Time series forecasting uses ordered business history to estimate future revenue or demand. Follow a monthly example, check the forecast against actuals and identify the decisions the results support.
Time series in forecasting uses a business measure's ordered history to estimate its future values, such as monthly revenue or weekly demand. Its usefulness depends on consistent periods, comparable definitions and a test against outcomes that were unknown when the forecast was made. The estimate supports planning; it does not explain every cause of future change.
A time series preserves the sequence of observations. January followed by February contains information that disappears if the same revenue amounts are sorted from largest to smallest.
The NIST introduction to time series analysis identifies trend, seasonal variation and dependence between observations as structures that time series methods may need to account for. For a business reader, the question is whether a pattern remains useful beyond the historical periods where it appeared.
A trend describes a sustained direction. Seasonality describes recurring timing, such as a holiday-related demand peak. An unusual event may instead be temporary, or mark a lasting change in the business. A chart can suggest these possibilities; it cannot settle them without context.
Keep the expected outcome separate from the approved target. The distinction in budget versus forecast matters when management wants a revenue number that exceeds the historical pattern.
Use a revenue ledger or demand workbook with period, business unit, product or region, measure, amount, currency and close status. Add the forecast date, target period and forecast version when estimates are stored beside actuals.
Choose one measure before preparing the history. Recognized revenue, booked orders, shipped units and requested demand are different series. Do not change between them halfway through the file.
The Census Bureau's time series guidance emphasizes regular intervals and comparable measurement. It also explains that trading-day and moving-holiday effects can affect period comparisons. A strong sales month may partly reflect the calendar rather than stronger daily demand.
Check that each period is complete. Investigate a missing month instead of silently treating it as zero. Label partial months and exclude them from a comparison with full months unless an explicit adjustment makes them comparable.
A US distributor wants a simple reference forecast for its monthly revenue review. The following fictional ledger contains full, closed months in 2026. The example is deliberately short to make every calculation visible; it is not enough history to establish an annual seasonal pattern.
| Month | Actual revenue (USD) |
|---|---|
| January | 90,000 |
| February | 120,000 |
| March | 90,000 |
| April | 120,000 |
| May | 150,000 |
| June | 180,000 |
For this illustration, the forecast for each next month is the arithmetic mean of the preceding three closed months:
Next-month reference forecast = Sum of revenue in the preceding three months / 3
At the March close, the April forecast is (90,000 + 120,000 + 90,000) / 3 = $100,000. At the April close, the May forecast becomes (120,000 + 90,000 + 120,000) / 3 = $110,000. At the May close, the June forecast is (90,000 + 120,000 + 150,000) / 3 = $120,000.
Each forecast uses only information available before its target month. These are successive forecasts made one month ahead, not a forecast of all remaining months made at the March close. That difference matters when the business plans staffing several months in advance.
At the June close, the same rule produces a July forecast of (120,000 + 150,000 + 180,000) / 3 = $150,000. It is an estimate for review, not an actual ledger entry. For detailed spreadsheet functions, use the forecast function guide.
Compare forecasts with their corresponding actual months. Define error here as actual minus forecast, so a positive error means the business underestimated revenue.
| Target month | Forecast revenue (USD) | Actual revenue (USD) | Error (USD) | Absolute error (USD) |
|---|---|---|---|---|
| April | 100,000 | 120,000 | 20,000 | 20,000 |
| May | 110,000 | 150,000 | 40,000 | 40,000 |
| June | 120,000 | 180,000 | 60,000 | 60,000 |
| Total | 330,000 | 450,000 | 120,000 | 120,000 |
The identity 450,000 - 330,000 = $120,000 proves the signed error total. Mean absolute error is 120,000 / 3 = $40,000 per month. Absolute error divided by total actual revenue is 120,000 / 450,000 = 26.7%, rounded to one decimal place.
The errors are positive and increasing. The trailing average is lagging the recent rise. That is useful evidence against accepting July's $150,000 estimate without review, but it does not prove the rise will continue.
The dedicated guide to evaluating time series forecasts covers the wider evaluation job. Keep the forecast horizon, error sign and denominator consistent when comparing results across departments.
Finance should ask whether the forecast supports the revenue outlook and cash planning assumptions. Operations should ask what demand would require from staffing, purchasing or capacity. The appropriate tolerance depends on the decision: a shortfall that is manageable for one team can disrupt another team's commitments.
Finance owns the revenue definition and ledger tie. Operations or sales owns the relevant business assumptions. The reviewer owns the forecast snapshot, horizon and error report. Agree on these responsibilities before debating which statistical approach to use.
If a pricing change, customer launch or capacity constraint is known, record it as an explicit assumption rather than expecting history alone to anticipate it. Time series regression addresses relationships with other measured variables; it is a separate question from this ordered-history example.
Bring a sample of your own monthly history, supplied forecasts and actuals to Covirage; it maps the columns, and its tools compute and check period totals and forecast differences. The external AI model explains the results with every figure traced to its rows. Create a free Covirage account, then set up your business and bring your file.
See forecast analysis for the supporting review context.
There is no universal minimum that makes a business forecast reliable. You need enough comparable history to examine the patterns relevant to your planning horizon and reserve later observations for testing. Repeated seasonal cycles are more informative than a single seasonal peak.
A forecast based only on past revenue can identify a recurring pattern or an unexpected departure. It does not establish the cause. Use price, volume, customer and operating evidence to investigate the change before assigning a business explanation.
A forecast estimates an expected outcome using information available at a stated date. A budget sets the approved financial plan. A historical forecasting rule can inform both, but management should keep expected performance separate from the target it wants the business to achieve.