Why two departments disagree about the same job

Ask four people in one company how many active jobs it has and you will get four numbers. This is not a story about bad data. In most operational businesses every one of those four numbers is being calculated correctly, from a system that is working exactly as designed.

The reason is worth understanding, because it decides whether the problem is fixable by a data project or not.

Each department is right, on its own terms

Finance counts a job from the moment money is committed, because that is when it has to be reported. Operations counts it from the moment a crew mobilises, because that is when it consumes capacity. Legal counts it from signature, because that is when obligation attaches. Procurement may count it from the purchase order.

Each of those definitions is not a preference. It is the definition that makes the department answerable for what it is answerable for. A finance team that recognised work at mobilisation would misstate its own accounts.

Which is why the reflex of asking everybody to align on one definition tends to fail. You are asking three of the four to hold a version that is wrong for their own accountability.

Why a data quality project does not fix this

Data quality work finds records that are wrong: a duplicate, a missing field, an inconsistent format, a value that violates a rule. All of that is real work and none of it touches this problem, because here no record is wrong.

A cleaning exercise run over these four systems will report that they are in excellent condition. Every field is populated, every value is valid, every total reconciles inside its own ledger. The disagreement lives in the space between them, which is precisely the space a per-system quality tool cannot see.

Informatica sells data management software, so this is vendor commissioned research. It is quoted here because the sample and date are stated and because it measures the barrier directly. Read "reliability" carefully though: a number can be perfectly reliable and still disagree with another perfectly reliable number.

Gartner measured something adjacent and the gap it describes is the same one.

The uncertainty in that figure is the interesting half. A company that is unsure whether its data is fit for purpose is usually a company where each department is confident about its own and nobody owns the seam.

What the disagreement actually costs

For years the cost was tolerable, because a human absorbed it. Somebody in the middle knew that the operations number excludes the two jobs awaiting signature, and adjusted. That knowledge was never written down and it was never in a system. It was a person.

Two things have changed. The first is scale: at ten jobs the reconciliation is a conversation, and at four hundred it is a full-time role nobody has budgeted. The second is that software has started to act.

That is the moment the disagreement stops being an inconvenience and becomes a risk. It is also the reason permissions are not the whole of agent safety: the agent is authorised, the call succeeds, and the object it acted on was not the one anybody meant.

What resolving it actually requires

Not a migration, and not a single system. Both of those attempt to make the four definitions into one, which fails for the reason above.

What it requires is a decision about precedence, taken by somebody with the standing to take it. For each fact the business needs to agree on, one department's version becomes the record and the others become views of it. What did it cost, who approved it, what governs it, where does it stand today. Each of those has an owner, and naming the owner is the whole of the work.

That is a governance decision wearing technical clothes, and it is why this cannot be delegated to an integration project. An integration can carry whichever answer it is told to carry. It cannot decide which one deserves to be carried.

This is one of the places where our own work stops and a person's begins, which we would rather state plainly than discover in month three. The others are described on how we work.

How to run that conversation without it stalling

1. 01Start from a real disagreement, not a definitions workshop. Take one week where the numbers differed and put the four versions on one page.
2. 02Ask each department what it is accountable for, rather than what its number means. The accountability explains the definition and takes the argument out of it.
3. 03Decide precedence per fact, not per department. Finance may own the money and operations may own the state, in the same record.
4. 04Write down the exceptions in the room. Every operational business has them, they are the reason the last attempt failed, and they are usually known to exactly one person.
5. 05Name who resolves the next conflict, before there is one. If that person does not exist, the agreement will hold until the first case nobody anticipated.

None of that requires software, and it is the part that determines whether any software helps. A company that has had this conversation can adopt almost anything. A company that has not will find that each new system becomes a fifth opinion.