Seasonal forecasting for retail on Dynamics 365

Retail forecasting has a shape that generic ERP forecasting handles badly: strong seasonality, short product lifecycles, promotions that move demand rather than create it, and a store-times-SKU matrix large enough that nobody can review it line by line. Dynamics 365 has three different tools that touch this problem, each aimed at a different size of retailer, and none of them is a complete merchandise planning suite. This guide is about choosing the right one and knowing where it stops.

The three tools

Demand Planning (part of Supply Chain Management, with its own app experience) is the current statistical forecasting product. It ingests historical demand at whatever grain you load — item, site, warehouse, customer group, calendar — runs a choice of models including ones that fit seasonality, and lets planners adjust the result in a worksheet before publishing it back as a demand forecast for master planning. This is the tool for mid-to-enterprise retailers on the Finance and Operations stack. The Demand Planning overview covers the mechanics.

Commerce replenishment is not a forecasting engine. It is an allocation and distribution engine: buyer's push to distribute a purchase across stores by weight, cross-docking to route inbound stock straight to stores, and replenishment rules that top stores up against minimums and maximums. It consumes a plan; it does not produce one.

Business Central's Sales and Inventory Forecast extension is a lightweight monthly forecast driven by an Azure AI model. It gives a small retailer a reasonable per-item forecast for reorder decisions with essentially no setup. It does not do store-level, weekly, or promotion-aware forecasting, and it should not be asked to.

Forecast at the grain you can actually manage

The first mistake retailers make with Demand Planning is forecasting every SKU at every store every week because the tool allows it. The result is millions of series, most with sparse, noisy history, and a planner who cannot review any of it.

The pattern that works:

• Forecast at category or sub-category by store cluster — stores grouped by format, size, or climate zone — at weekly grain. The history is dense enough for the seasonal models to find a real pattern.
• Disaggregate to store-SKU using recent sales mix, which Demand Planning does when you publish at a lower level than you forecast. Accept that the disaggregation is mechanical.
• Review exceptions, not rows. Configure the worksheet to surface series where the model's forecast diverges from last year's actuals or the planner's override by more than a chosen band.

Store clusters are also the natural unit for Commerce's buyer's push, so the forecast grain and the allocation grain line up.

Seasonality, promotions, and new items

Seasonal patterns are what the statistical models are good at, provided the history is long enough — a minimum of two full cycles, three is better. Retailers migrating from another system should load at least that much history rather than starting the clock at go-live.

Promotions are where most retail forecasts go wrong. A model that sees last Easter's promoted volume as baseline demand will over-forecast this Easter's unpromoted week. Demand Planning's approach is to let you separate the history into baseline and promotional components and handle the uplift as an adjustment, but it depends on the retailer having tagged promotional periods in the history. Commerce discount data can supply the tags; wiring that up is a project step, not a checkbox.

New items have no history. The standard technique is like-item modelling — copy the seasonal profile of a comparable item and scale it — which Demand Planning supports through profile assignment. Fashion retailers with hundreds of new SKUs a season will find the effort per item too high and typically need a merchandise planning ISV that handles attribute-based new-item forecasting.

What is not in the product

Be clear with the merchandising team before they see a demo:

• Open-to-buy — the financial budget that constrains purchasing by category and period — is not a Dynamics 365 feature. It is a spreadsheet, a Power BI model over the forecast and commitments, or a merchandise financial planning ISV.
• Assortment planning and range building are merchandising decisions the product records but does not optimise.
• Markdown optimisation is ISV or data-science territory.
• Size-curve and pack optimisation for apparel is ISV territory.

The forecast tools produce a demand number; turning it into a buy that fits the budget and the range is still human and partner work.

From forecast to orders

Once a forecast is published, Supply Chain Management's master planning consumes it to generate planned purchase and transfer orders, netted against on-hand and open orders, with the forecast-reduction rules deciding how actual sales consume the forecast. For retailers, the useful configuration is forecasting at the distribution centre and letting Commerce replenishment handle store distribution, rather than planning every store as an independent demand point. Planning at every store produces thousands of tiny transfer orders and a warehouse that hates the planning team.

On Business Central, the forecast extension feeds the item's reorder logic and the planning worksheet. It is adequate for a retailer with one or two locations and a few thousand items. Beyond that, the honest advice is that BC retailers use their POS ISV's replenishment features or an external planning tool; BC is the ledger and inventory record, not the planning brain.

Measuring whether it works

Forecast accuracy metrics are built into Demand Planning, but the retail measure that matters is stock-out rate on seasonal lines during the season and residual stock at season end. Track both by category and cluster in Power BI against the forecast that was published. A forecast that is statistically accurate but published too late to influence the buy has done nothing.