We know when writing formulas, errors are unavoidable, and troubleshooting could be time-consuming and cumbersome even for tiny errors like a missing parenthesis. With formula repair, now you may skip some of the troubleshooting effort as whenever you have a formula error, Power Apps will try to come up with a fix suggestion leveraging the power of AI. Once there’s a fix, a single click on the apply button will help you get rid of the error.

Here’s a short gif to demonstrate how it works today. If you cannot see formula repair, make sure you check your settings and toggle on “Formula repair”.

We are supporting a few error types now in preview including missing parentheses, inverted delimiters, syntax errors, incompatible types, etc. See following screenshots for examples. You can learn more about formula repair in our documentation.

We strongly recommend you try it out and let us know your feedback, including but not limited to how you like the experience, what errors you’d like to see be fixed by it.

The post Formula repair now in preview appeared first on Microsoft Power Platform Blog.

With Power Fx, our low code programming language, it makes it possible for hundreds of millions of people with the Excel-like skills to add advanced logic to their apps. However sometimes it’s not easy to write a formula, even for the most experienced Power Fx users, as it may sometimes take a lot of time searching for, learning about and debugging complex formulas.

With Power Apps Ideas, it’s easier for everyone to write formulas by leveraging the power of AI. We’re now supporting more formula patterns covering more controls and properties. We’ve improved our UI to prompt ideas contextually and also displaying ideas in a new, bigger panel. The old ideas pane (next to advanced tab) will still stay there for some time, and we will retire it over time.

Transforming natural language to formulas

After this release, you can write more formulas with natural language, examples include sorting, filtering, searching a table, conditionally format your fields in a gallery, generate text for a label or text input, generate content for a dropdown.

Transforming examples to formulas

With example to formula, you don’t need to worry about writing your own RegEx pattern or what will be the right Power Fx function to use, the AI model behind the scenes will detect the pattern from the desired input and generate one or a few formulas that can do the job for you to pick from.

We want to hear from you

With the AI assisted features, it will be easier than ever to write your formulas. You can learn more about those features via our documentation. And I encourage you to give it a try and let us know your thoughts by posting on our community forum. We look forward to your feedbacks and will keep enhancing ideas to cover more and more scenarios with high quality suggestions.

The post Announcing general availability of Power Apps Ideas appeared first on Microsoft Power Platform Blog.

Transforming natural language to formulas

1. Expanded support for more data sources – Data source connections are no longer limited to Dataverse; you can now use Power Apps Ideas with galleries and data tables connected to other data sources, such as SharePoint and Excel.
2. Supporting more ways to filter a data table – More data querying scenarios are supported including filter by Choice columns, Yes/No columns. Now you can try active ‘Contacts’ in Ideas to filter by Contact status.
3. Improved understanding of natural language – In our efforts to provide an experience with language that really is “natural”, we made significant improvements to our AI model to make it smarter and understand more natural input. For example, instead of typing ‘Contacts’ whose ‘Full Name’ starts with “Yvonne”, you can simply type Full Name starts with Yvonne to get the same formula.
4. Enhanced UI for formula suggestions – The formula suggestions are now displayed in cards, and there is a greater focus on the natural language description of what the formula is doing, so it will be easier for makers to review and understand the results.

Programming by example

1. Increased region and language availability – Programming by example has rolled out to all the Power Apps regions and languages so more makers can benefit from it.
2. Improved formula results – We made significant improvements in our AI model to make the resulting formula suggestions less complex while still providing the same functionality. Simplifying complicated formulas not only makes it easier to understand and maintain the formula, but also significantly enhances performance.

What’s next for Ideas

Transforming natural language to formulas:

• Continue to add support of more scenarios including generating formulas to manipulate content of a label, changing color and visibility based on conditions, support filter by lookup columns.
• Continue to enhance the model so it can better understand natural language queries and provide more accurate suggestions.
• Improve the usability and discoverability of Ideas so it will become more handy and easier to find.

