft with Power Automate.]

The Goldilocks algorithm

Guo brought the challenge to an engineering team in Microsoft Audit and Compliance within the Microsoft Cloud + AI organization.

They formed a volunteer team to explore audit digitization using machine learning built entirely on Microsoft Azure as a project for a Microsoft Hackathon, a yearly event where employees from across the company are invited to come up with freeform projects and solutions in an intensive three-day session.

We know recycled POE is an industry-wide problem. We could solve it using new technology in areas of ML that I was excited to learn about.

– Anuj Bansal, principal software engineer, Microsoft

“We know recycled POE is an industry-wide problem,” says Anuj Bansal, a principal software engineer for Commerce Financial Systems (CFS) who joined the Hackathon. “We could solve it using new technology in areas of ML that I was excited to learn about.”

If it worked, audit digitization using Azure technology had the potential to transform the auditing process, from just using samples at a minuscule proportion of the actual data volume to enabling the company to audit 100 percent of its data with greater accuracy in far less time.

The team presented a proof of concept for their solution at the Hackathon, which they called the Recycled POE tool, and won leadership support to further develop it into reality.

The first part was figuring out how to extract the data from the invoices, given the many different types of files and formats. Each POE typically generates around 10–12 images. ARC tests 10,000 purchase orders on average each year, and the volume grows by around 1 million images annually. They applied Microsoft Azure Cognitive Services to extract images from MS Invoice POE files to standardize the process across all document types.

Next came the bigger challenge: finding the most accurate algorithm.

They built a custom machine learning tool using an algorithm called a Hierarchical Navigable Small World (HNSW) graph to calculate the similarity score between images.

We’re transforming audit from reacting and sampling to a more live approach where you audit in a way that humans can’t.

– Jose De Jesus Sanchez Rico, senior software engineering manager, Microsoft

“We went through a lot of the standard algorithms,” Bansal says. “Some were too slow; some were too sensitive. Others were faster, but we saw too many false positives and didn’t get the best results. It was a journey to figure out which algorithm was the right one for us.”

With the “just right” algorithm, images are now processed using the HNSW graph in batches and saved to a Microsoft Azure SQL database. All of this runs in the background while auditors can focus on other tasks, and the results are delivered to them in a Microsoft Power BI report.

After nearly a year experimenting and fine-tuning, the new ML features were launched last September. The system can process 21,000 images per minute, allowing large-scale detection of similar images submitted as POEs for different purchase orders.

“The significance of this is that we’re transforming audit from reacting and sampling to a more live approach where you audit in a way that humans can’t,” says Jose De Jesus Sanchez Rico, a senior software engineering manager for Microsoft’s Core Functional Engineering who managed the team that developed the Hackathon project into production. “It’s having that perspective of what only ML can do for you.

Human learning

There’s still more that audit digitization can do for Microsoft.

The team is working on fine-tuning the results further by filtering out noise in the data such as logo images contained in documents and email signatures. The ML model can be strengthened still by feeding it more precision image metadata like time stamps and geotags.

The Recycled POE tool has also been recently integrated with another innovation of the audit digitization journey: a Microsoft Teams audit digitization assistant bot that auditors can access to perform common functions.

Taking the solution a step further, the team is also hoping to extend the Recycled POE tool to be used by invoice approvers as a proactive compliance check, rather than a reactive one, to catch mistakes before they’re made in the first place.

Jagannathan Venkatesan, a principal group engineering manager for Microsoft Finance Management, which oversees the Audit and Compliance team, says that he was especially impressed by the Hackathon team’s perseverance and eagerness to learn more about machine learning.

“There were setbacks and challenges from a performance perspective, but the team didn’t give up,” Venkatesan says. “We worked with the partner engineering teams, got guidance on how to improve, and we pulled it through.”

That makes them even better prepared for whatever comes next.

“We work in Finance, and we work with a lot of data, so data science and ML have always been areas where we want to learn,” Venkatesan says. “So, we took this as an opportunity to make an impact but at the same time, educate the team so that my engineering workforce is prepared for the technology shifts of tomorrow.”

• Find out how Microsoft applied Azure Cognitive Services to automate partner claim validation.
• Learn about automating revenue processing at Microsoft with Power Automate.

The post Boosting internal audits at Microsoft with audit digitization, machine learning appeared first on Inside Track Blog.