The Demand Signal Your Plan Is Actually Using

Most manufacturers measure forecast accuracy and never examine the demand signal underneath it. The signal is what the planning engine actually plans against after adjustments, overrides and uncleansed history. When it misrepresents real demand, every safety stock and reorder point downstream is solving the wrong problem.

Every supply chain organization measures forecast accuracy. Very few examine what the forecast is made of.

That distinction matters more than the accuracy percentage, because a forecast is a number the business produces and a demand signal is what the system actually plans against once that number has been adjusted, aggregated, overridden and combined with history nobody has cleaned in years.

The two are rarely the same thing. The gap between them is where a great deal of inventory and service performance is decided.

What is the difference between a forecast and a demand signal?

The forecast is the output of a process. Someone runs a statistical model, sales adds judgement, finance reconciles it to the plan, and a number lands in the system.

The demand signal is what replenishment logic consumes. It is the forecast plus everything that happens to it on the way down: the consumption mode, the aggregation level, the history the model was trained on, the promotional volume that was or was not loaded, and the manual adjustments made after the number was agreed.

An organization can have respectable forecast accuracy at an aggregate level and a badly distorted demand signal at the item and location level where replenishment decisions are actually made. Aggregate accuracy hides it, because errors in opposite directions cancel each other out in the total and do not cancel out in the warehouse.

Where does the signal get distorted?

Several places, and most of them are invisible in a forecast accuracy report.

• ‍History nobody cleansed. A stockout in the base period reads as low demand rather than unmet demand, so the model learns to plan for the shortage it caused. A one-time bulk order reads as a trend. Years of this produces a statistical baseline that describes the business's past supply problems rather than its customers' behavior.‍
• Promotional and event volume that arrives late or not at all. If promotional lift is not in the signal, the system plans for a normal week and the business expedites its way through the promotion. That expediting then enters history as demand, and the distortion compounds.‍
• Model settings frozen at implementation. Forecast models are assigned once, usually on the demand patterns that existed at go-live. Products mature, channels shift, seasonality changes. The model assignment does not.‍
• Aggregation at the wrong level. A forecast produced at a level above where replenishment happens has to be disaggregated, and the split rules doing that are often older than anyone can account for.‍
• Adjustments with no audit trail. A number is overridden between agreement and execution. The override may be correct. If nobody can see it happened, the accuracy measurement is being taken against a different number than the one the system used.

Why does this cost more than a few points of accuracy?

Because everything downstream is solving for the demand pattern the signal describes.

Safety stock is calculated against demand variability. Reorder points assume a consumption rate. Lot sizes are set against expected volume. If the signal misrepresents how demand actually behaves, every one of those parameters is correctly calculated against the wrong input, and the result looks exactly like the problem covered in our piece on excess and shortage coexisting: too much stock in the wrong places, too little where customers are ordering.

This is why inventory correction work that ignores the demand signal produces improvements that decay. The parameters get fixed against a distorted signal, the signal keeps distorting, and the position rebuilds.

What does fixing the signal produce?

Huntsman Performance Products had planning fragmented across geographies and functions with disconnected tools and ineffective use of MRP-generated orders. The work included creating global Supply Chain Planning Hubs for demand and supply balancing and integrating end-to-end planning inside SAP. Overdue demand and supply elements fell 60% and exception messages fell 46%.

Delicato Family Wines, facing limited visibility to inbound inventory and sales order commitments, recorded a 96% reduction in overdue demand elements, a 95% reduction in overdue supply elements and a 90% reduction in potential service level disruptions.

In both cases the demand side was addressed as part of the same work as the supply side, which is the point. Planning them separately is how the two stop matching.

What executives should ask

Not "what is our forecast accuracy." That number is reported, it is usually acceptable, and it does not answer the question.

Ask instead:

• At what level do we forecast, and at what level do we replenish? If those are different, what rule bridges them and when was it last reviewed?
• Has our demand history ever been cleansed of stockouts and one-time events?
• When was the statistical model assignment last revisited for products whose demand pattern has changed?
• Does promotional and event volume reach the planning system in time to be planned, or does it reach the warehouse as an expedite?
• Can we see every adjustment made between the agreed forecast and the number the system planned against?

If the honest answer to several of those is that nobody has looked recently, the demand signal is almost certainly costing more than the accuracy percentage suggests.

The 12-question self-assessment will give you an initial read on where forecasting and data reliability are breaking down in your environment.

Find out what your demand signal is really telling the system. Request an executive conversation.