Dynamics 365 Supply Chain Management Adds AI Forecast Explanations and SMAPE Accuracy Metrics

The update pairs calculated demand forecasts and historical data with symmetric mean absolute percentage error (SMAPE) metrics and AI-generated natural-language explanations of the factors influencing each forecast. Rather than leaving supply chain teams to stare at bare projections, the feature is designed to show the historical baseline, quantify the error rate, and summarize what drove the calculation in a single web view.

Rollout Schedule, Scope, and Current Status​

According to Microsoft's AI at Work roadmap and the corresponding 2026 release wave 1 plan on Microsoft Learn, roadmap item 576754 targets Dynamics 365 Supply Chain Management and Microsoft Copilot (Microsoft 365) on the Web platform for Worldwide (Standard Multi-Tenant) cloud instances. Note that while the roadmap metadata tags Microsoft Copilot alongside Supply Chain Management, Microsoft Learn's release plan describes an in-product demand-planning capability rather than a separate integration inside the standalone Microsoft 365 Copilot app.

While October 2026 is the scheduled general availability window, the roadmap record still lists the item's status as In development. Furthermore, Microsoft Learn's release-plan entry notes that the 2026 wave 1 release plan pages were archived in October 2026 as Dynamics 365, Power Platform, and Dataverse disclosures transitioned to the unified AI at Work roadmap, and it explicitly cautions that delivery timelines and projected functionality remain subject to change. Because roadmap dates are estimates and the entry has not yet been closed out as generally available, enterprise administrators and planners should verify availability directly in their environments rather than assuming the rollout has already reached every tenant. When it does land, Microsoft states the feature is enabled for end users automatically without requiring manual administrator activation.

Section Summary: Roadmap item 576754 brings AI-generated forecast explanations and SMAPE accuracy metrics to Dynamics 365 Supply Chain Management on the web. Public preview was slated for July 31, 2026, with a General Availability target of October 2026, though the roadmap currently keeps the feature marked "In development."

What the Feature Actually Does—and What It Doesn't​

It is easy for marketing language around artificial intelligence to blur the line between better explanations and better predictions. A careful reading of Microsoft's documentation makes an essential distinction: this update is built for forecast transparency and interpretation, not as a new forecasting engine that automatically corrects bad inputs.

According to Microsoft's feature specification, the update delivers three core functional elements in one workspace:

• SMAPE Accuracy Values: Planners can view the symmetric mean absolute percentage error (SMAPE) for their forecasts.
• Unified Historical and Calculated Views: Historical demand data and calculated forecast results are displayed together in a single location.
• AI-Generated Driver Explanations: Natural-language summaries explain the underlying factors influencing each calculated forecast.

Microsoft positions these capabilities as a way to build planner confidence, reduce the time spent investigating why a forecast shifted, and support faster decisions. Those efficiency and confidence gains represent Microsoft's projected business value rather than independently audited performance benchmarks. Microsoft's release-plan notice does not disclose which specific language model generates the text, the exact UI navigation path for roadmap item 576754, whether planners can edit or challenge the generated explanations, or whether special licensing prerequisites apply beyond standard product access.

Technical Context: From MAPE to SMAPE and Demand Planning Version 1.4​

To appreciate why Microsoft is highlighting SMAPE alongside AI explanations, it helps to examine how Dynamics 365 Supply Chain Management has historically handled forecast accuracy and how its next-generation Demand planning experience has evolved.

In Microsoft Learn's documentation for the built-in Master Planning demand-forecasting tools (Monitor forecast accuracy), the platform tracks two primary measures: historical forecast accuracy (which compares past forecasts against actual historical demand via the Show accuracy option on the Demand forecast details page) and estimated model accuracy, which is displayed under Model details - MAPE (Mean Absolute Percentage Error). When organizations use Azure Machine Learning for built-in demand forecasting, internal model accuracy is evaluated against a holdout slice of historical data controlled by the TEST_SET_SIZE_PERCENT parameter—for instance, setting the parameter to 20 tests the model against the final 20 percent of historical transactions.

