Time series analysis compares business measurements in date order to separate direction, recurring patterns and unusual movements. Work through a revenue example, the source checks and the decisions each pattern can support.
Time series analysis compares business measurements in date order to identify direction, recurring patterns and unusual movements. Start with comparable dated rows, then separate a change in the business from a change in the calendar or reporting definition. The result should tell a manager what moved, where to investigate and which historical comparison deserves attention.
A time series follows the same measure through successive periods. Monthly net revenue, weekly completed deliveries and daily support requests all qualify when the definitions remain consistent. Dates are part of the evidence: rearranging observations would change the story.
NIST's introduction to time series analysis emphasizes that observations over time can have structure, including dependence between nearby observations, trends and seasonal variation. Treating every month as an unrelated observation loses that context.
The unit of observation matters. A company's monthly revenue series combines all customers in each month. A customer-level series follows one customer through those same months. Use a consistent customer identifier to move between the two, and keep customers with no activity visible where a missing purchase is meaningful.
For the distinction between following a measure over time and comparing entities at one date, see time series versus cross-sectional analysis. A reporting pack may need both views, but their denominators answer different questions.
The level is the amount at a particular date or period. The trend is the broader direction through time. Seasonality is movement associated with a recurring calendar pattern. An irregular movement is what the chosen comparison does not explain; it still needs investigation.
These descriptions do not assign causes. A large month could reflect a contract renewal, delayed billing, an acquisition or a posting error. Nor does a remainder automatically mean random noise. It might expose a business event the comparison omitted.
Forecasting: Principles and Practice describes decomposition in terms of a trend-cycle component, a seasonal component and a remainder. For a business review, start by recognizing those ideas rather than selecting a specialist statistical method. A historical comparison can be useful without being a fitted decomposition.
The practical level and trend guide owns how to read direction beside a reported level. The seasonality guide explains when to use the same period last year instead of a trailing average.
Take a revenue export with posting date, invoice identifier, business unit, customer identifier, currency and net recognized revenue. Retain credit memos as signed amounts. Define whether the reporting date is the accounting posting date or another business date, and use that choice throughout.
Group the rows into complete calendar months. Include a calendar list so a missing month is visible rather than silently skipped. Distinguish a confirmed month with no revenue from a month whose file has not arrived; a blank is not evidence of zero sales.
Reconcile each monthly group to the approved ledger total before interpreting it. Then check that the monthly totals sum back to the selected export total. Remove any subtotal rows already embedded in the file so they are not counted alongside invoice details.
For rate measures, preserve the numerator and denominator. For example, the supplier delivery trend guide follows delivery performance by supplier and site. Combining sites requires their eligible delivery counts, rather than averaging percentages from sites of different sizes.
A US distributor reviews its first six complete months of 2026 against the same months of 2025. Both columns use the same business boundary and net revenue definition. These are illustrative company records, not an industry benchmark.
| Month | 2025 revenue (USD) | 2026 revenue (USD) | Change (USD) | Year-over-year change |
|---|---|---|---|---|
| January | 100,000 | 110,000 | 10,000 | 10.00% |
| February | 120,000 | 132,000 | 12,000 | 10.00% |
| March | 150,000 | 165,000 | 15,000 | 10.00% |
| April | 130,000 | 143,000 | 13,000 | 10.00% |
| May | 140,000 | 154,000 | 14,000 | 10.00% |
| June | 160,000 | 198,000 | 38,000 | 23.75% |
| Total | 800,000 | 902,000 | 102,000 | 12.75% |
Each change is current revenue minus prior revenue. Each percentage is that change divided by prior revenue. The total percentage uses 102,000 / 800,000, rather than an average of monthly percentages. Percentages here are rounded to two decimal places; dollar inputs and calculations remain unrounded.
The first five months are each 10.00% above the corresponding prior-year month. June differs. To make that exception explicit, apply a comparison assumption of 10.00% growth to each prior-year amount, then calculate the remainder.
| Period | Prior-year revenue (USD) | Comparison at 10% growth (USD) | Actual revenue (USD) | Actual less comparison (USD) |
|---|---|---|---|---|
| January-May | 640,000 | 704,000 | 704,000 | 0 |
| June | 160,000 | 176,000 | 198,000 | 22,000 |
| Total | 800,000 | 880,000 | 902,000 | 22,000 |
The check is 800,000 + 80,000 + 22,000 = 902,000. The middle amount is 800,000 multiplied by the stated 10.00% comparison assumption. This is an arithmetic bridge, not proof that the underlying trend is exactly 10.00%.
June is 22,000 / 176,000 = 12.50% above that comparison. Review June's invoices and credit memos to identify which customers and transactions account for the difference. The repeated monthly shape suggests a calendar pattern worth examining across more years; these rows alone do not establish stable annual seasonality.
Finance owns the period definition, ledger reconciliation and treatment of reporting changes. The business owner supplies the explanation for operational events and decides what response the evidence supports.
In this example, the sales leader should investigate June's extra revenue before increasing ongoing capacity. A repeat order pattern and a one-time catch-up invoice have different implications. Operations should compare physical activity with the revenue movement if staffing or inventory is the decision.
Forecasting comes after this review. Preserve the exceptional amount and its explanation so a future baseline does not automatically repeat it. If supplied forecasts use an ARIMA approach, ARIMA time series analysis addresses how to judge those outputs against actual results.
Bring a dated revenue or operating export, or a Google Sheet, with consistent measures and reporting periods. Covirage maps the columns, and its tools compute and check the historical comparisons; the external AI model explains the result with every figure traced to its rows. Create a free Covirage account in about a minute, then set up your business and bring a sample of your own data.
See forecast analysis with Covirage for the supporting business context.
Monthly recognized revenue for the same business unit is a time series. Keep the revenue definition, currency and reporting boundary consistent. A list of different customers' revenue for one month is instead a comparison across customers, even if the rows happen to be sorted.
The history must cover the pattern you want to examine. A short run can show recent movement, but recurring annual seasonality needs repeated annual observations. New products, acquisitions and reporting changes can limit how much older history remains comparable.
It identifies when a movement occurred and which segments contributed. Establishing the cause needs additional evidence, such as invoice details, price changes, campaign dates or capacity records. A movement that coincides with an event does not by itself establish that the event caused it.
Historical analysis describes patterns and checks whether comparisons are fair. Forecasting uses information available at a particular date to estimate later outcomes. The historical work helps define a reasonable baseline, but a forecast still needs evaluation against results that were not available when it was prepared.