Point of view · Semiconductor equipment supply chains

Forecasting the future, not the past.

The chipmaker decided a year ago. The equipment maker turned that decision into a build plan two quarters ago. The subsystem supplier's forecast is the last system in the chain to see it. This paper argues for forecasts built from the drivers that create demand, and judged by what being wrong costs.
Illustrative concept: UPSTREAM Chip demand Moves in cycles TIER 2 Fab investment Committed against roadmaps TIER 1 Equipment build plans Amplify the move YOU Subsystem orders Arrive last, and largestUPSTREAMChip demandMoves in cyclesTIER 2Fab investmentCommitted against roadmapsTIER 1Equipment build plansAmplify the moveYOUSubsystem ordersArrive last, and largest
Idea 1 · Drivers, not history

Demand is a system of drivers, amplified by the cycle.

History records the outcome of decisions made upstream, not the decisions themselves. Backlog, customer portal forecasts, design wins, qualification milestones and equipment build plans already sit inside the enterprise.

Treat them as primary inputs and the forecast sees the ramp when it is decided, not when bookings finally move.

Illustrative concept: SENSE · WHAT YOU CANNOT CONTROL SHAPE · WHAT YOU CAN Fab investment plans Equipment build plans Qualification mileston Customer portal foreca Design wins Long-lead buys Capacity Allocation Customer commitments Driver-based forecast one demand view A forecast as a rangeSENSE · WHAT YOU CANNOT CONTROLSHAPE · WHAT YOU CANFab investment plansEquipment build plansQualification milestonCustomer portal forecaDesign winsLong-lead buysCapacityAllocationCustomer commitmentsDriver-based forecastone demand viewA forecast as a range
Illustrative concept: FROM SHIPMENTS TO ESTIMATED TRUE DEMAND 100 Shipped volume -12 Double-booking removed -5 Expedite timing reset +15 Allocation shortfall added back 98 Estimated true demand IllustrativeFROM SHIPMENTS TO ESTIMATED TRUE DEMAND100Shipped volume-12Double-bookingremoved-5Expedite timingreset+15Allocationshortfall addedback98Estimated truedemandIllustrative
Idea 2 · True demand

Shipment history is shaped by supply, not only by customers.

In a ramp, customers double-book to secure allocation and expedites distort timing. In a downturn, qualifications slip and options are canceled. A forecast trained on shipments learns those effects, not customer intent.

Mark the periods when supply shaped what shipped, and the forecast can separate what customers wanted from what the network could deliver.

Idea 3 · The cost of being wrong

Publish the forecast that minimizes the cost of error, not its size.

Over-forecasting a ramp part ties up working capital; under-forecasting it costs expedites and revenue inside the window that matters. For a sunset part, the costs run the other way.

When finance and supply chain own those costs together, the published number becomes an economic choice rather than a statistical habit.

Illustrative concept: COST OF ERROR Under-forecast a ramp expedites, lost revenue Over-forecast a sunset cash tied up in stock Too low ← forecast error → too high Tuned to the cost of error Tuned to accuracy alone TWO ERRORS, TWO DIFFERENT PRICESCOST OF ERRORUnder-forecast a rampexpedites, lost revenueOver-forecast a sunsetcash tied up in stockToo low ← forecast error → too highTuned to the cost of errorTuned to accuracy aloneTWO ERRORS, TWO DIFFERENT PRICES
Inside the paper

Seven failure modes, and the architecture that answers them.

◇

Why history-trained forecasts miss the turn

Ramps, inflections and step changes from design-in cycles arrive before the history shows them.
≈

Revenue and units come apart

Price and content per system move through the cycle, so dividing revenue by an average price misleads.
▦

Each share is its own forecast

How demand splits across parts, customers and regions is forecast from the drivers that move it.
∑

Many forecasts, blended on merit

Candidate forecasts are weighed by how they perform and what their errors cost, part by part.
◷

Forecasts that declare their decision

Every forecast layer states the decision it supports and the horizon it is judged at.
⇢

From a revenue number to a part and a site

A defensible path from the executive revenue view to the units procurement must commit.
Who it is for

Written for suppliers to the equipment makers.

◎

VP supply chain

Long-lead commitments sized to the ramp that is coming, not the one that passed.
∑

CFO

Working capital and revenue risk priced into the forecast you publish.
▤

Demand planning leaders

Fewer overrides, and a forecast that explains which driver moved it.
⚙

COO

Capacity and allocation decisions made on demand intent rather than shipment history.
Email delivery

Read the full paper.

Get “Forecasting the Future, Not the Past.” by email. Use your company address; we’ll send a verification code, then a private link to the PDF.

We use these details to deliver the paper and respond to your request. This does not subscribe you to marketing emails. Privacy notice.