AI Readiness Assessment Questionnaire: The Themes

An AI readiness questionnaire should bring four themes into the open before anyone names a product. The themes are where the data lives, whether the process is written down, who checks the output, and what a good result would mean. A form that opens with a logo has already skipped the real work.

NIST groups the risk work into four parts, which are Govern, Map, Measure, and Manage.¹ We are not asking a small firm to build a risk program with those labels on the door. We are asking the questionnaire to cover the same ground in plain speech. Govern is who gets to decide, and Map is what the job really is. Measure is how you would know the output is wrong. And Manage is what you do when it is. A questionnaire that cannot find those four ideas, even in simpler words, is really a sales form.

The framework is how we score the answers, and the assessment is the longer reading. The growth guide is what the check offers after the score. Themes come first, and a band comes after that. A tool comes last, and sometimes it does not come at all.

We like a short questionnaire because length feels thorough and often hides the blank. An owner may answer a long run of soft questions and still not say where the record lives. When that happens, the form has dodged the only question that mattered. We would rather ask four blunt questions that are allowed to be answered with “we do not know.”

What a public checklist does not claim

The AWS checklist for small and midsize businesses, updated in September 2025, walks an owner through readiness questions.² It does not publish a percentage of businesses that are ready. The OECD SME AI adoption tool, current in 2026, is a self-assessment aid for smaller firms.³ It is not a national adoption rate.

That missing number is the point of this section, so we will not rush past it. Owners often arrive after reading a headline that sounds like a share of firms. They want their questionnaire to end in the same kind of share. Then they can tell a partner that they are above it or below it. Neither of those public pages will give you that share. AWS gives you questions and a way to talk about stages of work. The OECD tool gives a smaller firm a way to check itself. Suppose your questionnaire prints a readiness rate for “businesses like yours” at the bottom. Then ask who was counted, because if the answer is nobody, you are holding a costume.

We still use those pages, but we use them as prompts and not as a league table. The AWS questions help an owner who has only thought about features. And the OECD aid helps a smaller firm that does not want an account or a stored answer. What you leave with is your own set of answers. You do not leave with permission to say that a percentage of your industry is ready. And we will not write that sentence for you because the sources will not write it either.

Ask where the one place the facts live is for the job. Ask which step is written down and which step is only habit. Ask who can reject the output, and ask what you will stop doing if the answer is weak. Those four questions are the questionnaire.

Say them in the language of the job, or they stay abstract. The question “Where do the facts live?” becomes a plainer one about a real call. When a customer calls about the thing we promised, which page does the person on the phone open? The question about habit becomes a different one. What do we do when the usual person is out, and is that written anywhere a new hire could find? The question about rejecting becomes a third one. Who is allowed to say that this message does not go to the customer? Then the question about what we will stop becomes a last one. If this works, which notebook can we close for good?

A questionnaire that cannot survive that translation is still just a theme list. Themes are not a score, because they are the questions you refuse to skip. We send people to the growth guide once a real score exists. By then the next move is about getting better at the job and not about collecting more questions. Please do not open that guide to avoid the four questions, and open it only after you have answers you can stand behind.

We also want the questionnaire to record disagreement. If the owner and the person who does the job give different answers, the form should show both and not a negotiated average. An average is how a blank turns into a tick. Write “owner says the binder, dispatcher says her head” and leave it ugly. Ugly is honest. A clean tick that needed a compromise in the hallway is how you buy a tool for a process that does not exist.

NIST’s four words help us check that we did not drift into product trivia. If the questionnaire never asks who decides, it skipped Govern. And if it never asks what the job is in steps a stranger could follow, it skipped Map. If it never asks how you would notice a wrong output, it skipped Measure. And if it never asks what you will do when the output is wrong, it skipped Manage. You can fail all four tests with a very professional-looking form, and we have watched it happen. The form had sections and a logo but nowhere to write “we do not know.”

A worked example

Say we are at the desk of a small electrical contractor in the late afternoon, when the phones have finally gone quiet. There is an owner, a dispatcher, and a lead who actually walks the jobs. This is an example we walk through, and it is not a firm we audited. We are not going to pretend that we counted their jobs.

The owner was handed a questionnaire by a software seller. It opened with the product name and asked how many users she might want, whether she liked dashboards, and how soon she could start. None of that is a readiness question, so we set it aside and ask our four.

Where do the facts live for a booked job? The owner says the job system. The dispatcher says he keeps a whiteboard because the system is wrong by the time a customer changes a time. The lead says she trusts the text the dispatcher sends her in the morning, because the whiteboard and the system have both let her down. That gives us three answers, and the questionnaire does not get to average them. The one place the facts live is still unresolved, and that unresolved state is the answer we write down.

