The AI Readiness Assessment We Actually Run

Most AI readiness assessments are a sales deck wearing a clipboard. Forty questions, a maturity score, and a recommendation to buy whatever the assessor sells.

Here is the one we actually run. I am giving you enough detail to run it yourself, because the companies that do usually call us anyway, for the parts that are hard.

The Premise

AI projects do not fail on model quality. They fail on the boring stuff underneath: dirty data, undocumented workflows, missing metrics, and ungoverned access. So a real readiness assessment barely mentions AI. It examines the foundation AI will stand on.

Five checks. Red, yellow, or green on each.

Check One: Data

Where does the data that matters actually live, who owns it, and would you let a new employee see all of it on day one?

Green means the critical data has a known home, a named owner, and access rules. Red means the real answer is "spreadsheets, seven tools, and Dave's laptop." Most companies are red and do not know it, because nobody ever asked the question out loud.

AI on ungoverned data does not create risk. It reveals the risk you already had, at machine speed.

Check Two: Workflows

Pick the three processes that eat the most hours in your company. Are they written down anywhere, in enough detail that a competent stranger could follow them?

You cannot automate what you cannot describe. When a workflow lives only in someone's head, the AI project becomes an archaeology project first. That is fine, but budget for it.

Check Three: Metrics

If we automated your intake process tomorrow, what number would move, and do you measure that number today?

Green is a baseline that exists before the project starts. Red is "we will know it when we feel it." Without a baseline, every AI initiative succeeds in the demo and dies in the budget review, because nobody can prove it did anything.

Check Four: Guardrails

Who is allowed to put what data into which tools? Is there any policy at all? Any audit trail?

I am not asking for a governance department. One page of acceptable use, access tiers for anything that touches customer data, and a log of what automated systems did. Simple and secure beats clever and exposed. If the honest answer is that employees are pasting customer records into free chatbots, that is your first project, before any of the exciting ones.

Check Five: People

Has anyone on the team taken an AI tool past the toy stage into daily work? Is there an executive who owns this, with budget and authority?

Tools do not adopt themselves. The gap between AI capability and AI adoption is people, and the single best predictor of success I have seen is one named owner plus a handful of internal enthusiasts who already use this stuff without being told.

Scoring It

Five greens: you are ready, and honestly rare. Pick the highest-value workflow and go.

Mostly yellow: normal. Sequence the fixes, then build. Sixty to ninety days of foundation work beats a year of stalled pilots.

Two or more reds: do not buy anything yet. Anyone selling you an AI platform in this state is selling you a monument to future regret. Fix the boring stuff first. Build the boring stuff first, always.

What Comes Out the Other Side

The output of a real assessment is not a score. It is a ranked list: which workflows to automate in what order, what has to be fixed before each one, and what each is worth in hours or dollars. A number your CFO can hold you to.

That is the whole framework. Run it honestly and it will save you from the most expensive sentence in enterprise AI: "we bought the tool and then figured out the plan."