'Our data isn't ready for AI' is the new 'we'll do it next quarter.'

'Data readiness' used to be real advice, and the AI data readiness version inherited its authority. In the BI era, a dashboard was only as good as the warehouse behind it, and a reporting project built on dirty tables was money set on fire. A generation of executives learned the rule the hard way: clean the data first, then deploy the technology. It was correct, and it stuck.

Then generative AI arrived and the rule quietly stopped applying to most of the work, but the reflex stayed. 'Our data isn't ready' is now the most respectable way to defer. It sounds prudent, it comes with a project plan, and it commits you to nothing. I'd argue it's part of why the adoption gap between small companies and enterprises looks the way it does: the enterprise runs its data program and its AI program at the same time, while the smaller company queues one behind the other.

I hear this line from executives almost weekly, and I've learned to ask one follow-up: ready for what, exactly? Nobody is proposing you point a language model at your general ledger on day one. The work your team would actually start with doesn't touch the warehouse at all.

The AI your team would use tomorrow morning reads documents. It drafts from your proposals, summarizes your meeting transcripts, answers questions from your contracts, triages your inbox, and makes sense of the spreadsheet with the weird tabs that only Brenda understands. None of that requires a schema, a pipeline, or a single deduplicated record. The mess is the input. That is the whole point of these models.

A practical test: pick one document-heavy workflow, say proposal drafting or support replies or contract review. If the team can gather that workflow's material in an afternoon, a folder of examples and the source documents they work from, you are ready to start. The same logic runs through the first ninety days of an SMB AI transformation: one function, one workflow, depth before breadth.

Here's the honest caveat. Structured-data analytics still rewards clean pipelines. If the goal is reliable forecasting, reporting, or anything where a wrong number compounds, your data and analytics people are right to insist on quality work, and that project deserves real investment. But that's an argument for sequencing, not stalling. Run the cleanup where it's genuinely needed, in parallel, while the document-heavy half of your company starts now.