The AI readiness checklist: 20 points.

What makes a business ready for AI?

Documented workflows, data somebody can actually export, one named person with time to review what a machine produces, and a number on what the manual work costs you. That is the whole secret — not a data science team, not a bigger budget, not a platform migration.

The twenty checks below test exactly those four things. They are the same questions we ask in the first week of a paid AI Readiness Audit, published here because you can run most of them yourself. A business that passes fifteen of these will get more from a $12,000 build than an unprepared one gets from a $200,000 engagement.

How do you use this checklist?

Score one point for every item you can check off without squinting, zero for the rest — no half points. Twenty minutes if you know your business; an afternoon if you have to go look, and having to go look is itself a finding.

Be strict. The checklist only helps if the score is true, and nobody is watching you take it. If an item makes you say “sort of,” that is a zero with better posture.

Are your workflows ready? (Points 1–5)

Ready means known, written down, and owned. AI automates what you can describe; it cannot automate a vibe.

1. You can name your three most repetitive workflows without checking. If you have to ask a manager, start there — the asking is the audit.
2. At least one of those workflows is written down, step by step, somewhere other than someone’s head. A checklist, a wiki page, even a dusty SOP. Machines learn from documents, not folklore.
3. You know roughly how often each one runs. Ten times a week or ten times a day changes what a fix is worth by an order of magnitude.
4. You can describe the exceptions. Someone can say what the weird ten percent looks like — the rush order, the disputed invoice, the client who calls instead of emailing.
5. Each workflow has one owner with a name. “Operations handles that” is not a name. Automation without an owner is a stray dog.

Is your data ready? (Points 6–10)

Ready does not mean clean — nobody’s data is clean. It means you know where each fact lives and you can get it out.

1. You can say which system is the truth for each core fact. Customer, order, invoice, job status: one home each. Two systems that disagree is the most expensive kind of normal.
2. No critical number lives only in a spreadsheet on one person’s desktop. If payroll week depends on Carol’s laptop surviving, fix that before any AI conversation.
3. You can export your own data without calling a vendor. Try it this week. A system that holds your data hostage will hold your automation hostage too.
4. You know what you are obliged to keep private, and for whom. Clients, patients, employees — the obligations differ, and the build must honor them. Where data may and may not go is the subject of Where Your Data Goes.
5. Someone can pull last month’s numbers in under an hour, without IT. If a basic report takes a week, the machine will automate confusion at scale.

Is your team ready? (Points 11–15)

Most AI projects fail on people, not technology. These five checks find out before the invoice does.

1. Someone at the top actually wants this. A sponsor who will spend attention on it — not an owner who merely tolerates the experiment.
2. A specific person has two to three hours a week to review what the system drafts. Every system worth building routes judgment to a human. That human needs a calendar, not a title.
3. Your team has survived a process change in the past year. New POS, new scheduling tool, new anything. If the last change died in quiet mutiny, the pattern will repeat.
4. You can name who says “stop.” When the machine misbehaves, one person has the authority to pull the cord and the standing to be believed.
5. Training sticks here. Your people have learned a tool before and still used it in month three. Software that nobody opens is furniture.

Are your economics ready? (Points 16–20)

AI is an investment with a payback period, not a lottery ticket. These checks confirm you can do the arithmetic — because a vendor certainly will not do it against themselves.

1. You know what an hour of admin work costs you, fully loaded. Wage plus taxes plus benefits. Every automation case is built on this number.
2. You can name the three workflows that cost the most. Not the most annoying — the most expensive. They are frequently not the same list.
3. You have a budget range, even a soft one. For scale: our audits run $3,500–$8,500 and sprints $12,000–$45,000, fixed in writing. If those numbers are unthinkable, the answer is “not yet,” and that is a fine answer.
4. You expect ranges, not guarantees. Anyone who promises exact ROI is selling something other than results. We publish our assumptions; a promise dressed as a number is a red flag.
5. You are willing to stop a failing project at day 45. Not day 300. The discipline to quit early is worth more than the discipline to start.

How do you score it?

Sixteen or more: you are ready to build, and your problem is choosing what first — an audit will rank the options by payback. Ten to fifteen: you are ready to map but not yet to build; start with the audit, or work the failed checks yourself first. Nine or fewer: wait. The long version of “wait” — the signs themselves, and a 90-day plan that costs almost nothing — is in Signs You Are Not Ready for AI.

One pattern is worth naming because it is so common: a business that scores eighteen on workflows and economics but fails points 11 through 15 — the team section. That firm will buy a good system and starve it, because nobody has the hours to review what it drafts. If your zeros cluster in one section, fix that section before anything else; a lopsided score fails in more expensive ways than a low one.

Waiting is not falling behind. Most of what a low score reveals — an undocumented workflow, a hostage export, a missing owner — costs little or nothing to fix and pays off whether or not you ever build. The items in the data section are the usual culprits, and most are a policy decision, not a purchase. Fix the foundations, retake the list in a quarter, and you will have gained ground on competitors who bought first and prepared never. When you do take meetings, bring the ten vetting questions, and know what the market charges before anyone quotes you.

Is there an automated version?

Yes — the AI-Ready Score: fourteen questions, about seven minutes, and you get a 0–100 score plus the one constraint to fix first. No sales call follows it.

We built it as the instrumented version of this page. The checklist tells you what to look at; the Score weighs the answers the way we would in an audit and hands you a number you can retake next quarter. Run both — the gap between your self-score here and your Score there is usually the most instructive number of all.