Forem: Ippu Ito

Fix "Exception from HRESULT: 0x80131904" in Power Automate: Check the Recycle Bin

Ippu Ito — Tue, 03 Mar 2026 14:00:00 +0000

I recently encountered an issue where a "Create new folder" action in a Power Automate (SharePoint connector) flow suddenly failed with the error "Exception from HRESULT: 0x80131904".

TL;DR: The cause was "too many files accumulated in the SharePoint Recycle Bin".

Even though the visible number of files was under 5,000, the total count including the contents of the Recycle Bin exceeded the threshold, causing the error.

Below are my notes on the investigation and solution.

Error Occurred: "Exception from HRESULT: 0x80131904"

The flow execution result looked like this:

Exception from HRESULT: 0x80131904
clientRequestId: xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx
serviceRequestId: xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx

"0x80131904" is a common error code, and there are several reported cases of the same error string occurring in SharePoint Online.

Background: The SharePoint "5,000 Item Limit" Problem

SharePoint has a "List View Threshold," often referred to with the 5,000 items guideline.

If you hit this threshold, errors can occur not only in the screen display but also in "operations via API" (which is exactly what happened this time).

The Cause: Recycle Bin Overload

Initially, when I checked the number of items in the document library, it was around 3,700, so I didn't see any immediate red flags.

However, in the Recycle Bin of the same site, there were over 7,000 files deleted from the same document library remaining. This was the pitfall.

In SharePoint, there are cases where "items in the Recycle Bin increase too much, causing threshold-related errors." If you only look at the "visible file count" and feel safe, you might get blocked by what is clogged up in the background.

Resolution: Delete Unnecessary Files and Empty the Recycle Bin

To fix this, I permanently deleted unnecessary files and emptied the Recycle Bin. After that, the flow executed successfully to the end without any issues.

By the way, the SharePoint Online Recycle Bin has stages: there is the "Recycle Bin" that users see, and another layer (Second-stage Recycle Bin) beyond that.

To reduce the chances of falling into this trap, it is safer to make "emptying the Recycle Bin periodically" a rule in your operations.

Potential Workarounds (Unverified)

I solved it by "emptying the Recycle Bin" this time, but there are environments where you can't easily delete files. Here are the workarounds I plan to try if I get stuck with the same issue again.

1) Split the Document Library

Create a separate document library for archiving and move old files there.

If you can suppress the total volume of the library that Power Automate accesses, you should be less likely to trigger threshold-related errors.

2) Add Indexes to Columns

It is a standard best practice that SharePoint List View Threshold issues can be "avoided with indexes."

Therefore, adding indexes to columns used for search/reference in the flow (such as columns related to creation date or folder path) is also a candidate for resolution.

Note: 0x80131904 Can Occur for Other Reasons

In my environment this time, "massive files in the Recycle Bin" was the cause, but 0x80131904 itself is a generic SQL error and can occur in SharePoint Online due to other factors (like throttling or timeouts).

So, if you encounter the same error (Exception from HRESULT: 0x80131904), you should suspect the "item count (including Recycle Bin)" first. If that doesn't solve it, you will need to investigate other causes.

How to Add CSAT Surveys to Custom Topics in Copilot Studio (and Fix Analytics)

Ippu Ito — Tue, 24 Feb 2026 14:00:00 +0000

By default, CSAT is only in the 'End of Conversation' topic.

But what if you want to ask for feedback in the middle of a custom flow?

Here is a quick workaround.

Key Takeaways

You can copy the CSATQuestion action from the system topic to any custom topic.
Warning: Just answering the CSAT survey does NOT mark the session as "Resolved".
To fix the analytics (avoid "Abandoned" status), you must explicitly set the conversationOutcome variable to Resolved in your topic code.

CSAT

CSAT stands for Customer Satisfaction rating. In Copilot Studio, it measures user satisfaction using a 5-star rating system (scores from 0 to 5).

For more details about CSAT, see:
[https://learn.microsoft.com/ja-jp/microsoft-copilot-studio/analytics-csat]

CSAT collection is built into the system topic "End of Conversation" by default, but this time I wanted to test whether it could be collected from custom topics.

Implementation

First, open the system topic "End of Conversation".

Open the code editor.

Find the "CSATQuestion" action and copy this action.

Next, open the code editor for the custom topic where you want to integrate CSAT, and paste the action you copied earlier.

When you exit the code editor, CSAT will be integrated into your custom topic.

That completes the implementation process.

Testing the Implementation

When you publish the agent to Teams and call the topic we just created, CSAT will be displayed like this. Here, I selected 1,

When you return to the agent building screen and open the [Analytics] tab, you'll see the satisfaction score contains the value you just entered.

*It takes about 30 minutes to 1 hour for the data to be reflected in the Analytics tab.

"Outcome" and "CSAT (Satisfaction Score)" are Different

I commented out the conversationOutcome setting in the topic we created earlier,

Then called the topic again and this time selected 5.

The satisfaction score becomes an average of 3, but the new outcome shows as "Abandoned (session timeout)".

