Dataverse Prompt Columns store reusable AI insights

Key insights

• A Prompt Column stores its AI-generated result as a field on a Dataverse record.
• Prompts run when relevant input data changes, such as when a record is added or meeting notes are updated.
• Makers can write the prompt in natural language, test it, choose a model, and filter which records it applies to.
• The demonstration uses stored meeting highlights, lowlights, and recommended actions as context for a Copilot Studio agent.
• Backfilling existing records and additional input and output types are described as upcoming, not currently available.

Dataverse Prompt Columns generate AI insights from business data and save the results on records, so apps, workflows, analytics, and agents can reuse them. The feature is for makers building solutions on Dataverse and teams that want consistent summaries, classifications, or recommendations available beyond a single chat. Microsoft demonstrates the configuration and an agent scenario in a YouTube video.

What's new

Prompt Columns make an AI-generated result a field in a Dataverse table rather than an answer that exists only during an individual conversation. A maker writes an instruction in natural language, such as asking the system to identify risks in meeting notes or recommend follow-up actions. The generated insight is then stored with the business data.

That persistence changes how teams can use the output. An app can display it, a workflow can act on it, analytics can include it, and an agent can draw on it when answering a question. The same prepared information can therefore be available across more than one experience, rather than requiring each experience to interpret all the source text independently.

In the demonstrated customer-meeting example, three columns capture positive highlights, potential lowlights, and recommended next actions. The actions are prioritized by urgency. The agent uses the table as a knowledge source to answer questions about risky accounts and what an account team should do next.

Who it is for and what you need

Prompt Columns are a Dataverse capability aimed at makers configuring business data and the teams using solutions built on that data. The example uses a model-driven app and a Dataverse table named CustomerMeetingHistories, with customer meeting notes as the prompt input. The demonstration also connects a Copilot Studio agent to that table.

Microsoft's published material says Prompt Columns are generally available in Dataverse. The supplied material does not specify a required licence, plan, minimum environment configuration, regional availability, or tenant rollout details. Check the official documentation for requirements that apply to your environment before planning a deployment.

The example makes clear that the source data must be in a Dataverse table and that the maker configures a prompt and its inputs. The prompt can be tested and a model selected. Makers can also apply filters to target records—for example, the demonstration filters by customer name—rather than applying the prompt indiscriminately.

How it works

Prompt Columns execute when relevant input data changes. In the example, adding a meeting record or updating its notes triggers analysis; if the input has not changed, the prompt does not run again. This allows the stored insights to be ready for later use without regenerating them for every agent question.

1. Open the Dataverse table that contains the business records and create a new column.
2. Give the column a display name and choose Prompt as its data type.
3. Write the desired analysis in natural language, then test the prompt and choose a model.
4. Configure any applicable filters to determine which records should be processed.
5. Use the generated column in an app or make it available to other components, such as a flow or an agent.

In the demonstration, a new meeting record is entered through a model-driven app. Saving it produces the highlights, lowlights, and recommended actions in their respective columns, marked as AI-generated. The Copilot Studio agent then uses those stored fields to respond to account-health questions.

Stored insights can also support workflow automation. For example, the demonstration describes using a Power Automate flow to email an account team about a critical lowlight or create a task from a recommended next step. Those are example uses, not a claim that a specific flow is created automatically by the column.

Limitations and open questions

The feature prepares information for reuse; it does not remove the need to review the quality of the input or decide how AI-generated results should influence a business process. The material does not provide accuracy guarantees or a detailed governance procedure. Teams should test prompts against their own data and consider review requirements before using outputs for consequential decisions.

There is an important distinction between current behavior and the roadmap preview discussed in the demonstration. Backfilling historical records is described as coming soon, with a preview of an option to apply the prompt to existing and new records. The video does not establish that this capability is generally available.

Likewise, additional input and output options—including multiple lines of text, choice sets, rich text, files, and images—are described as coming soon. The material does not give delivery dates for these additions. It also does not specify licensing, usage limits, or the exact monitoring and administration controls available, so those details should be checked in Microsoft's documentation.

Sources

• Microsoft Power Platform blog: Prompt Columns
• Microsoft Learn: Create and use a Prompt Column
• Microsoft demonstration of Prompt Columns in Dataverse

Frequently asked questions

What does a Dataverse Prompt Column do?

It applies a natural-language prompt to business data and stores the generated insight as a field on a Dataverse record. Other apps, workflows, analytics, and agents can then reuse that value.

When does a Prompt Column run?

It runs when relevant input data changes, such as when a record is added or its meeting notes are updated. The demonstration says it does not run again if the input has not changed.

Are Dataverse Prompt Columns generally available?

The supplied Microsoft material says Prompt Columns are generally available in Dataverse. It does not specify licence requirements, regional availability, or tenant rollout details.

Can Prompt Columns process existing Dataverse records?

Backfilling historical records is described as coming soon in the demonstration, which previews an apply-to-existing-and-new-records option. The material does not establish that backfilling is currently generally available.