Slated for general availability in November 2026—with Message Center notices pointing specifically to November 16, 2026—the update will let makers apply a prompt column to existing table records in bulk, optionally filtered to a specific subset of rows. For Power Platform administrators and enterprise architects, that solves a tedious data-enrichment problem while introducing a very real governance question: how do you keep a sudden wave of historical AI jobs from chewing through your tenant's Copilot and AI Builder credit capacity overnight?
The Historical Data Blind Spot in Dataverse Prompt Columns
To appreciate why a bulk backfill matters, it helps to look at how Dataverse prompt columns behave today. According to Microsoft Learn documentation for Power Apps and Dataverse, a prompt column is an AI-powered data type that lets makers define a natural-language prompt tied to other columns in the same table. When the prompt runs, the AI model analyzes the referenced input fields—such as customer inquiry text, support notes, or Q1 sales summaries—and stores the generated text response persistently inside the prompt column for use in model-driven apps, workflows, and reports.
The catch lies in the execution trigger. As Microsoft Learn’s current FAQ on prompt columns states plainly, existing records are not processed by default. A prompt column only executes when a new record is created or when at least one of the referenced input columns on an existing record is modified. Even if a maker edits and improves the prompt definition itself, Dataverse does not recalculate existing rows until a subsequent update touches one of the input columns referenced by that prompt.
In practice, that left database administrators and makers with an ugly choice whenever they deployed a new classification, sentiment, or summarization column onto a mature table:
- Live with split-brain tables: Accept that only newly created or recently edited rows would contain AI-generated insights, leaving historical views and reports incomplete.
- Build synthetic update flows: Construct throwaway Power Automate flows or bulk-edit jobs whose sole purpose was to touch a referenced input column on thousands of legacy rows just to trick the prompt trigger into firing.
Neither option was clean. Touching records purely to trip an AI trigger pollutes "Modified On" timestamps and can unintentionally set off unrelated plugins, workflows, or audit logs.
What Roadmap ID 570865 and Message Center Advisories Reveal
Microsoft’s AI at Work roadmap item 570865, published on October 15, 2026, formalizes the fix under the title Microsoft Dataverse: Prompt column values in bulk for existing records (backfill). The roadmap entry lists the feature as In development for Web across Power Apps, Microsoft Dataverse, and Microsoft Power Platform governance and administration, targeting Worldwide (Standard Multi-Tenant), GCC, and GCC High cloud instances in November 2026. (As with all Microsoft roadmap projections, timelines and scope remain estimates until general availability.)
While the roadmap summary highlights the maker experience—applying a prompt to existing records in bulk, with optional row filtering so makers do not have to touch records one at a time—archived Microsoft 365 Message Center posts MC1488488 and MC1486292 fill in critical operational details for administrators:
- Target GA Date: Message Center posts MC1488488 and MC1486292 specify a general availability target of November 16, 2026, following earlier summer roadmap iterations (tracked in July 2026 under MC1442979) that had originally aimed for late August.
- Disabled by Default for Safety: According to Message Center post MC1488488, the environment setting Allow bulk processing of existing records will be off by default. Administrators must explicitly enable this setting in the Power Platform admin center before makers can launch bulk backfill runs.
- Asynchronous Background Execution: When triggered, the backfill capability runs asynchronously as a background job—and as noted in M365 Admin coverage of MC1486292, executes during maintenance hours—so bulk AI generation remains decoupled from real-time user transactions.
| Feature Attribute | Confirmed Details | Source |
|---|---|---|
| Feature Name | Microsoft Dataverse: Prompt column values in bulk for existing records (backfill) | Microsoft AI at Work Roadmap (ID 570865) |
| Message Center Tracking | MC1488488 / MC1486292 | Microsoft 365 Message Center Archive |
| Target Release Timing | November 2026 (GA slated for November 16, 2026) | Roadmap ID 570865 / MC1488488 |
| Cloud Instances | Worldwide (Standard Multi-Tenant), GCC, GCC High | Roadmap ID 570865 |
| Default Admin State | Allow bulk processing of existing records is Off by default | MC1488488 |
| Execution Mode | Asynchronous background job; optional row filtering supported | Roadmap ID 570865 / MC1488488 / MC1486292 |
| Cost & Capacity | Consumes Copilot Credits or AI Builder credits | Microsoft Learn / MC1488488 |
Section Summary: Roadmap ID 570865 and Message Center notices MC1488488/MC1486292 confirm that Dataverse is getting a native, filterable bulk backfill for prompt columns in November 2026, gated behind an opt-in environment setting in the Power Platform admin center.
Why Optional Filtering and Credit Governance Are Inseparable
If you have ever watched an unthrottled loop in Power Automate evaporate a monthly quota of AI credits before lunch, you already understand why Microsoft made Allow bulk processing of existing records opt-in by default.
Every time a Dataverse prompt column executes against a row, it consumes capacity—specifically Copilot Credits or AI Builder credits (which Microsoft Learn notes are transitioning toward Copilot Credits as the common usage-based currency). Running a prompt on new records as they trickle in during normal business hours is relatively predictable. Turning that same prompt loose on 250,000 legacy support tickets or five years of vendor notes all at once is a completely different financial event.