From these results, we can see that responding to CSAT does not change the conversation outcome to "Resolved," and if you want the conversation outcome to be "Resolved" or "Escalated," you need to configure the conversationOutcome setting.

Stop "Abandoned" Sessions! Fix Copilot Studio Analytics with YAML & System Topics

Ippu Ito — Tue, 17 Feb 2026 14:00:00 +0000

Is your Copilot Studio analytics dashboard showing too many "Abandoned" sessions? You might be ending conversations incorrectly. In this guide, I will show you how to properly configure conversationOutcome to ensure your Resolution Rate is accurate.

Analytics Session Outcomes

Conversational agents in Copilot Studio can record "conversation outcomes (Analytics metrics)" during interactions.

The types of outcomes are as follows:

Escalated: When the "Escalation" system topic is called or the "Transfer to Agent" node is executed
Resolved: When the "End Conversation" topic is called and the user enters the resolved branch
Abandoned: When the session times out and is neither "Escalated" nor "Resolved"

This time, I investigated how to record "analytics session outcomes" (referred to as "outcomes" below).

Preliminary Setup

To ensure our prepared topics are reliably called, we'll turn off the generative AI responses.

The "End Conversation" Node Alone Doesn't Result in "Resolved" Status

For example, if you create a topic that calls the "End Conversation" action as shown below: When you end the conversation using only this topic: The result will be marked as "Abandoned" (neither resolved nor escalated). This happens because the conversation ended without explicitly specifying an outcome such as "Resolved" or "Escalated" within the topic (session), and the session subsequently timed out. Below, I'll introduce two methods to specify these outcomes.

Method 1: Redirect to the "End Conversation" Topic

The first method is to use the system topic "End Conversation". This is likely the standard approach, as officially documented:

Implementation is simple: when the conversation reaches a natural conclusion (where you want to segment your analytics), select [Topic Management] → [Go to another topic], then choose the system topic "End Conversation". Note: System topic "End Conversation" ≠ "End Conversation" node Once this system topic is called: The outcome will be recorded. Note: It may take 30 minutes to 1 hour for the results to appear in the analytics tab.

Method 2: Explicitly Specify conversationOutcome

The second method is to explicitly specify either "Resolved" or "Escalated".

To specify an outcome, first open the topic's source code: Then simply specify the conversationOutcome for the node where you want to set the analytics result (in this case, the "End Conversation" node).

kind: AdaptiveDialog
beginDialog:
  kind: OnRecognizedIntent
  id: main
  actions:
    - kind: SendActivity
      id: sendActivity_1
      activity: "Ending conversation..."
      conversationOutcome: ResolvedConfirmed  # <--- This is the key!

When this topic is called: The outcome will be marked as "Resolved". Similarly, you can create a topic that ends with an escalation: Specify "Escalated" for the conversationOutcome: When this topic is called: The "Escalated" count increases by one in the analytics. Note: The values that can be specified are as follows:

conversationOutcome: Escalated
conversationOutcome: ResolvedImplied
conversationOutcome: ResolvedConfirmed
conversationOutcome: Abandoned

As such, when using "Analytics" to improve Copilot Studio, you need to pay attention to how conversations are ended.

Official documentation
https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/measuring-outcomes
https://learn.microsoft.com/en-us/microsoft-copilot-studio/analytics-improve-agent-effectiveness?wt.mc_id=copilotstudio_inproduct_analytics#outcomes-and-engagement

Deep Dive: How System Topics Handle conversationOutcome in Code

The system topic "End Conversation" also sets conversationOutcome. If you click [Open Code Editor] to view the source: You can see that "Resolved" is configured. Incidentally, if the user is not satisfied with the answer, the system topic "Escalate" is called: Looking at the source code of this system topic: You can see that conversationOutcome is set to "Escalated".

How to Check Session Outcomes

1. From Copilot Studio

To check from Copilot Studio, go to the [Analytics] tab, click [View details], then [Download sessions] to download a CSV file.

You can view the session outcomes and conversation content.

2. From Power Apps (Dataverse)

Copilot Studio conversation history is stored in the "ConversationTranscript" table:

By referencing this table, you can view detailed data such as the reason for ending (outcomeReason), number of events (turnCount), and other granular information.

When you call the operation to set conversationOutcome consecutively

Create a topic like this,

Edit the code so that conversationOutcome is set when sending the "End" message.

If you call this topic from the user's side in rapid succession,

The number of sessions increases according to the number of times it is called.

*The reason the number is 6 is because a test run was performed once before this test.

Forget Canvas Apps? Build Full-Code React + Azure OpenAI Apps on Power Platform

Ippu Ito — Tue, 10 Feb 2026 14:00:00 +0000

Power Apps recently released a preview feature called "Code Apps" that enables building web applications, so I decided to create an actual app to try it out.

Code Apps (Preview)

Code Apps is a new application format that allows you to host full-code web applications built with React and other frameworks on the Power Platform, while integrating them with Power Platform's authentication, connectors, and management capabilities for distribution.
While the Power Apps Component Framework creates components that run within Power Apps canvas or model-driven apps, Code Apps enables building complete web applications from the ground up.