However, traditional MAPE has a well-known mathematical blind spot in supply chain management: when historical sales drop to zero during intermittent demand periods, dividing by zero—or near-zero actuals—skews standard percentage error calculations.

Microsoft Learn's What's new or changed in Demand planning documentation shows how the standalone Demand planning app has been tackling this exact problem across recent releases:

• Demand planning version 1.0.0.1232 introduced an Explainability tab in forecast calculation job-run records that listed the chosen algorithm and MAPE for each dimension combination, such as item variant and warehouse location.
• Demand planning version 1.4 introduced Best fit model - version 4, which switches to SMAPE as the metric for selecting among competing forecast models—specifically noting that SMAPE improves model selection for products that have periods with zero historical sales.
• Version 1.4 Calculation Accuracy and Explainability: That same version 1.4 release notes document a calculation accuracy page showing the selected algorithm and resulting SMAPE per planning object, complete with model-distribution visuals and a filter-responsive average SMAPE indicator. Crucially, Microsoft notes that users can drill down into individual planning objects—such as a specific product and store—to view historical data next to the calculated forecast alongside a natural-language explanation of the driving factors. Version 1.4 also adds a full-horizon component breakdown separating a forecast into baseline, signals impact, and manual adjustments.

While the version 1.4 release notes describe capabilities that closely mirror the description of roadmap item 576754, Microsoft's documentation does not explicitly tie the roadmap ID number to the version 1.4 build label. Still, the technical trajectory is unmistakable: Dynamics 365 is moving away from opaque, single-metric batch outputs and toward drill-down workspaces that pair intermittent-demand-friendly error metrics (SMAPE) with human-readable driver summaries.

This also complements—rather than replaces—earlier cell-level Copilot tools already documented in Demand planning, such as the Copilot grid cursor (introduced in version 1.0.0.2794 and enhanced in version 1.0.0.3336 to separate signal impact from baseline values for a selected cell) and predefined analytical cursor prompts (version 1.0.0.1700). Where the grid cursor inspects individual cell edits and comments, the forecast accuracy explanation view focuses on evaluating the calculated forecast series and its SMAPE indicator as a whole.

Section Summary: Built-in Supply Chain Management forecasting historically relied on MAPE and holdout test sets (TEST_SET_SIZE_PERCENT), whereas recent updates in Demand planning (notably version 1.4) adopt SMAPE to better handle products with zero-sales periods and provide object-level drill-downs with natural-language explanations.

Practical Guidance for Supply Chain Teams and IT Admins​

When automatic rollout reaches your tenant, how should planning teams put AI explanations and SMAPE values to work without falling into the trap of blind automation? Based on general industry practice and the documented boundaries of the feature, here is a practical framework for evaluating the update:

1. Use SMAPE as a Triage Filter, Not an Absolute Guarantee: Because Microsoft does not publish a universal numerical threshold for a "good" SMAPE score—acceptable error varies wildly between fast-moving consumer staples and slow-moving spare parts—planners should use high SMAPE values to prioritize which product-and-location combinations need human review first.
2. Treat AI Explanations as Investigative Leads: A natural-language summary of forecast drivers is an aid to interpretation, not proof of real-world causation. When the AI explanation flags a specific trend or driver, planners should verify that narrative directly against the adjacent historical data view, known promotional calendars, or external signal inputs before locking in purchase orders.
3. Check Environment Governance Before Relying on Timeline Estimates: Because roadmap item 576754 remains listed as In development with an October 2026 GA target, ERP administrators should monitor the Microsoft 365 AI at Work roadmap and the Explore Feature Geography and Language reports referenced on Microsoft Learn to confirm when their specific Azure region and language pack receive the update.

By putting historical data, SMAPE error scores, and plain-English driver explanations on the same screen, Microsoft is tackling one of enterprise ERP's oldest bottlenecks: the trust gap between statistical modeling and human accountability. If the rollout delivers on its design, planners won't have to spend Monday morning playing data detective just to find out why the forecast changed—they can read the brief, check the historical curve right next to it, and get back to keeping the supply chain moving.