A blank field can make a spreadsheet look untidy, so it is tempting to replace it with zero. In operational analysis, that shortcut can change the meaning of the record. An unknown closing time is not an instant resolution. A missing payment-volume count is not a period with no payments. A case without a category is not proof that no exception occurred. Missingness deserves its own treatment.
The examples here are fictional and are not evidence about Corpay Prepaid, Fintwist, or any employer. The aim is to help analysts communicate data quality before presenting a service-quality conclusion. NIST’s statistical guidance emphasizes understanding data and assumptions; an attractive chart cannot repair an input whose meaning has been silently changed. A visible gap is often more honest than a complete-looking series.
Define the missing states you actually have Separate “not recorded,” “not applicable,” “not yet known,” and “failed to import” where the source can support those distinctions. A closure date absent because a case is open differs from one absent because an old system never stored it. Use explicit status fields rather than relying entirely on an empty cell. If you cannot determine the reason, record it as unknown instead of choosing the most convenient explanation.
Keep zero as a valid observed value when it really means zero. A confirmed count of no qualifying events is different from an unavailable count. The same applies to time: a same-day case may have a positive elapsed duration even if a rounded day count displays zero. State the unit and rounding rule so that a zero does not conceal a short but nonzero interval.
Report completeness beside the result Suppose a fictional dataset contains 100 closed cases, but only 82 have valid opening and closing timestamps. A duration average based on those 82 should say so. Reporting “average for 100 cases” would misstate the calculation. Include the eligible-record count, excluded-record count, and reasons. If missing records differ systematically from complete ones, the observed average may not represent the full group.
For example, older complex cases might be more likely to lack timestamps after a system migration. Excluding them could make the measured duration look shorter. That possibility is not proof of bias, but it is a reason to investigate. Compare the available characteristics of complete and incomplete records where authorized and appropriate. Do not assume missing data are random simply because you cannot see the missing values.
Avoid hidden imputation Imputation means filling a missing value using an assumption or model. It can be appropriate in some statistical work, but it should not happen invisibly through a spreadsheet default. Replacing a missing duration with the median, carrying forward a previous payment count, or treating blank categories as “no issue” all impose assumptions. Label any such method and explain why it fits the purpose.
For a small operational report, it may be better to leave the value unknown and show a range of possible effects. If the missing denominator could materially change the conclusion, say that the rate cannot yet be interpreted. Do not generate a precise percentage merely because a presentation template expects one. The audience should know whether the number was observed, calculated from complete data, or partly estimated.
Use bounds when they are informative Consider a fictional set of 200 expected payments with 6 confirmed affected payments and 4 unresolved reports, each tied to a different payment. The confirmed observed proportion is 3 percent. If every unresolved report became a confirmed exception, the affected proportion would be 5 percent. That range is a simple scenario bound under the stated assumptions, not a confidence interval or a prediction.
The assumptions matter: unresolved reports must refer to distinct eligible payments, and no additional unreported issues are included. If those conditions are not met, the bound changes. Explain what the range does and does not cover. Scenario arithmetic is useful when it reveals uncertainty; it becomes misleading when presented as a statistical guarantee about the real underlying process.
Make gaps visible in charts A missing period can appear as a gap, a labeled unavailable point, or a separate completeness panel. Do not connect a line across it without indicating that no observation exists. Avoid using the same visual treatment for zero and unavailable values. In a table, plain labels such as “not recorded” can be clearer than a cryptic code that readers may mistake for a number.
If a chart shows only complete records, state the rule near the chart and provide the excluded count. A footnote hidden in a distant document may not travel with a copied image. Keep enough context in the title or caption to prevent a screenshot from becoming an unsupported claim. Exported charts should preserve the period, population, units, and data-quality caveat.
Treat better recording as a possible explanation After a process change, reported exception counts may rise because detection improved. Missing timestamps may fall because staff began recording them consistently. Those changes can improve the evidence even when the headline metric initially looks worse. Separate data-quality improvement from service-performance improvement, and avoid penalizing a team merely for making previously invisible issues visible.
A responsible conclusion states what is known, how much is missing, and whether the missingness could change the interpretation. Unknown values are not failures of presentation. They are facts about the evidence. Preserving them helps keep a payroll service-quality report useful without pretending to know more about an employer or provider than the records support.