Programming by example:

• Improve the model quality so it can generate accurate yet simple formulas.
• Improve the usability of Train with examples, so when the suggested formula doesn’t meet your needs, you can easily tune the results by providing additional examples.

The post Power Apps Ideas: New enhancements and increased capabilities appeared first on Microsoft Power Platform Blog.

We know that manipulating text in a table can sometimes be tedious and heavy as you need to spend time working on which formula to pick, how to write them correctly, and spend hours if not days testing and tuning to get your desired output. Now, with Power Apps Ideas – Programming by example, you can just tell Power Apps what is the desired output and Power Apps will work out the formulas to get you there.

Following gif provides you a quick example. Say I want to change my date to only show month (represented by words). I can simply select my date field, and enter a desired output which is “May”, then I can get a formula in seconds. By applying the formula suggestion I see the full date changed to month only. Furthermore, if I just want to show the first 3 letters of the month, all I need to do is to feed one more example to Power Apps Ideas by clicking “Train with examples”. Once done, I can get a new formula immediately and apply to check the results which perfectly meets my expectation.

This feature is powered by PROSE (Programming by Examples and Natural Language), researched and developed by the Microsoft PROSE team. And the same technology is already used in other Microsoft products as well, e.g. Flash fill in Excel, Intellicode suggestions in Visual Studio and Table extraction in Power Query, etc. Now it is also available in Power Apps, currently for US region only. Make sure you have turned on the “Enable Ideas panel” flag if you cannot see it in your apps in a US environment.

Before you get too excited, there’re also some limitations. It currently supports label controls within a gallery and works best for dates and strings. You can refer to this document for more details.

As a next step, we plan to expand support to more scenarios and more data types. We look forward to any feedback you have on this feature. Try it out and let us know your feedback and ideas in the Power Apps community forum. 

The post Generating Power Fx formulas from examples appeared first on Microsoft Power Platform Blog.

Overview

We know to make an app powerful, most time you have to deal with Power Fx formulas so you could bind objects together, e.g. saving forms, search, updating UI based on button clicks or text typed in a text control, etc. And it doesn’t require you to become a pro-developer to start writing a Power Fx formulas. However, certain amount of time is still requested when you’re trying to write a formula, e.g. finding which function to use and what is the right syntax, tuning and refining.

In the example, which is to enable user to use the search box to search customers by name and sort results, experienced makers can easily write the following formula.

That means they need to at least understand

1. The Search() function, what can it do and the syntax of it
2. The Sort() function, what can it do and the syntax of it.
3. How to use Text input value in Power Fx
4. Knows the logical name for Full Name in Dataverse.

Today, with Power Apps Ideas, one can just type in their thoughts in plain English “search Contacts with ‘Full Name’ in TextSearchBox1 and sort results by ‘Created On’ latest on top”. No need to worry about which function to use, no need to learn the syntax of the Search() and Sort() function, no need to learn how to use a Text Input control in formula, no need to go back to Dataverse table and check for logical name, and the same formula will be generated for use.

Transforming natural language to Power Fx formula using GPT-3

Trained with 175 billion parameters, GPT-3 is an advanced natural language AI model that implements deep learning to be able to both understand and produce human-like text based on a prompt in natural language. Microsoft has a strategic collaboration with OpenAI, the developers of GPT-3, to be able to apply the model in products like Power Apps.

By fine-tuning GPT-3 to understand how Power Fx formulas are constructed, we can leverage the model’s existing strengths in natural language input to give Power Apps makers the ability to describe logic just like they would to a friend or co-worker, and end up with the right formula for their app.

Now let’s take a deeper dive and explore how the above example works. After you finish typing and click Get ideas, your input will be sent to the model. Based on your input and context from your app, like which tables you have, controls on the screen, the fine-tuned model can generate a list of most relevant Power Fx formulas that may address your business needs. PowerApps will also do a post-check to filter out results that cannot pass formula syntax check. In addition, to de-mystify the formula and explain what it means, we also generate a natural language description for each formula suggestion to make it easier for both no-code and pro-dev makers to pick from.

