The AI readiness checklist for a 20-person company

At twenty people you have no ops department, no data team, and no slack in anyone's calendar. That makes AI tempting, because leverage is exactly what you lack. It also makes a failed AI project more expensive for you than for a company that can absorb the write-off.

Readiness is not about being technical. Every item below is something an owner or office manager can check in an afternoon, without a consultant. If most of them are true, AI is worth the spend. If most are false, fix those first; it will be the cheapest AI work you ever do.

The short version

Your data

AI works on what you can show it. The bar is lower than people think, but it is not zero.

• It exists in a system, not in heads. Customer records, orders, tickets, or files live somewhere a computer can reach: a CRM, a shared drive, a spreadsheet. "Ask Sandra" is not a data source.
• Someone can export it. If nobody on the team can pull a CSV or grant API access to the tools you use, every project starts with a fight against your own stack.
• It is roughly consistent. Perfect is not required. But if the same customer appears three ways in three tools, dedupe comes before AI.

Your process

The workflows you want to improve have to hold still long enough to improve them.

• The steps are written down. Even one page per workflow. If the process only exists as habit, document it first; the writing usually exposes the real problem.
• The steps are stable. If how you handle an order changes every month, automation will always be one rewrite behind reality.
• Someone owns each workflow. One person who can say how it works today and approve how it should work tomorrow. Committees stall AI projects faster than bad data does.

Your team

Tools do not adopt themselves. The team conditions matter more than the model choice.

• There is an internal owner. Not a developer, an owner: someone who cares whether the thing gets used and has the standing to nudge people.
• People have a few hours to give. Mapping workflows and testing outputs takes real time from the people who do the work. If the team is at 110%, schedule the project for when they are not.
• Leadership will accept "not yet." If the project is only allowed to conclude that AI is needed everywhere, you will buy shelfware. The honest answer is sometimes to wait, and that has to be a permitted outcome.

The checklist

Run through it honestly. Count the items that are true today.

• Our core records live in systems a computer can reach, not in inboxes and memory.
• Someone on the team can export data or grant access without outside help.
• The workflows we want to improve are written down, even briefly.
• Those workflows have been stable for at least a few months.
• Each workflow has one named owner.
• Someone internal will own the project and its adoption.
• The team can spare a few hours a week while we build.
• "You don't need this yet" is an acceptable answer.

Six or more: you are ready, and the question becomes where AI pays off first, which is what the AI Opportunity Diagnostic answers in five days. Four or five: start with the gaps; most close in a few weeks. Three or fewer: do not buy anything yet. Fix the list, then come back to AI with a foundation it can actually multiply. Not sure a given task belongs on your shortlist at all? Start with how to tell if a task is worth automating.