"We're not really AI-ready yet." I hear a version of this in almost every first call — usually as a preface to explaining why the conversation might be premature. It rarely is. The businesses that most need automation are, almost by definition, the ones whose current process is a mess: three spreadsheets that don't talk to each other, a shared inbox nobody fully owns, a report that gets rebuilt by hand every Friday because the export button doesn't quite work.
That mess isn't a prerequisite that has to be cleared before the real work starts. Untangling it usually is the real work.
Where "AI-ready" comes from
The phrase gets used by vendors selling platforms that assume clean, structured data on day one — a data warehouse, a consistent taxonomy, an API-first stack. If your business doesn't look like that, their tool won't work well, and "you're not AI-ready" is the polite way of saying "our product doesn't fit you." That's a real constraint, but it's a constraint of that specific product, not a statement about whether automation can help your business.
What actually determines fit
Three things matter far more than how clean your systems already are:
1. Is the process repeatable, even if it's messy? If a person does roughly the same sequence of steps each time — even with variations, even with judgment calls mixed in — that's automatable. The mess is a design constraint, not a disqualifier. A system built to handle inconsistent inputs from the start looks nothing like a rigid workflow tool that breaks the first time reality doesn't match the happy path.
2. Is someone doing it by hand right now? If the answer is yes, there's already a working (if slow) version of the solution. That's actually the easiest starting point — we're not designing a process from scratch, we're formalizing one that already exists in someone's head and their inbox.
3. Does getting it wrong occasionally matter less than getting it done at all? Some workflows need to be perfect — financial reconciliation, compliance filings. Others just need to save the 90% case and flag the exceptions for a human to review. Most automation projects live in the second category, and that's exactly where messy starting conditions are fine.
Automatable doesn't mean tidyWhat discovery actually looks for
In a discovery session, we're not auditing your data hygiene. We're watching for the shape of the current process — the decision points, the exceptions, the systems it touches — so we can design something that ingests it as-is. Clean data makes a build faster and sometimes cheaper. It has never once been the reason a project wasn't possible.
The actual bar
The bar isn't "is our data clean." It's "does this eat enough hours, often enough, that fixing it pays for itself." If the answer is yes, the mess comes with the project — it doesn't block it.