Demand forecasting process: from inputs to review

A demand forecasting process turns source exports into an approved expectation of customer demand, then checks that expectation against actual results. The essential controls are a shared demand definition, documented adjustments, and a saved forecast snapshot that later revisions cannot overwrite. Without those controls, the review measures changing files rather than forecasting performance.

Define the decision before collecting inputs

Write down what the forecast must support: supplier orders, capacity commitments, or inventory availability. Specify the item, location, customer channel, unit of measure, target period, and horizon. A monthly total is insufficient if purchasing needs a weekly delivery decision.

Choose a demand signal. Requested customer quantities, invoiced sales, and shipments answer different questions when orders are backlogged or inventory is unavailable. Record returns and cancellations separately so the team can distinguish new demand from later reversals.

Hyndman and Athanasopoulos put problem definition first: the forecast's use and users determine the task. The general demand forecasting guide covers demand concepts; the process here focuses on the files, handoffs, and checks needed to run a repeatable review.

Establish the export and approval handoffs

Use the same minimum evidence at each cycle. Agree the cutoff before anyone supplies an override.

Keep the demand expectation separate from a constrained shipping plan. If supply can fulfill less than expected demand, preserve both figures and identify the gap. Otherwise, a capacity limit silently becomes a lower demand forecast.

Build an adjustment bridge that adds up

A US industrial distributor approves its August 2026 demand forecast using the following quantities in individual units. Sales adjustments reflect customer schedule changes; event adjustments reflect separate, documented campaigns. These are illustrative business inputs, not claims about typical forecast accuracy.

Put the following columns in A:H with headers in row 1. Approved forecast is baseline plus the two accepted adjustment columns. Signed error is approved forecast minus actual demand, so a positive error is an overforecast.

Enter these Excel formulas in E2, G2, and H2, then fill through row 4:

=SUM(B2:D2)
=E2-F2
=ABS(G2)

The approval bridge must reconcile: 380 + 10 + 40 = 430 units. Make the following check return zero:

=SUM(E2:E4)-SUM(B2:D4)

Microsoft's SUM documentation confirms that it accepts ranges and ignores text in referenced cells. Validate quantity columns as numeric before relying on the totals. Covirage is not affiliated with Microsoft.

Keep the individual reasons behind the net sales adjustment. The total of 10 contains both an increase of 20 and a reduction of 10. A net-only comment would hide the opposing customer changes.

Freeze the forecast before the period begins

Save the approved rows with an issue date, target month, version, approver, and adjustment references. Preserve a stable identifier for item-location-period-version. Review duplicate identifiers before summing anything; a repeated approved row doubles demand without changing the source transactions.

The file used for performance measurement must be the version available when the relevant decision was made. If the team revises August after new August orders arrive, retain that revision as a separate version. Do not replace the earlier approval.

For statistical method comparisons, Forecasting: Principles and Practice recommends evaluation on data excluded from fitting. The same separation matters operationally: later information must not improve the historical forecast being scored.

Finance can keep the approved forecast distinct from its financial target using the budget versus forecast distinction. Neither file should quietly inherit the other's purpose.

Close actuals and review the adjustments

Actual demand must use the same units, dates, and item scope as the approved rows. Reconcile the 415 units to the closed demand export before interpreting error. Record late changes to actuals so a revised score has an explanation.

The signed errors reconcile to 430 - 415 = 15 units. Absolute errors sum to 5 + 10 + 20 = 35 units because shortages and excesses do not cancel in this measure.

WAPE = Sum of absolute forecast errors / Sum of actual demand

Approved-forecast WAPE is 35 / 415 = 8.4%, and signed bias is 15 / 415 = 3.6%. Percentages are rounded to one decimal place.

=IF(SUM(F2:F4)=0,NA(),SUM(H2:H4)/SUM(F2:F4))
=IF(SUM(F2:F4)=0,NA(),SUM(G2:G4)/SUM(F2:F4))

Test whether adjustments helped by calculating baseline absolute error in I2 and filling down:

=ABS(B2-F2)

The baseline errors are 15, 30, and 10 units, totaling 55. Baseline WAPE is 55 / 415 = 13.3%. The adjustments improve the overall result, but SKU C worsens from 10 to 20 units of absolute error. Review its event assumption instead of treating the total improvement as proof that every override was useful.

Use bias by SKU and site for persistent directional errors. Products with many empty periods need the separate checks in intermittent demand forecasting.

Where it goes wrong

• Overwriting approvals. A live workbook that changes after the cutoff removes the evidence needed to assess the original decision.
• Overlapping adjustments. Sales and marketing may enter the same customer event twice. Link both requests to the same evidence before accepting quantities.
• Comparing different demand definitions. Requested orders against shipments creates apparent forecast error whenever fulfillment moves between periods.
• Scoring only the total. Opposing item errors cancel, and successful adjustments can conceal unsuccessful ones.
• Leaving review actions ownerless. Record who will verify the assumption, what evidence is needed, and which next approval it can affect.

Review your forecast files in Covirage

Bring a sample of your own baseline, adjustment log, approved snapshot, and closed demand export to Covirage. Covirage maps the columns; its tools compute and check the reconciliations, and the external AI model explains the comparisons with every figure traced to its rows. Create a free Covirage account and set up your business with your own files.

See Covirage for supply chain teams for the supporting operational context.