Get started with Power Apps Ideas

If you have an environment created in the “United States” region, you will find a new Ideas tab on the right pane after you select on a gallery or data table. Click on that and follow instructions to type in your thoughts.  For more details about how to use this feature, please check this document.

The feature is only available in US region and “en-us” language due to the limitation of GPT-3. We will investigate how to expand it to more regions and languages and keep you posted.

Limitations

To begin with, the preview capability is optimized when you are working with Search(), Sort(), SortByColumns(), Filter(), FirstN(), LastN() formulas on Gallery and Data table controls. It also includes some basic functions for conditions like StartsWith(), EndsWith(), DateDiff(), DateAdd(), etc. The simpler and clearer your requests, the better the generated outputs will be. Take a look at the default queries on Ideas pane to see examples of what is supported now. Follow best practices and use intellisense to complete your typing. For example, if you have a table called “Accounts” and you want to find records created before last week. You can type “find ‘Accounts’ whose ‘Created On’ is before last week”.

For preview, this feature will initially be available in the United States region and support English only. Check this document for full list of the limitations.

• Data sources: only supports Dataverse now, we will soon expand to support other data sources
• Data type: Text, Whole Number, Date and Time, Date Only, Decimal Number.
• Relationships: Single table only. We are also working on supporting lookup fields.

Next steps

We understand this is a very limited scope and we do plan to expand the capabilities to support more functions, controls, data types, regions and languages. We look forward to any feedback you have on this feature.

Try it out and let us know your feedback and ideas in the Power Apps community forum. 

The post Writing Power Fx formulas with natural language appeared first on Microsoft Power Platform Blog.

Both new capabilities can be found under the Prediction model card in AI Builder. One of them is prediction of multiple outcomes, and the other is prediction of a numerical field. These new features make the prediction model richer, and be able to solve more business problems for you. Previously, AI Builder prediction model could predict binary outcomes, such as whether a shipment would arrive on time or not. Now you can use it to predict more than 3 outcomes, like will a shipment arrive early, on-time or late. You can also predict a number, like how many days it would take to close a customer ticket.

More information: Overview of the prediction model – AI Builder | Microsoft Docs

With the release of these capabilities, they’re now considered premium features. You will need an AI Builder add-on license or an active trial to use them.

Multiple outcome prediction

To predict multiple outcomes, you just need to create a prediction model, and when picking the target column, the Choice columns are available in the dropdown list.  Select one column to see the available prediction outcomes. You could select all the outcomes, or just the most relevant  to your business problem. You could also map them to two outcomes if you want to do a binary prediction by clicking the Manage Outcome button in the top navigation bar.

Numerical prediction

To predict a number, go ahead and create a prediction model. At the select outcome step, select a numerical field as your target column. We support all types of numerical columns in Dataverse, including whole number, decimal number, floating point number, and currency.

Use your prediction model in Power Apps and Power Automate

To use the model, follow the steps in AI Builder to finish training, and if the result looks good, publish the model. Once it’s published, you can use your model in Power Apps or Power Automate.

More information: Use your prediction model – AI Builder | Microsoft Docs

For additional  AI Builder resources, visit Power Apps and learn how your users can make their apps AI enabled, or go to Power Automate and learn how your users make their workflow solutions better with AI. Please visit our pricing pages to learn more about the AI Builder capacity add-on for your apps and workflows. Visit our Licensing page for more information on licenses and trials.

The post New AI Builder capabilities are now generally available appeared first on Microsoft Power Platform Blog.

AI Builder prediction models now support a preview capability to predict multiple outcomes. Now you can use AI builder intelligence to predict things like is shipping going to be early, on time or late. Or will the business flow be completed as a success, a failure or cancelled.

