Demand sensing: bu ne anlama geliyor
Demand sensing is often sold as forecasting at higher frequency: the same model, run daily instead of weekly. That is a faster forecast, and it answers the same question it always did.
Sensing is a different question. A forecast asks what demand will be over a horizon. Sensing asks what has changed since the forecast was made.
The inputs are not the same
A statistical forecast reads history: sales, seasonality, promotions, trend. Sensing reads what is happening now — point-of-sale movement in the last three days, order book changes, channel stock positions, a competitor going out of stock.
Some of these do not exist as history at all. A competitor stockout that started on Tuesday is not in eighteen months of sales data, and no amount of re-fitting will find it.
The decisions are not the same either
A forecast drives planning: production schedules, purchase orders, capacity. Those decisions have long lead times and want a stable number.
Sensing drives allocation and replenishment: which store gets the constrained stock, whether to expedite, whether to hold. Those decisions are made daily and want a signal that moves.
Feeding a sensing signal into a planning process usually makes it worse, because the planning process was tuned for stability and now receives noise.
Where it pays and where it does not
Sensing earns its cost when the response window is shorter than the forecast cycle and an action is genuinely available. Fast-moving retail with daily replenishment: yes. Made-to-order manufacturing with a twelve-week lead time: rarely, because by the time the signal is read the decision has already been committed.
The test is simple and often skipped. If the signal changed today, what would you do differently before it stopped mattering? If the honest answer is nothing, sensing is producing information with no decision attached.
The measurement problem
Forecast accuracy is measured against actuals. Sensing is harder, because a sensing system that works changes the outcome it would be measured against — you allocated differently, so the stockout did not happen, so the signal looks like it was wrong.
The measurement that holds up is decision-level: how often the sensing signal changed an action, and what happened when it did against comparable cases where it did not.
Why the distinction matters commercially
Organisations buy sensing to fix forecast error, and forecast error is usually not the constraint. If the forecast is reasonable and decisions still arrive late, the problem is between the forecast and the action — approval steps, batch runs, an ERP that only accepts changes overnight.
Sensing does not fix that. It makes the delay visible sooner, which is useful, and is not what was bought.