D365 Supply Chain: Inventory Optimization Most Mid-Market Manufacturers Skip

Planning parameters set at go-live become planning parameters forever

The most common inventory problem in a mid-market Dynamics 365 Supply Chain deployment is not a missing feature. It is that item coverage was configured once during migration, with the same coverage group applied to every released product, and nobody has revisited it since. Two years later the plant carries stock it does not need and still expedites the same twenty parts every month.

Item coverage is set per item and, critically, per site and warehouse dimension. Requirement creates one supply order per demand line, Period batches demand into buckets of N days, Min/Max refills to a maximum when on hand falls below a minimum, and Manual removes the item from planning altogether. Applying Requirement to everything because it is the default produces a purchase order per sales line: technically correct, operationally unusable, and the reason planners quietly go back to a spreadsheet.

The time fences underneath are where the money actually leaks:

• Coverage time fence. How far into the future planning looks. Set it shorter than your longest purchase lead time and long-lead parts are never planned in time, however good the forecast is.
• Positive days. How far forward planning will look at existing supply and on-hand to cover a demand line. At zero, planning ignores the container landing next Tuesday and orders again. Left at a large migrated default on a fast-moving consumable, planning consumes stock that is genuinely needed three months out.
• Negative days. How long planning waits for a late supply order before proposing a replacement. Too small and every delayed shipment spawns a duplicate. Too large and real shortages never surface.
• Freeze time fence. The near horizon inside which planning stops proposing changes, so buyers are not re-cut nightly on orders already confirmed with suppliers.

These four values, set per coverage group rather than per item, do more for inventory than any forecasting exercise. Segment coverage groups by lead time and volatility rather than by product family and the planned order noise drops immediately.

The safety stock journal nobody runs

In most deployments, minimum quantities are numbers a planner typed during implementation, based on the old system, based on the system before that. Demand has moved. The numbers have not.

Supply Chain ships a safety stock calculation that reads historical issue transactions over a period you choose, applies a service-level multiplier to observed variability, and proposes new minimum coverage quantities per item, site and warehouse, which you review in a journal and apply selectively. It handles seasonality by calculating from a comparable historical window rather than a rolling average, which matters if your Q4 looks nothing like your Q2. Running it quarterly takes an afternoon, and most mid-market manufacturers have never run it once.

ABC classification is four fields, not a project

Released products in Supply Chain carry ABC value, ABC margin, ABC revenue and ABC carrying cost, and a periodic job assigns them from percentage thresholds you define. In most mid-market systems those fields are empty.

Empty classification means every item gets identical treatment: the same coverage rules, the same counting cadence, the same planner attention. That is expensive in both directions. A-class parts, where a stockout stops a line, deserve tight coverage time fences and frequent counting. C-class fasteners deserve Min/Max, a generous period, and to be left alone. Without classification populated, planners default to treating everything like an A part, and the system generates work no one has time to action.

Action and delay messages are the output, not the noise

Running planning and then only reading planned orders wastes most of what the engine produces. Action messages tell you to advance, postpone, increase or decrease orders you already have. Delay messages tell you which demand is at risk and why. Both are the difference between planning that rebalances existing supply and planning that only buys more. If you are on Planning Optimization rather than the built-in engine, run the fit analysis first: it reports exactly which parts of your configuration the service does not support.

Van stock is inventory, and it is usually wrong

This is the gap that sits on the boundary between Supply Chain and Customer Engagement, and the one we see most often.

Dynamics 365 Field Service models truck stock as warehouses in Dataverse, one per vehicle, with its own inventory transfers, adjustments and RTV records. Work order products decrement that warehouse when the work order is closed. Dual-write moves customers, products and work order headers cleanly enough, but Field Service inventory movements are usually not part of that integration at all.

The result is a plant where the ERP still shows a pump seal on hand that a technician installed six weeks ago. Planning does not reorder it, the next work order for that part fails, the technician drives to a wholesaler, and someone expenses it outside the system, so the consumption never enters demand history either. The safety stock calculation above then understates that requirement permanently, for exactly the parts that break most often.

The fix has to be deliberate: model every van as a warehouse in both systems, post Field Service inventory journals into Supply Chain on a schedule rather than in real time, and reconcile weekly rather than at year end. Making spare parts consumption visible to planning is one of the larger inventory wins available to a manufacturer with a field team, and it is a recurring theme in our manufacturing work and in the field service implementation we ran for a European manufacturer.

Counting: the annual shutdown you probably do not need

Warehouse management supports cycle counting plans scoped by location or product, threshold-based counts triggered when on-hand drops below a quantity or percentage, and spot counts. Many mid-market plants still shut down for two days in December instead.

A full count once a year gives you one accurate day and eleven months of drift. Counting A items monthly and C items annually gives you a number planning can rely on all year. The usual objection is that nobody has time to count, which is a data-capture problem rather than a policy problem: a Power App on a rugged device with barcode scanning, writing counts back on a schedule, is a short build, and that shop-floor capture layer is most of what our Power Platform work consists of.

One caveat that invalidates everything above: run inventory closing monthly. Until it runs, issue transactions sit at running average cost, so margin and inventory value are both wrong, and any carrying-cost argument you make to the CFO rests on numbers finance does not trust.

Where to start

In order, because each step makes the next cheaper: populate ABC classification and split coverage groups by lead time and volatility; set positive days, negative days and the freeze fence deliberately per group; run the safety stock calculation and apply what survives review; get field consumption posting back; move counting from annual to class-based cycles.

Those are configuration and process decisions, not development. What does need building is the layer around them: the capture app, the write-back, the exception alerting, the dashboard that makes excess and obsolete stock visible. That is the work our services are built around.