When Two Reports Disagree, Somebody Has to Decide Which Is Wrong

There is a meeting that happens in most mid-market businesses, usually monthly, in which two numbers for the same thing appear in the same pack. Revenue by channel. Active customers. Units shipped. The numbers do not agree, and the first fifteen minutes of the meeting are spent establishing which one to believe rather than deciding anything.

The reflex is to treat this as a reporting defect — a system needs fixing, a report needs rebuilding, someone should reconcile the two. That reflex is why it recurs. The gap between two numbers is almost never a technical fault. It is an unresolved question about definitions that has been quietly delegated to whoever built each report.

What it costs, roughly

Estimates for the cost of poor data quality are large and vary widely by methodology. Gartner’s frequently-cited figures sit between $12.9m and $15m a year for the average enterprise, drawn from surveys of organisations sophisticated enough to be buying data quality tooling. MIT Sloan Management Review research has put the revenue impact at 15–25%.

Those numbers are worth treating carefully. They come from large enterprises and self-reported surveys, and a mid-market business should not simply scale them down. What they establish is direction and order of magnitude, not a figure to put in a business case. The useful version of this number is the one you calculate yourself, and it is calculable.

Why the two numbers exist

In almost every case, both reports are computing something reasonable. They differ because they were built at different times, by different people, answering slightly different questions, and nobody wrote the questions down.

The recurring causes are unglamorous.

Different points in the lifecycle. Sales counts at order. Finance counts at invoice. Marketing counts at qualified opportunity. Each is right for its purpose, and the difference between them is real business — orders that have not shipped, invoices not yet raised.

Different treatment of the awkward cases. Refunds, credit notes, intercompany, cancelled-and-rebooked. One report nets them off; the other does not. Neither is wrong, and the divergence grows with volume.

Different definitions of the entity itself. Is a customer a company or a purchasing site? Is a group with four subsidiaries one customer or four? Two systems answer differently because two teams needed different answers, and both answers are now embedded in reporting.

Different time boundaries. One report uses calendar month, another the accounting period, a third a rolling four weeks. In a month with an awkward weekend, that alone produces a visible gap.

The reason it never gets fixed

It is not difficulty. Reconciling two figures is generally a day’s work for someone who understands both.

It is that nobody has the authority to lose. Settling the definition means telling one function that its number — the one on its dashboard, in its targets, possibly in its bonus calculation — is not the number the business will use. That is a decision with a loser, and in the absence of someone empowered to make it, the organisation defaults to the arrangement where both numbers survive and the discrepancy is re-explained at every meeting.

So the discrepancy becomes folklore. New joiners are told, in their first month, that “the sales number always runs higher, you just have to know”. The knowledge transfers informally and the underlying question is never asked again, because everyone competent has already adapted to it.

What actually resolves it

The fix is governance, and it is smaller than the word suggests. It does not require a data governance programme, a council, or a tool.

Name an owner per definition, not per system. Someone owns what “customer” means. Someone owns what “revenue recognised” means. Their job is to decide, not to build reports. This is the step that is usually skipped, because owning a system feels like a role and owning a definition feels like a formality until you need one.

Write the definition down where the number appears. Not in a data dictionary nobody opens — next to the figure, in the pack. Including the awkward cases: what happens to refunds, to intercompany, to a cancelled order rebooked next month.

Let the other numbers exist, and label them. Sales should keep its order-book number; it is the right number for running a sales team. What changes is that it is labelled as the order-book number rather than presented as revenue. Most of these disputes evaporate the moment both figures are named accurately.

Decide once, in a room, with a name against it. The output is a short list of settled definitions with an owner each. Not a project.

Why it is worth doing before anything else

This is the least fashionable work available. It produces no capability, demonstrates nothing to a customer, and cannot be shown in a board pack except as an absence of argument.

It is also the thing that determines whether everything built on top of it holds. Every analytics investment, every automated decision, every forecast inherits these definitions. A business that has not settled what a customer is will find that question resurfacing, more expensively, in every system it subsequently buys — and by then the disagreement is encoded in four places instead of two.

The last question in the box above is the one to start with. Most management teams have never added up the hours spent reconciling rather than deciding. It is usually a larger number than anyone expects, it is entirely recoverable, and it costs a decision rather than a budget.