Financial forecasting models: choosing business drivers

Financial forecasting models estimate company results by connecting expected activity with prices, costs and other assumptions. Driver-based forecasts explain operational causes, trend forecasts provide a historical benchmark, and scenarios show what changes under alternative conditions. Choose the approach that supports the decision and survives comparison with actual results.

Start with the decision and the available rows

A purchasing decision needs units and unit costs; a hiring decision needs workload, capacity and start dates. A revenue total alone cannot answer either question. Keep the difference between an approved target and a current expectation clear: budget vs forecast explains which benchmark belongs in each review.

Use monthly ledger results together with the operational records that explain them. Useful fields include business unit, product or service, period, quantity, net realized price, variable unit cost, fixed expense, forecast version and assumption owner. State whether quantities mean orders, shipments or recognized sales; those events can fall in different periods.

The SBA's business-plan guidance calls for prospective income statements, balance sheets and cash flow statements, with detailed first-year projections. A profit forecast is therefore one part of a company outlook. Collection terms, inventory and capital spending still need separate cash and balance-sheet assumptions.

Compare three approaches

These approaches can work together. A driver forecast can be the central estimate, a trend can challenge it, and scenarios can test decisions around it. Avoid choosing whichever approach produces the preferred profit.

For the detailed revenue method, use sales projection; forecasting revenue and cost connects those inputs across a projected income statement. Here the question is which assumptions and tests should govern the company forecast, including its cost consequences.

Work the same business both ways

Cedar Ridge Distribution is an illustrative US distributor selling one standardized product group. Its monthly history is below. Realized price is $100 per unit, variable cost is $60 per unit, and fixed operating expense is $25,000 per month. This simplified example has no other profit lines.

Revenue equals units times realized price. Operating profit equals revenue minus variable cost minus fixed expense. Each row reconciles using those relationships.

For April, extending the observed monthly increase of 100 units gives 1,300 units. Keeping the historical price and costs gives a trend-based operating profit of $27,000. The operating team instead supplies 1,250 units and a documented supplier increase to $62 per unit.

The driver forecast profit checks as 125,000 - 77,500 - 25,000 = 22,500. Actual profit checks as 124,000 - 76,880 - 25,000 = 22,120. This is a short teaching history, not evidence that a straight-line trend is suitable for this distributor year-round.

Measure errors on genuinely future periods

Keep each forecast as issued before the month started. Hyndman and Athanasopoulos explain that forecast accuracy must be evaluated on observations excluded from fitting, rather than on how closely a method reproduces its own history.

Suppose the company records the following revenue estimates, each issued before its respective month. April comes from the worked table; May and June are additional evaluation observations. Error here means forecast minus actual, so positive error means overforecasting.

Mean absolute error = Sum of absolute forecast errors / Number of observations

Mean signed error = Sum of forecast-minus-actual errors / Number of observations

Driver mean absolute error is (1,000 + 2,000 + 1,000) / 3 = $1,333.33. Trend mean absolute error is (6,000 + 8,000 + 22,000) / 3 = $12,000. Driver mean signed error is zero; trend mean signed error is $12,000. Dollar averages are rounded to cents; other figures are exact.

Zero signed error does not mean zero mistakes: positive and negative errors cancel. Compare both measures, and keep the evaluation horizon consistent. These observations favor the driver approach for this example; they do not establish a permanent winner across every business unit.

Use scenarios to set contingent actions

Keep April's price and fixed expense unchanged, then vary only the assumptions explicitly shown below.

The downside profit is 110,000 - 71,500 - 25,000 = 13,500. It combines lower volume with a higher unit cost; it is not a pure volume sensitivity. Document whether that combination could occur before using it to delay purchases or staffing.

Scenario forecasting explores plausible combinations and supports contingency planning. These labels are not probability bands. Assign an action and owner to each case rather than presenting a range without a decision.

Where it goes wrong

• Later knowledge enters an earlier forecast. Revised inputs make historical accuracy look better. Preserve the issued version and cutoff date.
• History crosses a business change. Acquisitions, discontinued products or accounting reclassifications can break a trend. Reconcile the scope before extending it.
• Driver detail lacks evidence. A precise volume assumption is still uncertain if nobody can support it. Record its source, owner and review date.
• Revenue looks accurate while costs drift. Evaluate profit and individual cost drivers alongside revenue; offsetting errors can hide a weak plan.

Check your forecast files in Covirage

Bring your forecast workbook, assumptions and matching actual-results export; Covirage maps the columns, and its tools compute and check forecast differences and error measures. The external AI model explains the computed results with every figure traced to its rows. Create a free Covirage account and bring a sample of your own data.

See FP&A reporting for the reporting context.