Home·Insights·Data readiness: the right but inconvenient answer

Data readiness: the right but inconvenient answer

The honest data audit usually says «not yet». The companies that hear it and act on it are the ones whose AI works.

Digital & AI2026-10-017 min readAtlas Strategy Group

Ask an AI vendor whether the company's data is ready and the answer is a slide with the word «journey». Ask honestly, and the answer is usually «not yet» — and this is the answer that separates the companies whose AI compounds from the ones whose AI produces demos. Data readiness is the least glamorous topic in the field and the most predictive one: almost every disappointing AI project traces back not to the model but to what it was fed.

What «ready» actually means

  • Findable: the data exists and someone can locate it — in named systems, with named owners, not in the departed manager's laptop. Most mid-sized companies fail this test before any technology is discussed.
  • Consistent: the same thing is called the same thing across systems; the customer has one identity; the statuses mean what they say. The company with four definitions of «active client» does not have data yet — it has four habits wearing a database's clothes.
  • Complete enough: not «big» — sufficient. Ten months of clean history on the process being automated beats ten years of noise. Readiness is scoped to the use case, which is why the audit follows selection rather than preceding it.
  • Legally clean: permissions to use it for the intended purpose exist — checked before the project, not during the lawyer's review of its launch.

Why nobody wants the honest answer

The honest audit lands badly because it delays: the budget is approved, the strategy is announced, and here is a document saying the foundation needs two quarters first. The convenient route — start anyway, fix the data «as we go» — is how the two quarters become two years inside the project, invisibly, at project prices. The companies that succeed treat the «not yet» as the project's first deliverable: a scoped, time-boxed cleanup aimed at exactly the use cases selected — the assessment's shortlist, not the whole company's history; the readiness work buys the right to skip everything else.

Data readiness is the only part of an AI strategy where «we’ll fix it later» reliably costs more than fixing it now.

The pragmatic version

The pragmatic audit is scoped to the ambition: for each candidate use case, the four tests — findable, consistent, sufficient, clean — answered in weeks, with a costed remediation list as the output. It is the entry gate of the opportunity assessment, the quiet half of the P&L question, and the standard opening phase of the AI practice — because the model choice, the vendor choice, the architecture choice all matter less than the one it rests on, and «not yet» is only a bad answer when it is ignored.

All insights →