Your data model is a record of old arguments

Software can be replaced in a quarter. Infrastructure can be replaced over a weekend if you are brave. A data model cannot be replaced at all, because it is not really a technical object. It is the accumulated residue of every decision an organisation has made about how it wants to see itself.

Look at any long lived cost centre hierarchy. Somewhere in it there is a branch that makes no sense on the current org chart. Trace it back and you find a business that was acquired in 2016, run separately for reporting because its founder negotiated that, and then never folded in because the person who would have had to fold it in reported to that founder. The hierarchy is not wrong. It is a minute of a meeting nobody wrote down.

Why this makes rationalisation slow

The usual instinct is to treat the model as a design problem. Get the right people in a room, agree the target structure, map old to new, cut over. That works for the eighty per cent of the model that everybody agrees on and stalls completely on the twenty per cent that encodes an unresolved disagreement.

You cannot map your way out of a governance question. If two divisions define gross margin differently because they sell differently, a mapping table does not settle which definition the group reports. Somebody has to decide, and that somebody is usually two levels above the project.

So the work of rationalising a model is mostly the work of surfacing which structures are load bearing arguments and getting them settled by the people entitled to settle them. The technical migration is the easy half.

Why the reference model does not save you

Every vendor and every large consultancy has an industry reference model, and they are genuinely useful as a starting shape. What they cannot carry is the reason your model differs from theirs. Some of those differences are debt. Some of them are the business.

We have seen a manufacturer adopt a clean reference chart of accounts and lose, in the process, the granularity that let them price contract variations. Nobody noticed for two quarters, because the reports still balanced. The model was correct and useless, which is the worst combination available.

How to tell the debt from the business

Three questions, asked of each structure that looks odd.

Who reads it. If a dimension has no consumer, no report, no rule, no allocation, it is debt. This clears more than people expect.

What breaks if it goes. If the answer is a specific decision that a specific person makes on a specific cadence, it is business. If the answer is that a report would look different, ask who reads the report.

When was it last changed. A structure that has been static through two reorganisations is either fundamental or forgotten, and the owner will know which within about a minute.

The part that compounds

Whatever you decide, write down why. Not the mapping, the reason. The next team to touch this model will be six people who were not in the room, in four years, under time pressure, and they will either find your reasoning or repeat the argument from the start. Almost every expensive model rebuild we are asked to run is an argument being had for the second or third time, because the first answer survived only in the structure and not in words.