Official documentation here:
[https://learn.microsoft.com/en-us/power-apps/developer/code-apps/]

In this article, I'll walk through creating a Code Apps application and building a web app that runs on Power Apps.

Prerequisites

There are three prerequisites:

Install the latest version of Power Platform CLI
Enable Code Apps in the environment where you'll upload the app
Note down the GUID of the environment where you'll upload the app

Install the Latest Version of Power Platform CLI

First, install the latest version of Power Platform CLI (PAC) in your development environment.
*Note: Older versions don't support the [pac code] command.

Enable Code Apps in the Target Environment

Next, enable Code Apps on the Power Platform side. From the Power Platform admin center, click [Settings] for the target environment.

Navigate to [Features].

Enable [Power Apps code apps] and save.

Note Down the Environment GUID

Finally, note down the GUID of the environment where you'll upload the app.

That completes the prerequisites.

Building the App

Launch the Sample App

First, let's follow the official reference guide to launch the sample app provided by Microsoft locally. Start by cloning the sample project from GitHub. *Note: URL is current as of the time of writing.

git clone https://github.com/microsoft/PowerAppsCodeApps.git

Next, install the required packages using npm install.

Create an authentication profile for the target environment using the pac auth command. Use the environment GUID you noted down during the prerequisites.

pac auth create --environment [Environment GUID]

If you see the following message, authentication was successful.

Next, initialize using pac code init. The "Hello World" portion sets the app name.

pac code init --displayName '[App Name]'

Then, launch locally with the following command.

npm run dev | pac code run

Once it launches successfully, access localhost:3000.

The web app is running!

Building a Chat App with Azure OpenAI (AOAI)

Let's take it a step further by customizing the sample app to create a chat application. Edit App.tsx in the project and save. *The code is available here. Note: This code has not been tested.
Launch the app again, and you'll have a web application that enables conversations with Azure OpenAI. Since I've implemented SSE (Server-Sent Events) support, you can now have these types of real-time conversations that weren't possible in canvas apps.

Upload the App to Power Platform

Finally, let's upload the created app to Power Apps. Use the pac code push command to push the app to your environment.

npm run build | pac code push

The app you created will now appear in the app list!

When launched, you can interact with your custom web app directly within the Power Apps interface.

This exciting new feature, "Code Apps," is likely to become the standard approach for internal applications where users have paid licenses, rather than using Azure Web Apps.

The code is available here. Please note this is for demonstration purposes only and has not been tested:
[https://gist.github.com/ippu-i/cbff9bc767ba9b6f43380159af846dc4]

Stop Initializing Arrays (In Single Loops): A Cleaner Power Automate Hack

Ippu Ito — Tue, 03 Feb 2026 14:00:00 +0000

Just a quick note: I found that Power Automate’s “Compose” action can handle arrays, which is pretty handy.

The standard way to use an array

For example, if you want to store SharePoint Online (SPO) list items one by one into a variable, I used to do it like this with an array variable.
Note: You can also use the “Select” action, but I’ll skip that for this post.

If you output the variable contents, you’ll of course see multiple values stored like this.

You can use “Compose” instead

It turns out you can actually replace this process with the “Compose” action. If you add “Compose” inside Apply to each like this and then output it using the outputs() function,

it outputs multiple values in an array format like this.

Personally, what I like about this is that since you don’t use “Initialize variable,” you can keep everything contained within the scope.

⚠️However, you can’t put it in a nested loop

That said, you can’t put it inside a nested loop, so the scenarios where you can use it may be a bit limited.

Sharing this as a small tip that might come in handy someday.

Dataverse Functions Deep Dive: Limitations, Transactions, and Performance

Ippu Ito — Tue, 27 Jan 2026 14:00:00 +0000

Since a new feature called "Dataverse Functions" (or rather, a renamed feature) was added to Dataverse, I decided to put it to the test. Note: Formerly known as "Instant low-code plug-ins".

This is a very welcome feature that effectively means "you can now write heavy server-side logic using Power Fx instead of C#."

However, since the underlying architecture is still a hardcore "Dataverse Plugin," it inherits all the specific "limitations" and "pitfalls" associated with them.

In this post, I’ll cover everything from the basics of how to create and use them, to verifying "transaction behavior," "processing speed," and the "limits of recursion."

About Dataverse Functions

In a nutshell, this feature allows you to "create server-side logic using Power Fx."

Until now, whenever someone mentioned "Plug-ins," it was considered the "sanctuary of Pro Developers," requiring compiling C# code. This feature democratizes that process, allowing you to write logic using the Power Fx language you already use in apps.

It used to be called "Instant low-code plug-ins" and was created via the "Dataverse Accelerator," but it has been renamed to "Functions" and feels much easier to create now.

