Dataverse Prompt Columns: classify row data with AI

Key insights

• A Prompt Column can use another column from the same Dataverse row as prompt input.
• The example classifies comments as POSITIVE or NEGATIVE and saves the category in the record.
• Adding a specific instruction to return only the category made the test output concise.
• The demonstrated prompt runs when a new record is created; the result may take a few seconds to appear.
• The test prompt used about 0.2 Copilot credits; the demo does not establish licensing, rollout status, regional availability, or production costs.

Microsoft Dataverse Prompt Columns can use AI to process table data and save the result in the same row, avoiding a separate flow whose only job is to call a prompt and write its answer back. This can matter to makers who want a generated classification, such as a sentiment label, stored alongside the record that the AI analyzed. Ronan Vico Technologies demonstrates the setup with a comment table and a positive-or-negative category.

What it does

A Prompt Column is configured with instructions for an AI prompt and can draw on data from the current Dataverse record. In the demonstration, the input is a comment and the desired output is one of two labels: POSITIVE or NEGATIVE. The generated text is saved in the new category column rather than being handled only as a temporary prompt response.

The practical change is where the prompt-and-save work happens. The example does not create a Power Automate cloud flow to detect a new row, send the comment to a prompt, then update that row with the answer. Instead, the maker creates a column whose prompt is associated with the table data and lets Dataverse populate its value after a record is added.

This is a narrow replacement, not evidence that Prompt Columns replace every flow. The demonstrated task is to produce one AI-generated value from record data and place it in that record. If a process needs additional actions beyond that pattern, the material does not show how those actions would be handled.

Who it is for and what you need

The demonstrated setup is aimed at Dataverse makers configuring a table and its columns. The example requires a table with a comment field, a new column for the classification, and a prompt that uses the comment as input. It also uses sample values to test the prompt before saving.

The video shows a model choice, GPT-4.1 mini, for this simple classification. It does not establish which licenses, plans, permissions, or environment settings are required, nor does it verify whether the option is available in every tenant or region. Check the settings and model options available in your own environment before planning a deployment.

The transcript does not identify a rollout stage or preview status. It also does not explain how to monitor usage or determine credit costs for a production workload. Treat the demonstrated model and test as an example, not a guarantee of access or a cost estimate.

How it works

1. In the Dataverse table, add a column for the result, such as a comment category. Set its data type to Prompt, as shown in the demonstration.
2. Choose to add a new prompt and build it from scratch. Include the comment column as prompt input by selecting it from the available row columns.
3. Write the instruction around the specific task. The example asks the AI to categorize the comment as POSITIVE or NEGATIVE. It also uses a required instruction to return only the category, without extra wording.
4. Test the prompt with sample data. The demonstration enters “THIS IS A GOOD VIDEO!” and sees that the first version returns a full explanatory sentence. The added output rule produces the single label POSITIVE instead.
5. Save the prompt and then save the column. The demo names the prompt to describe the category it fills and notes that it can be edited later.
6. Create records with comments and allow time for the prompt to run. The example shows the category appearing after a short delay, with one positive and one negative comment classified.

The sample also exposes a row-context setting called PromptColumnRecordId. In the demonstration, the current record ID and a filter limit the input to the row being processed, so the prompt receives that row’s comment rather than comments from the whole table. The presenter mentions that a no-filter option could retrieve data across the table, but does not use it for the classification example.

Limitations and open questions

The output is generated by AI, and the example shows why prompt wording matters: without a strict response instruction, the model returns an explanation rather than just a category. Testing with representative examples and checking the saved results are sensible parts of configuring the demonstrated task; the video does not establish accuracy rates or guarantees.

The result may not appear immediately. In the live test, the presenter waits a few seconds for processing and refreshes the page before showing the categories. The material does not specify processing time, behavior when a prompt fails, or exactly which later edits to source data cause the prompt to run again.

There is no detailed comparison of Prompt Columns with other automation patterns, no production-scale test, and no licensing or rollout documentation in the supplied sources. The example supports using the feature for a simple AI classification stored on the row; it does not establish that it is the right choice for multi-step actions or every AI-backed Dataverse process.

Sources

• Ronan Vico Technologies’ Dataverse Prompt Column demonstration

Frequently asked questions

What does a Dataverse Prompt Column do?

It uses a prompt to process data from a Dataverse row and saves the generated text in the Prompt Column. The demonstration uses it to classify a comment as positive or negative.

Can a Dataverse Prompt Column replace a Power Automate flow?

It can avoid a dedicated flow for the demonstrated task of prompting on row data and writing the answer back to that row. The material does not show it replacing flows that perform additional actions.

How do I make a Prompt Column return only a category?

Add a clear instruction requiring the model to return only the allowed category labels. In the example, that changed a full explanatory response into the single label POSITIVE.

What license do Dataverse Prompt Columns require?

The supplied material does not state the required licenses or plans. Confirm requirements in the documentation and environment settings for your deployment.