Make the denominator and the boundary visible

I ask what one observation represents before combining measures. A daily average and an order-level average can both be valid while answering different questions. When the number of orders differs between days, treating the daily values equally does not produce the combined order average. My 2020 approval failed to preserve that distinction in the label. I now want the component totals, counts and definition available alongside the result so an authorised reviewer can reconstruct it. Weighting is not a universal instruction to use one formula everywhere; the appropriate basis depends on the intended question and the available evidence. Finance and commercial owners still decide how the measure informs their decisions. The data function should make the calculation and its scope explicit, including when a comparison with an earlier report no longer has the meaning the heading suggests.

An event timestamp, a local calendar date and an agreed trading cutoff are different pieces of information. I work with technology, finance and operations to establish which one a report is intended to use. The original timestamp should remain traceable while the reporting date is derived under the agreed convention. A conversion should not silently rewrite an accounting period or assume that every service closes at midnight. I ask the team to test boundary examples and retain the relevant timezone rule rather than rely on the server’s current local setting. My role is to make the business definition consistent and inspectable, not to choose financial policy alone. A correct conversion can still answer the wrong business question if the cutoff was never agreed or if the label implies a different date basis to the reader.

A return record with no reason is not evidence that the customer had no problem. I preserve missingness in the account and distinguish the share among recorded answers from the share among all reviewed returns. Service owners can then decide what further investigation is justified without treating an unrecorded explanation as a known preference. I do not automatically impute the observed distribution to the missing group, because the people or processes producing the blanks may differ. Improving collection also needs privacy and service judgement: more compulsory fields are not necessarily a better customer process. The analytical task is to make the uncertainty usable and to show what the existing records support. It should not manufacture a complete reason profile simply because a full chart is easier to present than an explicit unknown category.

As CDO I lead the team and recommendations behind the analytical account. As an executive director I also participate in collective oversight of the enterprise. I state those roles separately because a statutory appointment does not turn my functional interpretation into an unquestionable board conclusion. I want other directors to understand the population, time basis and missing evidence well enough to challenge the claim without becoming analysts themselves. My employment and shareholding also remain relevant to any outside role; they are not erased by experience in the boardroom. The receiving company would need its own assessment of suitability and independence. A useful data contribution should help directors recognise which question a number answers and which professional owner must resolve the remaining issue, rather than add another layer of technical authority over management.