Main Use Cases for Dataverse Functions

Server-side Execution: Since it runs on the server rather than the app (client) side, it offers significant performance advantages when handling large amounts of data.
Written in Power Fx: No need for .NET knowledge. You can build backend logic with the same feel as writing formulas in Canvas Apps.
Reusable: Once created, a Function can be called from both Power Apps and Power Automate. This allows you to create "common components" that can be reused, preventing your logic from becoming scattered.

The key selling point is: "Citizen developers can now create heavy server-side processing and shared logic using the Power Fx they are accustomed to."

⚠️ Limitations of Dataverse Functions

Since these Functions run on the traditional "Dataverse Plug-in" infrastructure, they inherit strict constraints.

1. The 2-Minute Timeout

Functions (Custom APIs) are synchronous by default. The caller (e.g., the App) waits for the result, but there is a strict limit of "2 minutes (120 seconds)."

If the process exceeds 2 minutes, the server will forcibly terminate the execution (Timeout Exception).

2. Depth Limit (Call Chain) is capped at 8

There is a limit to "nesting" (Function A calls Function B, which calls C...) or chains where data updates trigger other logic.

Generally, if the Depth exceeds 8, the platform assumes an "infinite loop" and throws an error. (Depending on the environment, 16 might be allowed, but it is a golden rule to design within a depth of 8).

3. Power Fx "Dialect" Restrictions

Not all functions available in Canvas Apps can be used here.

UI-related functions like Launch() (screen transition) or Maps() are strictly off-limits. This is purely for "data calculation and logic."

Basics: How to Create and Use

Now, let's look at the basic creation and usage.

Creating a Dataverse Function

As an example, I'll create a function that "extracts an invoice number from an email body using Regular Expressions."

First, create a solution, then:

Select [New] -> [Automation] -> [Function].

Fill in the Display Name, Description, Parameters, and the Formula (Power Fx), then press [Save].

// Use the Match function to get all matches and return them as formatted JSON text
{
    MatchedItems: JSON(
        ForAll(MatchAll(InputText, "INV-\d{6}-\d{4}"), FullMatch),
        JSONFormat.IndentFour
    )
}

The "Function" is now complete. If you want to edit it later, simply select the item where the Type is "Function".

You can then edit the formula.

Using Dataverse Functions from Power Automate

To call the created function from Power Automate, use the "Perform an unbound action" step.

Select the created function in [Action Name] and set the arguments.

The invoice number is successfully extracted using regex.

Using Dataverse Functions from Power Apps (Canvas)

To call the created function from a Canvas App, you need to add the Environment table to the app. You can call the function from there.

Calling Other Functions & Recursion

It is also possible to call a created function from within another function.

You can create a new function and call an existing one via Environment, just like in Canvas Apps.

Self-referencing is possible = Recursion! (Or so I thought…)

Once saved, a function can even call itself.

So, theoretically, you could construct a recursive process to calculate a factorial like this:

However, while this function works if n is 8 or less (it's 5 in the image),

If you increase n beyond that...

An error occurs.

This hits the "Depth Limit" mentioned in the limitations section.

Since the max depth is 8 (technically 16 in some cases), building recursive processes like digging through folder structures isn't realistic.
Note: There probably aren't many use cases for this anyway.

Does Dataverse Functions Support Transactions?

Reading through various documents, it seems Dataverse Functions should behave transactionally, so I conducted an experiment.

I created a function that simply adds records to a "Child table" and a "Parent table". (I will explain later why I didn't use Patch).

Calling this normally succeeds.

Now, what if I force the function to crash with a "divide by zero" error after creating the Child record?

As expected, neither the Parent nor the Child record was created.

Based on this behavior, it seems to function correctly as a transaction.

Note: Further verification is needed to see if this applies to all Create/Update/Delete operations.

Tip 1: Comparing with Power Apps (Canvas) Behavior

Incidentally, if you execute the same logic (error halfway through) directly in a Canvas App:

Surprisingly, the Parent record creation (the process after the error) succeeds. This behavior might need its own detailed investigation.

Tip 2: Cannot Save Function if using "Patch" for Creation

Currently, if you try to use Patch to add data within a Function, you will encounter a save error. Therefore, when adding data, you must use the Collect function.

However, "Updating" records with Patch works fine.

You cannot specify the second argument of Patch because the Defaults function is not supported in Functions.

Validating Bulk Updates (Canvas vs. Functions)

Finally, let's compare "Processing Speed."

As mentioned in the article below, when performing bulk updates from a Canvas App, using Patch was generally the fastest method.

https://ippu-biz.com/development/powerplatform/powerapps/update-all-data-dataverse

When doing this from Functions, doing a bulk Patch is tricky (no UI, and UpdateIf isn't stable), so you'll likely end up looping with ForAll.

Surprisingly, even executing Patch 500 times in a loop took only 13 seconds.

With({records:Sample500},
    ForAll(records As record,
        Patch(Sample500 , record, {str:"a"});
    )
);

This is because the request from the client is sent only once, significantly reducing communication overhead.

If you need to update a large amount of data with fine-grained control via loops, it seems best to handle it on the Functions side.