In this post, we demonstrate how multiple outcomes prediction works by building an example of an end to end scenario for how AI Builder can help to optimize delivery performance and improve customer satisfaction for an e-commerce company.

Example

Let’s say I’m a maker in a typical e-commerce company in Brazil. My CDS database contains typical entities like Order, Product, Customer, etc. I’m noticing that there have been a lot of complaints about delivery delays. But the cost of upgrading shipping from regular to express is very high so I don’t want to upgrade every order that is likely to be late. Instead, I just want to upgrade orders that are going to be late for more than 2 days and for the rest likely to be late orders, I’ll offer some coupons to the customers.

To meet this business goal, I need to create a prediction model which can tell me which orders are going to be late and which are going to be very late (more than 2 days).

Create output fields

I’m going to create an output field to denote what constitutes a delayed delivery. In my Order entity, I have the estimated delivery date and the actual delivery date. Every record where actual delivery date is later than estimated delivery date is considered as a late delivery, and if actual delivery date is more than 2 days later than estimated, the order is considered as very late.

I will create two calculated fields:

Delivery Delta – this is the difference in days between Delivered Date and Estimated Delivery Date.

Delivery timeliness – this field categorizes the Delivery Delta column into Early, On time, Late and Very Late.

And I’m going to user Delivery timeliness as my target outcome.

Create prediction model

First, let’s create a new prediction model from AI Builder section of Power Apps. More information about how to do this: Creating a prediction model.

We use my BC_orders entity and the Delivery timeliness field. Note, multiple outcomes prediction is still in preview. You can also click Manage outcomes on the top navigation bar and select group to two outcomes to switch to binary prediction if needed.

Next, I select my related entities, Customer and Product.

Also, make sure that Delivery Date and Delivery Delta are not selected, because the model should not have information that is only known after the product has been delivered. If you do keep it selected, then the model will show that it has 100% accuracy, but it will not be able to properly predict new orders. This is known in machine learning as target leak, and you can learn more about it here.

Here, let’s skip adding a filter as this data should be sufficient to train the model.

Once all that is done, it is time to train the model. For this model, we have a performance grade of B, which means it’s correct most of the time. The most influential data is also reasonable. I know I can keep improving my model to reach an A grade. However, I decide to use this B model for now to unblock my business.

Publish and use your model

As part of publishing, we actually create three outputs:

• Predicted – The predicted outcome
• Probability – The probability of the true outcome, from 0 – 1
• Explanation – A JSON text field that contains the record level influencers

After model is published, select Use model, and then select Run now to get the prediction results immediately. Let’s give it a few minutes to let the model run on our data set.  The model will predict for any records with empty Delivery timeliness field.

After above step is done, we can use the view feature to quickly see the predicted output. Refer to this document for more info.

Now, we’re ready to connect the prediction model to our Power Automate flow. For this example, we can set the flow to be triggered when the Delivery timeliness is Late then send this customer a coupon.

Now, go ahead and try this scenario on your own!  Your feedback will  help us continue to build on and improve AI Builder. We want to hear from you!

You can refer to this document for guidance about how to import the sample data set so you can try it on your own!

The post AI Builder now supports predicting multiple outcomes (Preview) appeared first on Microsoft Power Platform Blog.

AI Builder prediction models now support a preview capability to predict a number. Now you can use AI builder intelligence to predict things like product ratings, price estimates, time to completion, and so on.

In this post, we demonstrate how numerical prediction works by building an example of an end to end scenario where we’d use AI Builder to help an online business to optimize purchase conversion rate.

Example

Let’s say our online business is quite competitive, and there’s a need to improve purchase conversion. How can we increase the percentage of customers visiting our website who then make a purchase?