That is where the optional filtering mentioned in Roadmap ID 570865 becomes an essential cost-control lever rather than just a convenience feature. Instead of backfilling an entire historical table blindly, makers can scope the bulk run to the rows they actually need—for instance, open incidents from the current fiscal year or high-value accounts—leaving dormant archives untouched.
Microsoft Learn already documents how column-level filter conditions work on standard prompt columns to curb unnecessary credit burn. When a filter condition is configured on a prompt column, Dataverse evaluates the filter before invoking the model. Furthermore, Dataverse automatically provisions two hidden companion tracking columns for every prompt column you create:
(columnName)_PromptColumnStatus(columnName)_PromptColumnDetails
Knowing the status codes recorded in (columnName)_PromptColumnStatus helps administrators and makers verify what ran, what was skipped, and what failed:
- 0 (
NotStarted): Record created; AI analysis has not begun. - 1 (
InProgress): AI analysis is currently running, with the start timestamp logged in the Details column. - 2 (
Completed): Prompt execution succeeded, with the completion timestamp logged in Details. - 2000 (
Completed- Filter Skipped): Execution completed without running the AI model because the record did not meet configured filter conditions. - 2001 (
Completed- Input Unchanged): Execution completed without running the AI model because none of the referenced input columns were updated. - 3 (
Failed): AI generation failed, with error specifics written to the Details column (commonly caused by missing read permissions on referenced input columns or exhausted Copilot/AI Builder credits).
While Microsoft has not yet published whether bulk backfill jobs will introduce additional batch-level progress dashboards or new status codes, these existing per-row status columns and the Power Automate Automation center > AI Builder activity credit logs give administrators an established paper trail for monitoring consumption and failures.
Section Summary: Because every backfilled row consumes Copilot or AI Builder credits, combining the Power Platform admin center opt-in toggle with selective row filtering is critical to preventing unexpected capacity exhaustion.
Architectural Guardrails and Current Limitations to Keep in Mind
Before planning a November data-enrichment sprint, makers and administrators should check their tables against the structural rules documented in Microsoft Learn for Dataverse prompt columns:
- Five-Column Ceiling: You can create a maximum of five prompt columns per table.
- Unsupported Input Data Types: A prompt can reference multiple input columns from the table, but formula columns, file columns, image data types, and other prompt columns cannot be used as input columns. If a maker selects any of those data types in the prompt designer, Dataverse simply ignores their values. That means you cannot chain prompt columns together where Prompt Column B backfills from the output of Prompt Column A.
- Required Primary Column Text Input Variable: Every prompt definition automatically includes a required text input variable (
PromptColumnRecordIdby default) applied as a filter on the primary column (unique identifier) of the data source. Removing this variable or unbinding it from the primary column will block the column from saving. - Unmanaged Customizations Lock: To create or modify prompt columns, the environment setting Block unmanaged customizations must be disabled. If it is enabled, makers will hit an
msdyn_AIConfigurationcomponent restriction error. - No Native Auditing: Microsoft Learn explicitly notes that prompt columns are not audited through standard Dataverse auditing, making the companion
_PromptColumnStatusand_PromptColumnDetailscolumns your primary diagnostic record.
How to Prepare Your Environment Before Rollout
Whether you manage a commercial Worldwide tenant or a GCC / GCC High government environment, preparing for Roadmap ID 570865 takes only a few deliberate steps:
- Audit Your Credit Capacity First: Sign in to Power Automate, navigate to Automation center > AI Builder activity, and review your current Estimated consumption trends (or export the
.csvlog). Verify that your environment has enough Copilot Credits or AI Builder capacity allocated before anyone enables bulk processing. - Test Prompts for Free in the Designer: In Power Apps, open your target table, edit the prompt column, and use the Test button inside the prompt designer. Testing against sample records (or pasting a specific record ID into
PromptColumnRecordId) does not consume credits and displays an estimated credit cost for a single run in the lower-right corner. Multiply that single-run estimate by your filtered row count to forecast the cost of a historical backfill. - Verify Prerequisites and Governance Toggles: Confirm that Copilot and AI Prompts are enabled under environment feature settings, and keep Allow bulk processing of existing records turned off in the Power Platform admin center until you have agreed on which tables and row filters are approved for historical backfill.
By pairing asynchronous background execution with an admin-controlled opt-in switch and row-level filtering, Microsoft is giving Dataverse teams a practical way to bring legacy records up to parity with new data—without forcing anyone to run another midnight dummy-update flow ever again.
For more discussions on Power Platform administration, Dataverse architecture, and managing Microsoft 365 Copilot credit capacity, explore our enterprise IT and database management threads right here on WindowsForum.com.
References
- Microsoft Dataverse: Prompt column values in bulk for existing records (backfill) Microsoft 365 Roadmap · 2026-10-15T22:47:30.226767Z
- Prompt columns in Microsoft Dataverse - Power Apps | Microsoft Learn learn.microsoft.com
- MC1488488 - Microsoft Dataverse - Bulk generate prompt column values for existing records (backfill) | Microsoft 365 Message Center Archive mc.merill.net