Conclusion

Dataverse Functions is still in preview and has various limitations, but it proved to be a very useful feature that I expect to use more frequently in the future.

Agent Evaluation in Copilot Studio: Test Methods, Thresholds, and Regression Checks

Ippu Ito — Thu, 25 Dec 2025 13:10:04 +0000

I tried out the automated testing (evaluation feature) for agents created with Copilot Studio, which is now available.
It is a huge deal that we can now check response accuracy in bulk using a test set (CSV), a task that previously had to be done manually.

The Importance of Automated Testing in AI Agent Development to Ensure Quality in Copilot Studio

In AI agent development, quality assurance is a very important theme.

Unlike traditional rule-based development, generative AI is always accompanied by variability in answers and hallucinations. A single mistake can easily lead users to have a negative impression, thinking, "This is why AI is..."

Therefore, a rigorous testing process is crucial, but it is not realistic for humans to manually keep checking nearly infinite input patterns in natural language.

This is why the introduction of "Automated Testing," which I will introduce here, becomes an essential element in creating "trusted agents" that can be operated in production with peace of mind.

Automated Testing in Copilot Studio

Automated testing in Copilot Studio is performed from the [Evaluation] tab.

For testing, first prepare data containing the following four items:

Questions for the agent
Ideal responses (Expected responses)
Evaluation method (Test Method)
Threshold for success (depending on the evaluation method)

Select the "Test Method" from the following options according to what you want to verify:

・General quality

The most advanced evaluation method using generative AI. An LLM acts as a judge and scores comprehensively from the following four perspectives. It is ideal for testing generative AI-like responses where "there is no single correct answer."

Relevance: Does it answer the intent of the question accurately?
Groundedness: Is the answer based on the data source (free of hallucinations)?
Completeness: Is all necessary information covered without omission?
Politeness: Is the tone polite and free of inappropriate expressions?

・Compare meaning

Determines whether the "intent of the text" matches. It does not require a word-for-word match; if the underlying meaning and ideas match the expected value (ideal response), it is considered a pass, even if the expression is different. Effective when you want to emphasize semantic correctness.

・Similarity

The AI uses "cosine similarity" to calculate the closeness between the agent's response and the expected value with a score of 0 to 1. Used when you want to evaluate mechanically, including semantic closeness, not just exact word matches.

・Exact match

Checks if the response is "completely identical" to the expected value. A 100% match is required, including characters, numbers, and symbols. Used for confirming data where absolutely no variability is allowed, such as model numbers, codes, or fixed standard phrases.

・Partial match

Checks if the response contains expected "specific keywords" or "phrases." Useful for requirement checks, such as whether a guidance sentence like "Contact support for unclear points" is included, or if a mandatory URL is present.

Now, let's actually perform an automated test.
In this article, I will walk you through 1) Creating a test set, 2) Execution procedure, 3) How to judge results, and 4) How to deal with garbled text.

Prerequisites: Creating an Agent to Test

As preparation, create an agent to be tested. This time, I will create one with the setting of an "Agent working at a cafe."

First, create an agent that can converse in Japanese (or your target language),

Enter the following prompt in the instructions:

You are the AI Virtual Barista for "Blue Cloud Coffee."
In this cafe, primarily visited by cloud engineers and developers, your role is to answer questions from customers and provide a comfortable "deployment wait time."

## Character Settings and Behavior
- Tone & Manner: Intelligent and friendly, with a bit of humor. Converse by moderately mixing IT terminology and jokes for engineers.
- Politeness: Basically use polite language, but do not be too stiff; maintain a flat sense of distance like talking to a colleague engineer.
- Expertise: In addition to coffee knowledge, be particularly accurate in answering infrastructure information that engineers care about, such as Wi-Fi and power supply environments.

## Response Guidelines
1. Add a welcome message mixed with some IT terms to your greeting. (e.g., "Thank you for accessing," "Connection established," etc.)
2. When guiding the menu, do not just explain the taste but add benefits for engineers (boost effect from caffeine, relaxation effect, etc.).
3. Respond humorously to errors or unclear questions, mixing expressions like "404 Not Found (Answer not found)" or "Internal Server Error (Thinking)."
4. Do not force the creation of information that is not in the knowledge base; honestly convey, "That data is not indexed in my database."

## Prohibitions
- For topics about competitors (other cafe chains), lightly brush them off as "stories from another region."
- Do not perform code debugging or actual programming support. Decline by saying, "I leave that to Stack Overflow."

Add the following text file and description to Knowledge.

Text file below:

Store Name: Blue Cloud Coffee
Concept: A cafe where cloud engineers can relax, fusing technology and coffee.
Address: 1-2-3 Tech Park, Minato-ku, Tokyo
Business Hours: Weekdays 8:00-20:00, Weekends/Holidays 10:00-18:00
Closed: Every Tuesday
Wi-Fi SSID: BlueCloud_Guest
Wi-Fi Password: Coffee2025!
Payment Methods: Completely cashless (Credit card, Transit IC, QR payment only). Cash cannot be used.
Menu:
- Serverless Espresso: 400 yen (Rich and strong bitterness)
- API Latte: 550 yen (Plenty of milk and sweet)
- Deploy Donut: 300 yen (Excellent compatibility with coffee)
Power Supply: Available at all seats. Monitor rental is also available.

