Two directors bring the same KPI to a meeting with different numbers. Both are confident. Both can show their working. Twenty minutes later nobody has discussed the business, and someone has been asked to go away and reconcile.

This is rarely one person's mistake. It is how reporting grows. Knowing how a single measure drifts into several definitions is the first step to fixing it, and it matters most when you are about to modernise.

How one KPI becomes four

Nobody sets out to calculate Gross Margin four ways. It happens one sensible decision at a time, usually under deadline.

  • Hard-coded rates. An analyst needs cost of sales before the ledger detail is ready, so they estimate it as 58% of revenue and type =B2*0.58 into a cell. It goes into the board pack, and years later the rate is still there with nobody sure why.
  • Copied reports. A regional team needs the same report with a different view. They copy it and adjust the rate to local experience. Now two versions exist, each right for the person who made it, and they quietly disagree.
  • Local fixes. Someone spots a figure that looks wrong for one period and patches their own copy. The fix never reaches the original, or the systems that feed the dashboard.
  • Different tools, different logic. The same measure can be calculated separately in Excel, SQL Server, SSRS and Power BI, each by a different person against a slightly different source. Exclusions and period cut-offs drift apart.

None of these is reckless. Each solved a real problem on the day. The cost shows up later, when the versions meet in the same room.

Why it matters more in a transformation

Today, conflicting definitions are an irritation. People know which report to trust, or which number to quote to which audience. That knowledge lives in their heads.

A transformation removes that safety net. The new reporting needs one definition of each measure, built once and shared. Whoever builds it has to choose. If they choose without the owners, the first month-end on the new reporting becomes the meeting where everyone finds out their number has changed.

  • The build team uses whichever version it finds first, so the choice is accidental.
  • User testing turns into a debate about definitions, and releases slip.
  • People keep their old spreadsheets running to check the new numbers, so the duplication survives the programme.
  • Confidence in the new reporting drops before it has had a chance to earn it.

How to resolve it

Resolving conflicting definitions is mostly a people problem with a technical starting point. Your team needs to put every version side by side. The right person then chooses.

  • Find every version. Ask the team how the number is actually calculated everywhere it appears, not just where it carries the name. Hard-coded rates are the giveaway.
  • Put them side by side. Show each version, where it lives and what relies on it. A small difference looks different once you see that one version feeds the board pack.
  • Name an owner for the measure. Usually a finance or business lead, not the BI team. The BI team explains the options; the owner decides.
  • Agree one definition and write it down in plain English, including the source, what is excluded and the period rules.
  • Decide what happens to the other versions. Each one is replaced by the agreed definition, kept for a stated reason or retired with its report.
  • Build it once. The new reporting uses the agreed definition, and owners of reports that used another version are told before go-live.

Keep the decision record with the definition. When someone asks in a year why the margin moved, the answer should be a link, not an archaeology project.

Not every difference is a mistake

Some variation is legitimate. A management view may exclude one-off items that the statutory view includes. A commission model may use a different rate for a different product line.

The problem is when those variants share a name. If both are called Gross Margin, readers cannot tell which one they are looking at, and the argument starts again every time the two meet.

The goal is not to force every number to be the same. It is to make every difference deliberate, named and owned. A useful test: if two versions differ, can someone explain why in one sentence? If so, give the variant its own name. If not, it is drift.

Start with the map

You cannot resolve conflicts you have not found. Most organisations find them one at a time, after the numbers have already disagreed in front of someone senior.

Mapping how each number is calculated turns that into a list owners can work through, before the rebuild locks a choice in. It also tells you how big the problem is, which helps when you are asking for time to fix it.