Hiring guide
Cash forecast accuracy: explain the variance without rewriting history
By CFO Index · Published
To assess a cash forecast, compare what was known when it was issued with what actually happened. Do not judge yesterday's prediction against a spreadsheet already updated with today's bank balance. A fractional CFO should explain both the size of the difference and what the team can learn from it.
Freeze the forecast being tested
Keep the issued file or snapshot, preparation timestamp, bank-account scope, currency and forecast horizon. Record whether you are testing the next week's prediction or a week forecast a month earlier; those are different tests. Reconcile actual opening and closing cash to the same accounts and cutoff. Identify transfers inside the reporting perimeter so they do not appear as new external receipts or payments. A missing bank feed is a data limitation, not automatically a forecasting error.
Build a bridge that reconciles to the closing balance
Use actual minus forecast as the sign convention for cash balances and receipts. For payments, a larger outflow has a negative effect on closing cash. Hypothetical week: opening cash was €50,000 in both versions. Forecast receipts were €40,000 and payments €55,000, giving €35,000 closing cash. Actual receipts were €32,000 and payments €51,000, giving €31,000. The bridge below explains the €4,000 shortfall. All figures and explanations are invented for illustration.
| Difference | Effect on closing cash | Evidence or follow-up |
|---|---|---|
| Opening cash difference | €0 | Same reconciled opening balance |
| Customer receipt moved to next week | −€10,000 | Confirm revised date and retain invoice ID |
| Unexpected customer receipt | +€2,000 | Match receipt to the customer record |
| Supplier payment moved to next week | +€6,000 | Keep the obligation in the next forecast |
| Unplanned repair payment | −€2,000 | Trace the approved payment and its cause |
| Total difference | −€4,000 | €35,000 forecast becomes €31,000 actual |
Separate timing from a changed amount or missing item
Use a short cause list: timing, changed amount, omitted item, data or model error, and unresolved. Split a transaction when more than one cause applies. A customer paying €8,000 of a €10,000 expected receipt creates both a received amount and an outstanding balance to investigate; it is not automatically a €2,000 permanent loss. Record evidence and an owner for uncertain explanations. British Business Bank's forecast guidance puts receipts in the period cash is expected to reach the bank. The review method here applies that cash-timing basis to a preserved forecast.
Keep delayed items visible until they resolve
In the example, the €10,000 receipt and €6,000 supplier payment remain future cash events unless new evidence changes them. Moving both to next week changes next week's net movement by €4,000, assuming neither was already included there. Use stable invoice or transaction IDs to prevent duplication. Do not repeatedly describe an old delay as a new surprise: track its original forecast date, each revision and actual settlement. Delaying a supplier payment improved this week's balance but did not eliminate the obligation.
Do not let offsetting errors pass as precision
The receipt difference is minus €8,000 and the payment difference improves cash by €4,000. Their net is minus €4,000, but the two category-level absolute differences sum to €12,000. Neither figure tells the whole story. Review major receipts and payments separately, and distinguish signed bias from absolute error. A percentage based on a near-zero net cash movement can be misleading; state the denominator or use currency amounts. Do not invent an acceptable accuracy benchmark without considering the decisions and horizon involved.
Turn one recurring cause into a process change
If customer dates are repeatedly optimistic, test whether the collection owner can provide better evidence before the forecast cutoff. If approved purchases are omitted, change the input hand-off rather than adjusting every future week by an unexplained contingency. Assign an action and a date to check whether it helped. Preserve the original forecast even when a genuine model defect is corrected; retain a separately labelled corrected version if needed to understand the defect.
Define a small, verifiable CFO assignment
Ask the provider to reconcile one completed week, explain material differences, list unresolved items and show that delayed transactions appear once in the new forecast. Then compare several forecasts made at the same lead time to see whether the same bias recurs. The handover should include the signed bridge, evidence references and a repeatable review routine. This evaluates the forecasting process; it does not promise that customer behaviour or future cash balances can be predicted exactly.