Your data isn't ready. That's normal.

"We'd need to sort our data out first" is one of the most common reasons businesses give for not starting. It is usually true, and it is usually the wrong conclusion to draw from it.

The belief underneath it is that there is a state called "data ready", reached through a long tidying project, after which AI becomes possible. That state does not exist. Established businesses accumulate mess as a by-product of operating, and the mess is never finished being cleaned.

The question is narrower than it sounds

What matters is not whether your data is good. It is whether the specific information required by one specific piece of work is available, findable and accurate enough for that purpose.

Those are very different questions. A firm might have a chaotic CRM, inconsistent file naming and three overlapping spreadsheets — and still have everything needed to automate document summarisation, because that work only touches the documents.

Assessed as a whole, the business looks unready. Assessed against a defined task, it is fine. Most "we need to sort our data out" conclusions come from making the first assessment when only the second one matters.

Four gaps, only two of which are blocking

1. Messy but present Inconsistent formats, varied naming, duplicates. Irritating, and largely tractable — this is precisely the kind of variation modern systems handle. Not a blocker.

2. Scattered The information exists but lives in four systems and an inbox. Not a blocker either, though it moves cost from the model into the integration work, and the estimate should say so.

3. Unrecorded The information was never captured. Why a client left, why a job was priced that way. This is a genuine blocker, and no amount of technology recovers it. Start capturing it now and revisit later.

4. Wrong Records exist and are inaccurate. The worst case, because it is invisible: a system built on it produces confident, plausible, wrong output. This one has to be fixed first.

Messy data is a cost. Missing data is a constraint. Wrong data is a hazard. Treating all three as the same problem is what turns a six-week project into a year.

What to do instead of a tidying project

Pick the work you want to improve. List only the information that work touches. Assess that list against the four categories above. In most cases the honest finding is that two fields are wrong, one thing was never recorded, and everything else is merely untidy.

That is a fortnight of targeted work, not a data transformation programme. And it produces something a general clean-up never does: a clear statement of what the project depends on, which is exactly what you need to know before committing to it.

The part worth doing anyway

One exception is worth acting on regardless of any current project: start capturing the things you know you are not capturing. Reasons for decisions, outcomes, why a case went the way it did.

That information has no value today and cannot be recovered later. It is the cheapest possible investment in your own future options, and it is the one thing on this list that genuinely does get harder the longer it is left.