One approach is to identify customers visiting a product page that have a high probability to exit the page before completing a purchase. Using AI builder prediction modeling, we can identify this customer as ‘likely to drop’, and then trigger a Power Automate flow to send a coupon to this customer. This targeted interaction enabled by AI Builder prediction modeling can help convert site visitors to customers.

Identify target field

For this scenario, we’ll use the  Online Shopper Intention entity. This entity contains historical online shopper behavioral data from the past year. There are two fields which are particularly related to the issue we want to predict.

ExitRates: Probability that a user would leave the current webpage.

BounceRates: Probability that a user would navigate away from the current website after viewing only one page

Let’s use ExitRates, as users who viewed more pages before they exit are probably better candidates to convert to customers.

Create prediction model

First, let’s create a new prediction model from AI Builder section of Power Apps. More information about how to do this: Creating a prediction model.

We use my Online Shopper Intention entity and the ExitRates field. Note, numerical prediction is still in preview, so numerical fields are annotated with a ‘Glimmer’ in the Field drop down menu.

Next, we’ll exclude BounceRates from the training fields as it might have a high correlation with exit rates.

Here, let’s skip adding a filter as this data should be sufficient to train the model.

Once all that is done, it is time to train the model. For this model, we have a linear performance score of 83. Linear performance scores measure closeness between predicted data and actual data. It can be between 0 – 100%, with higher values indicating the predicted value is closer to the real value.  Generally, a higher score means the model should perform better. However, be wary of perfect scores, as this can indicate an overfit model.

Publish and use your model

As part of publishing, we actually create three outputs:

• Predicted – The predicted outcome
• Probability – The probability of the true outcome, from 0 – 1. But this field would be empty as it doesn’t apply when you’re predicting a number.
• Explanation – A JSON text field that contains the record level influencers

After model is published, select Use model, and then select Run now to get the prediction results immediately. Let’s give it a few minutes to let the model run on our data set.  The model will predict for any records with empty ExitRates field.

After above step is done, we can use the view feature to quickly see the predicted output. Refer to this document for more info.

Now, we’re ready to connect the prediction model to our Power Automate flow. For this example, we can set the flow to be triggered when the ExitRates is higher than 15%, and send an email to this customer with coupon information.

Now, go ahead and try this scenario on your own!  Your feedback will  help us continue to build on and improve AI Builder. We want to hear from you!

You can refer to this document for guidance about how to import the sample data set so you can try it on your own!

The post AI Builder now supports numerical prediction (Preview) appeared first on Microsoft Power Platform Blog.

Why this change

We’ve received customer feedback that the accuracy score alone wasn’t sufficient to help make publishing decisions.  Some people were confused by the performance score.

Let’s look at an example model that predicts business loan approvals.  Here are 2 contrasting scenarios that have the same accuracy score (71%), but vastly different practical performance:

The model has a 71% accuracy rate, but scenario 1 represents a significant improvement, while scenario 2 is only marginally better than a random guess. Therefore, the accuracy score alone isn’t helpful, but the accuracy grade (A, B, C, D) based on  improvement relative to a random guess can help you decide if your model is ready to use.

Existing models

For any existing trained or model, you won’t see the new performance information until you retrain it.

Performance grade

After you train your model, the performance grade appears on the model details page. Accuracy may vary based on the data you use, but the grade is more consistent for a given model.

Let’s say your model predicts whether a business loan is approved or denied. If your model grade is B, it means the model performance is generally good.  You have to decide whether it is good enough based on your own unique circumstances. Does it have room to improve? Yes – probably. You can follow instructions in the AI Builder documentation to keep improving your model if you want to achieve better accuracy.

More information about what each grade means and how they’re calculated: Prediction model performance.

Training details performance tab

In addition to training summary, we’ve introduced a new Performance tab in the training details pane that appears when you click see details below the performance grade. The following performance metrics are shown:

• Accuracy grade
• Accuracy score

More information about these metrics: Prediction model performance.

The post AI Builder – New prediction model performance page appeared first on Microsoft Power Platform Blog.