Where Would AI Break First in Your Company?
An AI project in an SMB rarely fails because the model is wrong. It breaks somewhere nobody looked before starting, and that place is rarely technical. A tool starts quickly. What takes months is everything around it: who owns the process, where the data lives, who decides, who corrects. In a company of 30 to 150 people, those answers are spread across several heads and nobody has the full picture, which is why the blockage shows up late.
AI readiness is often presented as a maturity level, a score from 0 to 5. A level tells you where you are, not where the first break will happen, and the break is what costs you the quarter. The useful question is: where would it break first? Ten factual questions, grouped into five areas, asked about one process — the one you would most like AI to help with.
Ownership. Does one person own this process end to end, with the power to change how it works? If two people share it, neither can decide to change it, and the AI output has nowhere to land. Second: who is accountable for a wrong result produced with AI? “It depends” is the most common answer, and it means nobody is accountable for anything.
Data. Where does the data this process depends on live? If it is scattered across personal files, inboxes and chat threads, the tool will work on whatever one person pasted in. Next: do you know what must never be pasted into an external tool? A rule that exists but was never communicated protects nothing.
Decisions. Who approves an AI tool for work? In many companies, it is whoever asked first. You also need a sponsor at management level, willing to back an uncomfortable change with time or budget. Without one, a change that makes someone’s week harder eventually stops.
Current use. Your teams probably already use AI tools. Do you know which ones, and what for? Do they have a shared way to check results? Advice traded between colleagues is a start, not verification.
Correction and measurement. When a result is wrong, is there a known route to report it and fix the process, with someone who acts? And do you know today’s baseline in time, cost or error rate? Without that baseline, you cannot say afterwards whether the project helped. You will have an impression.
For each question, score 0 if the answer is no or nobody is named, 1 if it is partial or informal, 2 if it is in place and people know it. Add up by area. The three weakest areas are your first break points. Count the zeros separately too: a company with a partial answer in every area is not in the same position as a company with four blocked answers, even if the totals look similar. Zeros are blockages; ones are work in progress.
If you would rather not add it up by hand, I built an assessment: ten questions, about five minutes, and you get your three weakest points with one concrete next step for each. You can receive the results by email or download them directly.
Take the AI readiness assessment
One caveat, with or without the tool. The questions are factual, but you are answering about your own company, and a leader tends to mark as “in place” what they announced. Answer as a new arrival would discover it on their first day, and if you hesitate between two scores, take the lower one.