Separate pharmaceutical demand from channel stocking

Shipment growth can conceal unchanged demand

The monthly sales review celebrates a 30 percent increase in primary shipments. Procurement raises its requirements, production adds a campaign and finance expects the growth to continue. Meanwhile, distributor sales remain flat. The channel has bought more stock, but the organisation has not yet established that underlying demand has increased.

Pharma demand forecasting starts with the meaning of the observation. A model trained on shipments predicts shipments unless the team deliberately adjusts the data and defines a different target. Changing the algorithm cannot repair a confused definition of demand.

For Indian manufacturers and distributors, the practical first step is to put sales levels, inventory movements and reporting coverage beside one another. This creates a basis for deciding whether the next replenishment should grow, remain stable or fall while excess channel inventory is consumed.

Agree the sales level and reporting boundary

Primary sales commonly describe manufacturer shipments to the trade, while secondary sales describe the next commercial movement in the channel. Organisations use these labels differently, so document the exact transaction and party boundary. Secondary sales are not automatically patient consumption. Further stocking, transfers or institutional ordering can still influence them.

Use consistent units and product identities across sources. A manufacturer may report cartons while a distributor reports saleable packs. A pack-size conversion can create apparent growth without any change in the underlying quantity. Record conversion rules with effective dates and treat strength or formulation changes as separate product decisions.

Also record the population covered by each feed. Sales from 70 reporting distributors cannot be compared directly with shipments to 100 distributors and called a discrepancy. Compare like-for-like coverage first. If an estimate is required for missing partners, label it and preserve the method so later actuals can replace it without rewriting the original forecast silently.

WHO's storage and distribution guidance discusses stock control and reconciliation for medical products. That provides an operational context for trustworthy inventory data; the forecasting approach below is an original management design, not a WHO forecasting requirement. WHO storage and distribution guidance.

Worked example of a channel inventory build

Consider an illustrative product measured in packs. At the start of the month, reporting distributors hold 4,000 packs. They receive 13,000 packs from the manufacturer and sell 10,000 packs downstream. Assume no returns, transfers, losses or timing differences in this simplified example.

Closing inventory is 4,000 + 13,000 − 10,000 = 7,000 packs. Primary shipments increased from a previous 10,000 to 13,000, a 30 percent rise. Secondary sales stayed at 10,000. The extra 3,000 packs accumulated in the channel.

If the next month's secondary demand is again expected to be 10,000 packs and the agreed closing inventory target remains 4,000, the illustrative replenishment requirement is 10,000 + 4,000 − 7,000 = 7,000 packs. Forecasting next month's shipments at 13,000 would build still more stock under those assumptions.

This calculation is a planning bridge, not an instruction to reduce every order. The team must inspect product availability, location, expiry, commercial commitments and whether the inventory is actually eligible for sale. Seven thousand packs in one geography do not necessarily serve demand in another. A blocked lot cannot support the same replenishment decision as usable stock.

The calculation should also expose its assumptions. If a tender shipment was received at month-end but not yet recorded downstream, the next period may contain a timing reversal. If a distributor is preparing for a documented expansion, the target inventory itself may need an approved change. The forecast record should distinguish those explanations.

Reconstruct demand signals without inventing sales

A stockout can suppress observed sales. Treating a zero-sales week as zero demand may cause repeated under-ordering. However, filling the missing demand with a confident estimate can be equally misleading when the product has intermittent orders.

Record stockout intervals, available quantities and known unfulfilled orders. Keep observed sales and estimated lost demand as separate fields. Where the evidence is weak, use a range or scenario and let the planning decision show its sensitivity. Do not overwrite the transaction history with an imputed number.

Promotions and commercial pushes need similar treatment. Identify the start and end dates, affected products and partners, expected mechanism and whether the activity changes consumption or merely purchase timing. A shipment spike followed by a trough can be one stocking event rather than two unrelated changes in demand.

Returns require their own classification. A credit note may appear weeks after the physical return. Counting both as separate demand reductions distorts the forecast. Link the records where possible and explain the chosen timing basis. The distribution software guide addresses the transaction controls that make this reconciliation dependable.

Evaluate the forecast at the horizon that matters

A forecast that predicts next week well may be unsuitable for material procurement three months ahead. Define the decision horizon before comparing models. A long supplier lead time, a campaign production cycle and a short replenishment route require different evaluation windows.

Start with transparent baselines, such as a recent average or a seasonal comparison where there is enough relevant history. Test more complex methods against the same historical cut-off dates and the same product population. Keep future information out of earlier training windows. Product launches and discontinuations need explicit treatment because the past may describe a different commercial situation.

Report both error and direction. Repeated over-forecasting can build inventory even when absolute error looks acceptable. For an illustrative two-period case, forecasts of 110 and 130 against actual sales of 100 and 100 produce absolute errors totalling 40 and a positive signed error of 40. Dividing absolute error by total actuals gives 20 percent weighted absolute percentage error for this small example. The positive signed error reveals persistent overstatement.

Avoid percentage metrics where the denominator is zero or tiny without explaining how those cases are handled. Review service and inventory consequences alongside model accuracy. A numerically better forecast can still produce a worse operating plan if allocation, batch sizes or replenishment rules are poorly designed.

Give overrides an owner and a review date

Commercial teams often know about events the statistical baseline cannot see. Capture an override as a separate layer: original forecast, adjusted quantity, affected period, rationale, evidence, owner and review date. “Management target” and “expected demand” should remain distinguishable.

After actuals arrive, evaluate whether the override helped. Repeated optimistic adjustments without supporting evidence should lead to a planning discussion, not an automatic reduction in every salesperson's input. The objective is to learn which information adds value and at what horizon.

Bring material changes into sales and operations planning, where demand, capacity, inventory and finance can be reconciled. If supply remains constrained, use the approved shortage allocation process rather than letting forecast rank become an undeclared customer priority rule.

Start with one product family and a reconciled feed

Choose a family with identifiable packs, reasonably consistent reporting and a known replenishment decision. Establish a weekly data-quality review before judging model performance. Track missing partners, late submissions, unit mismatches and unexplained inventory differences separately from forecast error.

The pilot should retain the forecast as it was issued, not only its latest revision. Compare successive versions and identify which information changed the decision. This makes the planning process explainable when a later review asks why production was increased.

NextEdge Labs can help connect the transaction feeds and build the reconciliation and override workflow. The first useful output is a demand view that planners can explain pack by pack and period by period, with uncertainty visible enough to guide action.

Summary and next actions

A pharmaceutical forecast becomes more useful when it separates underlying demand, channel inventory and replenishment timing. Establish those meanings before investing in a more complex model, and evaluate the forecast against the decision it is intended to support.

• Document the exact transaction boundary and unit behind each primary and secondary sales feed.
• Reconcile opening inventory, receipts, sales and adjustments for one reporting population.
• Compare a simple baseline with the current forecast at the actual replenishment horizon.
• Add reasons, owners and review dates to overrides, then check their contribution after actuals arrive.