Which step is written, and which step is habit? The written step is the quote, and the owner is proud of the quote template, as she should be. The habit is everything that happens after the customer says yes. Who tells the lead about the parking note, and who calls the customer when a part slips? Who decides that a job is done enough to invoice? The dispatcher says “we just do it.” The lead says she calls the owner when she is unsure, and that means the owner’s phone is the process. A step that lives in a phone is not a written step, so we mark it as habit.

Who can reject the output? This is the question that makes the room go quiet, and in a good way. If a tool drafted the morning text to the lead, who could say it is wrong and stop it? The dispatcher assumes he would notice, but the lead says she would notice on site, which is too late. The owner says she would want to approve messages at first. And then she admits she will not read them once the week gets loud. So nobody currently has a real reject button, and the people theme stays blank. We do not fill it in with optimism.

What will they stop doing if the answer is weak? The owner wants to stop repeating the same parking notes, and that is a real stop. The dispatcher wants to stop rebuilding the day from scratch each morning, which is also real. The lead wants to stop arriving without the part. If the questionnaire cannot connect a tool to one of those stops, the tool is only a hobby. So we write the stop next to the blank and let the blank carry a cost they can feel.

We do not end this example with a score, because themes are not a score. The framework is where those answers get words like absent or partial. And the assessment is where a band gets explained. What this conversation produced is a form with blanks the owner cannot talk her way out of. That makes it a finished questionnaire, and a finished sale would have been a different meeting and a worse one.

How we’d run it

First we throw out any question that starts with a product. If a vendor’s brochure could answer the sentence, it does not belong on our form. The owner’s seller asked about dashboards, but dashboards are a feature. And features are what you discuss after the themes have survived.

Second we ask the four questions in the words of the job, and we pick one job only. At this desk the job is a booked call, from the yes all the way to the lead packing the van. We do not run the questionnaire over the whole company, because a form about “the business” produces slogans. A form about one job produces places, names, and blanks.

Third we ask each person on their own before we let them repair each other’s answers. The dispatcher will defer to the owner if she is in the room from the start. And the lead will go quiet if the form starts to sound like a test of her. We want the first answer, which is the one they would give a new hire. Then we put the answers side by side and let the conflict stand.

Fourth we match the answers to Govern, Map, Measure. And Manage without making anyone say those words out loud. Who decides is Govern, and the steps of the booked job are Map. How the lead would know the morning text is wrong is Measure. And what the dispatcher does when it is wrong is Manage. If one of those is missing, the questionnaire is not done, no matter how many other boxes are ticked.

Fifth we decide whether a score is even useful yet. If the facts still live in three different places, a band will only decorate the confusion, so we stay with the questions. If the four themes have answers we can point at, we move to the check and then to the growth guide. That guide is about what to do once you know where you stand, and it is not a way to skip the form.

We keep the session short enough that the lead is not late to a job and the dispatcher is not dispatching from the hallway. A questionnaire filled in by the owner alone on a Sunday night will be complete and also made up. The person who does the step has to be in the room.

Common mistakes

The mistake we see most is starting with the tool. The seller’s form feels efficient because it ends in a quote. But that kind of efficiency is how you skip the record, the habit, and the person who can say no. You will pay for that skip every month.

Another mistake is treating a public checklist as a readiness rate. The AWS page, updated in September 2025, does not publish a percentage of businesses that are ready. And the OECD aid, current in 2026, is not a national adoption rate. If your notes say “we are ahead of most firms,” then you wrote a sentence your sources refused to write.

A third mistake is smoothing over disagreement. When the owner, the dispatcher, and the lead name different records, the honest questionnaire shows the split. A single tick marked “system of record: yes” is a fiction they agreed to so the meeting could end. Ending the meeting is not the goal. The blank is the goal.

We also see forms that ask about attitude, and “How excited is your team about AI?” is not a theme. Excitement does not tell you where the facts live. A doubtful lead who can reject a bad text is more ready than a cheerful room with no reject button. So we ask about permission and not about mood.

The last mistake is a questionnaire with no “stop doing” line. If nothing will be stopped, then nothing was decided, and you only gathered opinions. You did not assess the job.

What we’d do

We would use a one-page form with those four questions and a place to write two names when the answers conflict. Our opinion is that a long assessment questionnaire is often a way to feel serious without naming the whiteboard. The owner does not need more questions. She does need the dispatcher’s answer and the lead’s answer on the same sheet, in ink, where a dashboard question used to be.

If we were sitting with them next week, we would not bring a tool. We would bring the four questions again and see whether the facts now live in one place. If they do, we would score it, and if they do not, we would still not buy. A questionnaire that cannot produce a no is not an assessment, because it is only a path to a signature.