Description below:

Use this knowledge when the user is seeking basic information about the cafe "Blue Cloud Coffee."
Specifically, refer to this when answering the following questions:
- Basic information such as store address, business hours, and closed days
- Connection information such as Wi-Fi SSID and password
- Available payment methods (cashless support, etc.)
- Coffee and food menu and prices
- In-store facilities and atmosphere (power supply, for engineers, etc.)

Import the test cases you want to use for testing the agent.

The construction of the agent to be tested is now complete.

Creating a Test Set (Test Items)

Next, create the test set (test items) to be used for automated testing.

From [Create test set] in the [Evaluation] tab,

Download the test set template (CSV) file.

Here is the downloaded file. Lines up to 14 are comments, and line 15 onwards are the actual test items.

This time, I created the test items as follows. From left to right: "Question to AI," "Ideal (Expected) Answer," "Test Method," "Score to be considered a success."

Upload this, and preparation is complete.

It is convenient to save the uploaded test set with a name, as you can test it as many times as you like afterwards.

Executing Automated Tests

When you press the [Evaluation] button in the previous state,

You will be asked for a connection (connector), so connect with your account and execute.

When the test is completed, the success rate is displayed like this.

Details of Test Results

Finally, let's look at the details of the test results.

First, the question about closed days. Since I selected "Compare meaning" as the test method and set the success threshold to 70, it was judged as a success. Comparing the expected response and the agent's response, it looks quite acceptable.

Next, the question about cash payment. Since I selected "Partial match" as the test method, it resulted in NG because the ideal response (Cash cannot be used) was not included in the answer.

This is an example of poor test item creation; originally, this should have been carried out with "Compare meaning" or similar.

And the question about business hours. When I set the test method to "Similarity" and the threshold to 70, it resulted in NG.
However, since the AI answered correctly here as well, this is a test item where the "threshold" should be lowered, or the test method should be changed to "Compare meaning."

So, this is a very powerful automated testing feature.

Even when you make minor corrections to prompts or update knowledge (like SPO), you can immediately detect "if previous answers are broken" just by running this test, so I recommend introducing it at a mandatory level for actual operations.

Power Automate Trigger Condition Formulas Cheat Sheet: Ready-to-Use Examples for Business Hours & Email Filtering

Ippu Ito — Thu, 25 Dec 2025 13:08:00 +0000

Here's a collection of Power Automate trigger condition formulas that I personally use frequently or get asked about often.

Trigger Conditions

Power Automate trigger conditions are a feature that allows you to set conditions for when a trigger should execute. In the new designer, you can access this from the [Settings] tab: In the classic designer, you can specify this from the [Settings] menu: By the way, this formula uses the same syntax as when editing the "Filter array" action in "advanced mode": The major benefit of using trigger conditions is that the flow execution itself doesn't occur, which helps reduce execution counts and resource consumption. [https://learn.microsoft.com/en-us/power-automate/limits-and-config] In this article, I'll compile the trigger condition processes that I personally use most frequently.

Date and Time

Date and time condition formulas are commonly used with "Scheduled" triggers.

Day of the Week

Let's start with days of the week. These are frequently used for flows that should run only on weekdays or only on weekends (like batch processes). * Public holidays are difficult to handle with trigger conditions (while not impossible if you specify fixed values, it's quite cumbersome...), so holidays are often filtered out using conditional logic within the flow itself.

// Only run from Monday to Friday
@and(
  greaterOrEquals(dayOfWeek(addHours(utcNow(), 9)), 1),
  lessOrEquals(dayOfWeek(addHours(utcNow(), 9)), 5)
)

// Only run on Saturday and Sunday
@or(
  equals(dayOfWeek(addHours(utcNow(), 9)), 0),
  equals(dayOfWeek(addHours(utcNow(), 9)), 6)
)

// Only run on a specific day of the week (in this example, Wednesday only)
@equals(dayOfWeek(addHours(utcNow(), 9)), 3)

// Result of dayOfWeek function: Sunday=0, Monday=1, Tuesday=2, Wednesday=3, Thursday=4, Friday=5, Saturday=6

Time

Next is time. These are commonly used for flows that should run during business hours or outside of business hours.

// Within business hours (in this example, business hours are from 09:00 to 18:00)
@and(
  greaterOrEquals(addHours(utcNow(), 9, 'HH:mm'), '09:00'), // (Time must be in 24-hour format with leading zeros)
  lessOrEquals(addHours(utcNow(), 9, 'HH:mm'), '18:00')
)

// Outside of business hours
@or(
  less(addHours(utcNow(), 9, 'HH:mm'), '08:00'),
  greater(addHours(utcNow(), 9, 'HH:mm'), '18:00')
)

Outlook

Next is Outlook. These are used for flows that should run only when the subject contains specific text, or when emails are received from specific senders.

