You Are Probably Planning the Same Demand Twice

Manufacturers invest heavily in demand forecasting, digital planning tools, and increasingly sophisticated supply chain systems. Many still carry excess inventory, unstable schedules, and material shortages regardless.

The forecast is rarely the problem. What happens to that forecast once real customer orders start arriving usually is.

Every unit of demand should be planned once. In practice, many planning systems don’t enforce that. They treat the forecast and the confirmed order as two separate demands, add them together, and hand production a number that was never real.

The Same Demand, Counted Twice

Forecast netting, sometimes called forecast consumption, is the discipline of reducing forecast quantities as confirmed orders replace them. It sounds like a configuration detail. It behaves like a structural safeguard.

Say a monthly forecast predicts 1, 000 units. By mid-month, customers have placed confirmed orders for 600 of them.

Without netting, the planning system sees a forecast of 1, 000 and orders of 600 and plans for both: 1, 600 units of demand, even though only 1, 000 will ever be sold. Production increases to match a number that doesn’t exist. Inventory builds. Working capital sits on a shelf waiting for orders that were already counted once.

With netting, the 600 confirmed units consume the forecast rather than sitting alongside it. Only 400 units of forecast remain open. Total planned demand stays at 1, 000. The number that was actually true all along.

Nothing about the forecast changed. What changed was whether the planning system was allowed to believe the same demand twice.

This Isn’t an ERP Setting. It’s a Planning Discipline

It’s tempting to treat forecast netting as a technical parameter, something to configure once during ERP implementation and leave alone. In food and beverage manufacturing specifically, that assumption is expensive.

A cleaner demand signal doesn’t just tidy up a report. It’s the input every downstream planning decision depends on. The master production schedule is built from it. Material Requirements Planning explodes it into raw material need. Procurement buys against it. If the number entering that chain is inflated, every decision made from it inherits the error — extra WIP that has nowhere to go, ingredients ordered against demand that was never real, shelf-life exposure on stock built to satisfy a forecast that had already been fulfilled. On the floor, that surplus shows up as changeovers that didn’t need to happen and lines running batches nobody was waiting for.

None of that shows up as a forecasting failure. It shows up as overproduction, schedule instability, and a planning team constantly firefighting a problem that started upstream of them, and eventually, as the orders that don’t ship on time because the plan was never really built around what customers had actually confirmed.

Customers Rarely Order Exactly When You Expect

Confirmed orders don’t arrive neatly on the date the forecast predicted, which is why most ERP systems support more than one consumption method.

Backward consumption lets orders consume forecast from earlier periods, useful when customers tend to order later than anticipated. Forward consumption lets orders consume forecast from future periods, for customers who order ahead of schedule. Most food and beverage manufacturers end up running both, with consumption windows wide enough to absorb normal variation in ordering behaviour without letting genuinely new demand slip through uncounted.

Get the windows wrong in either direction and netting stops protecting the plan. Set them too narrow and orders stop consuming forecast at all, recreating the double-count. Set them too wide and genuine incremental demand gets silently absorbed, understating what the business actually needs to produce.

Netting Matters Differently Depending on How You Build

How much this matters depends on where a product sits on the make-to-stock to make-to-order spectrum.

For forecast-driven, make-to-stock production (a large share of retail-facing food and beverage volume), netting is close to essential. Without it, duplicate demand turns directly into finished goods inventory sitting in a warehouse. 

In assemble-to-order environments, where forecasts drive component and WIP inventory but final assembly waits for the order, netting keeps component planning honest even though the finished product itself is order-triggered. In make-to-order production, where nothing starts until the order lands, netting plays a smaller day-to-day role but still shapes longer-range capacity and procurement planning.

Few manufacturers run one of these models exclusively. Most run all three at once, often on the same lines. Forecast netting has to hold up across that mix, not just in the cleanest case.

Where It Quietly Breaks Down

The failure mode is rarely dramatic. It’s usually a handful of small, ordinary decisions that compound: forecasts and orders counted simultaneously somewhere in the chain, consumption windows set once and never revisited as ordering patterns shift, master data that’s slightly out of date, or sales, demand planning and production working from different versions of the same number. Individually, each looks minor. Together, they produce exactly the symptoms manufacturers usually blame on forecast inaccuracy: inventory imbalances, schedule instability, and planners intervening manually to catch what the system should have caught first.

The fix isn’t a more accurate forecast. It’s making sure the forecast and the order are never allowed to be true at the same time.

Where AI Extends This, Rather Than Replaces It

Fixed consumption windows are a reasonable approximation of customer behaviour that doesn’t actually stay fixed. Ordering patterns shift with promotions, seasonality, and customer consolidation, and static rules drift out of alignment with them.

This is where AI and advanced planning add genuine value, not by replacing forecast netting, but by making the consumption logic itself adaptive. Instead of a manufacturer manually reviewing and resetting windows on a schedule, the system can adjust consumption behaviour continuously, based on how customers are actually ordering right now rather than how they ordered when the parameters were last set.

The future of this isn’t a better forecast. It’s a demand signal that stays accurate as the business around it keeps changing.

Conclusion

Forecast netting will never get the attention AI forecasting or digital transformation initiatives get. It’s not the most visible part of a planning system.

It may still be the most consequential. A forecast and a confirmed order are supposed to be two views of the same demand, not two separate demands added together. Get that wrong, and every system downstream (MRP, procurement, the production schedule itself) inherits a number that was never true.

For manufacturers chasing tighter inventory, steadier schedules and better service levels, the opportunity isn’t always a new forecasting model. Sometimes it’s making sure the forecast you already have is only ever counted once.