// Only trigger for emails from a specific sender (in this example, sample@example.com)
@equals(triggerOutputs()?['body/from'], 'sample@example.com')

// If there are multiple specific email addresses
@or(
  equals(triggerOutputs()?['body/from'], 'sample1@example.com'),
  equals(triggerOutputs()?['body/from'], 'sample2@example.com')
)

// Only trigger for emails from a specific domain
@endsWith(toLower(triggerOutputs()?['body/from']), '@example.com')

// If there are multiple specific domains
@or(
  endsWith(toLower(triggerOutputs()?['body/from']), '@example.com'),
  endsWith(toLower(triggerOutputs()?['body/from']), '@example.org')
)

// Only trigger if the subject contains a specific keyword (in this example, "invoice")
@contains(triggerOutputs()?['body/subject'], 'invoice')

SharePoint, OneDrive

For SharePoint and OneDrive, these are primarily used for flows that trigger only when files with specific extensions or files containing specific text in their names are uploaded to document libraries.

// Trigger only for specific file extensions (e.g., .pdf files)
@endsWith(toLower(triggerOutputs()?['body/{FilenameWithExtension}']), '.pdf')

// Trigger only when filename contains specific text (e.g., "invoice")
@contains(triggerOutputs()?['body/{FilenameWithExtension}'], 'invoice')

The following sections are for testing each formula. Please proceed if you're interested.

Testing

Date and Time

For testing date and time functionality, I set up a scheduled trigger that runs every minute.

Day of the Week

First, I configured the trigger condition to run only "Monday through Friday":
When tested on 5/29 (Thursday), the flow triggers successfully.
In contrast, when configured to run only on weekends:
Since it's Thursday, the flow doesn't trigger.

Time

Next, when configured to run only between "08:00 ~ 22:05":
The flow doesn't trigger after 22:06.
Conversely, when configured to trigger after 22:10:
The flow starts triggering from 22:11 onwards.

22:11 was skipped due to the timing of saving the flow.

Outlook

The trigger used is "When a new email arrives".
First, when configured to trigger only for a specific email address:
Even when two emails arrive simultaneously:
The flow triggers only once (only for the specified email address).
Next, when configured to trigger only for a specific domain:
Even when emails from two different domains arrive (one being Gmail):
The flow triggers only once (for the specified domain only).

SharePoint

For the SharePoint experiment, I used the "When a file is created" trigger.
First, when configured to trigger only for PDF files:
Even when uploading two files simultaneously:
The flow triggers only once (for the PDF file).
Next, when configured to trigger only when the filename contains "invoice":
Even when uploading two files:
Only the invoice file triggers the flow.

[https://ippu-biz.com/en/development/powerplatform/powerautomate/get-items/]

[https://ippu-biz.com/en/development/powerplatform/powerautomate/filter-sharepoint-list-boolean/]

[https://ippu-biz.com/en/development/powerplatform/powerautomate/excel-filter-query/]

[https://ippu-biz.com/en/development/powerplatform/powerautomate/filter-using-datetime/]

How to Use Structured JSON Output in AI Builder for Power Automate

Ippu Ito — Wed, 24 Dec 2025 13:00:00 +0000

AI Builder has added a feature similar to structured outputs, so I decided to try it out.

What is JSON Output?

Broadly speaking, JSON output (structured output) means "specifying the LLM's response into a designated JSON format data."

Until now, we used prompts like the following to specify the output, but the stability of the output wasn't quite there.

Generate the output in the following JSON format.

# output json schema
{
    "type" : "object",
    "properties" :
(omitted below)
}

Original OpenAI models and Azure OpenAI GPT models added "Structured Outputs" (formerly JSON mode), and now a corresponding feature has been equipped in AI Builder as well.

So, I decided to play around with it immediately.

*Official documentation is here:
[https://learn.microsoft.com/en-us/ai-builder/change-prompt-output]

Building JSON Output

For example, let's say we receive an email like this and want to use AI to extract "candidate dates and times."

First, select the "Create text with GPT using a prompt" action and choose [New custom prompt].

Set up a nice prompt.

Then, select the "Output" icon, click [JSON] -> [Edit].

Enter an example of the JSON you want to output, click [Apply], and then [Save custom prompt].
*You can also specify the output JSON using a schema.

Finally, go back to Power Automate and input the string from earlier into the Input field to complete it.

When you run the flow, it returns the response in the "specified JSON" like this.

The Benefits of JSON Output

The advantage of outputting in JSON is that "it is easy to handle the results generated by GPT as data."

For example, in this case, by combining the generation results from GPT with Outlook's "Create event" action,

you can easily set the generated values to "Start time" and "End time,"

and easily add a schedule using a flow combined with GPT.

So, that was the extremely convenient JSON output.

How to Use the New Dataverse “Prompt Column” to Automate Categorization with AI

Ippu Ito — Tue, 23 Dec 2025 01:52:19 +0000

A new "Prompt column" has been added to Dataverse, so I verified how it works.

Prompt column

The new Dataverse feature, "Prompt column," is a column where generative AI automatically populates values by referencing information from other columns in the table.

It is faster to demonstrate it in action than to explain it, so here is a verification of its behavior.

Preparation

Let's assume we have an "Inquiry" table like the following.

Suppose we want to add a "Category" column to this table and "automatically" categorize items into "Request," "Question," "Consultation," etc.

Previously, I would combine Power Automate and AI Builder, but this time I will achieve this using the Prompt column.

Configuration

First, select [New column], choose [Prompt] for the [Data type], and

Click [Add new prompt].
*Currently, the format is "Text" only.

Then, configure the prompt as shown below and save it.

Finally, save the column to complete the configuration.

Verification

Add the category column to the form for verification.
*Set it to read-only so it is not modified by mistake.

Next, enter an appropriate title and body in the form, and when you click [Save],

The result categorized by the AI will be populated.

If the title and body indicate a "Question," the category becomes "Question."

This is convenient!

While the Prompt column is extremely useful, it appears to operate using AI Builder credits. Therefore, exercise caution if you create it in a table where a large number of records are added, as unexpected costs may occur.

Bonus 1: Structured output (JSON mode) is also possible

Structured output is also possible, so use this if you want to strictly constrain the output.

For more on structured output:
[https://ippu-biz.com/en/development/powerplatform/ai-builder/change-prompt-output/]

Bonus 2: "PromptColumnExtension" error occurs without AI Builder license or credits

If users without an AI Builder trial license run this, or if executed in an environment with no credits, a PromptColumnExtension error will occur.

Forem: Ippu Ito

Fix "Exception from HRESULT: 0x80131904" in Power Automate: Check the Recycle Bin

Error Occurred: "Exception from HRESULT: 0x80131904"

Background: The SharePoint "5,000 Item Limit" Problem

The Cause: Recycle Bin Overload

Resolution: Delete Unnecessary Files and Empty the Recycle Bin

Potential Workarounds (Unverified)

1) Split the Document Library

2) Add Indexes to Columns

Note: 0x80131904 Can Occur for Other Reasons

How to Add CSAT Surveys to Custom Topics in Copilot Studio (and Fix Analytics)

CSAT

Implementation

Testing the Implementation

"Outcome" and "CSAT (Satisfaction Score)" are Different

Stop "Abandoned" Sessions! Fix Copilot Studio Analytics with YAML & System Topics

Analytics Session Outcomes

Preliminary Setup

The "End Conversation" Node Alone Doesn't Result in "Resolved" Status

Method 1: Redirect to the "End Conversation" Topic

Method 2: Explicitly Specify conversationOutcome

Deep Dive: How System Topics Handle conversationOutcome in Code

How to Check Session Outcomes

1. From Copilot Studio

2. From Power Apps (Dataverse)

When you call the operation to set conversationOutcome consecutively

Forget Canvas Apps? Build Full-Code React + Azure OpenAI Apps on Power Platform

Code Apps (Preview)

Prerequisites

Install the Latest Version of Power Platform CLI

Enable Code Apps in the Target Environment

Note Down the Environment GUID

Building the App

Launch the Sample App

Building a Chat App with Azure OpenAI (AOAI)

Upload the App to Power Platform

Stop Initializing Arrays (In Single Loops): A Cleaner Power Automate Hack

The standard way to use an array

You can use “Compose” instead

⚠️However, you can’t put it in a nested loop

Dataverse Functions Deep Dive: Limitations, Transactions, and Performance

About Dataverse Functions

Main Use Cases for Dataverse Functions

⚠️ Limitations of Dataverse Functions

1. The 2-Minute Timeout

2. Depth Limit (Call Chain) is capped at 8

3. Power Fx "Dialect" Restrictions

Basics: How to Create and Use

Creating a Dataverse Function

Using Dataverse Functions from Power Automate

Using Dataverse Functions from Power Apps (Canvas)

Calling Other Functions & Recursion

Self-referencing is possible = Recursion! (Or so I thought…)

Does Dataverse Functions Support Transactions?

Tip 1: Comparing with Power Apps (Canvas) Behavior

Tip 2: Cannot Save Function if using "Patch" for Creation

Validating Bulk Updates (Canvas vs. Functions)

Conclusion

Agent Evaluation in Copilot Studio: Test Methods, Thresholds, and Regression Checks

The Importance of Automated Testing in AI Agent Development to Ensure Quality in Copilot Studio

Automated Testing in Copilot Studio

Prerequisites: Creating an Agent to Test

Creating a Test Set (Test Items)

Executing Automated Tests

Details of Test Results

Power Automate Trigger Condition Formulas Cheat Sheet: Ready-to-Use Examples for Business Hours & Email Filtering

Trigger Conditions

Date and Time

Day of the Week

Time

Outlook

SharePoint, OneDrive

Testing

Date and Time

Day of the Week

Time

Outlook

SharePoint

Related Articles

How to Use Structured JSON Output in AI Builder for Power Automate