What makes a good AI readiness questionnaire?
Most AI readiness questionnaires ask people to rate themselves. "Are your processes documented?" gets a yes, because a folder exists somewhere. A person who knows the answers also knows the result, and the questionnaire teaches nothing. Our first draft had the same fault. We rewrote it on four rules.
- Ask about events and facts. The last unusual case. The last AI experiment that worked. How the last mistake came to light. People remember events, and they describe them the same way whoever asks.
- Options describe situations, not grades. "We would feel it" and "We track time per task" both sound like normal answers. For the reading they are different findings.
- "I do not know" is an answer. A leader who does not know how many AI tools are in use has given us the finding.
- One answer feeds several findings, and one finding rests on two or three answers. The count of tools, who decides about them and how sure the leader is about data together give the shadow AI finding. An answer that another answer refutes goes back as a question: a company cannot both depend on one vendor and switch vendors within days.
The same discipline runs through our interviews: say the answer back in the person's own words before it becomes a finding. Listening builds trust. It is not verification.
The AI readiness assessment framework behind the questions
The questions sit on three parts.
One scale. The autonomy scale we use across this site: Manual, Assisted, Collaborative, Supervised, Autonomous. A stage describes who carries the work and who checks it today. It is not a score of how modern the company is.
Four checks. Whether the next stage can hold depends on four conditions: information that is structured, current and findable; processes that are defined and repeatable; systems AI can reach; agreement on what AI should accomplish. The readiness assessment guide explains each check. On top of them we read quality and control: what happens to AI output, how mistakes surface, and what a system may never do on its own.
Six areas. After three facts about the company (industry, size, your role), 24 situational questions walk through six areas. One open question closes the list: what you would change, within your control.
| Area | Questions | What it reads |
|---|---|---|
| How AI got into the company | 4 | The adoption pattern, visibility of tools, shadow AI |
| What AI does in the work | 7 | The stage, verification, trust, effect, vendor dependence, pilots, the plan |
| Knowledge and process | 4 | Where the rules live, single points of failure, consistency, exceptions |
| Systems and alignment | 4 | Handoffs between systems, access, a shared goal, the last attempt |
| Quality and control | 3 | How mistakes surface, what stays with people, whether a gain is measured |
| What stands in the way | 2 | Past blockers, who would carry the change |
How did AI get into your company? Four questions
1. How did AI get into the company? Several answers are allowed, and most companies tick two or three. The pattern decides how AI tends to fail.
- People picked their own tools. The real wins start here. Later the prompts stay personal and company information goes into tools nobody chose.
- Vendors added AI to tools you already use. The work did not change, and the useful part, the prompts and settings, belongs to the vendor.
- Leadership launched an initiative. Without one written goal and a team that carries it, it spreads into side projects.
- One leader is carrying it. The fastest way to get results, and the most fragile way to keep them.
- An outside firm ran a pilot. Unless the result became a process, knowledge and a tool the team owns, the next attempt starts from zero.
- It has not arrived yet. A clean start.
2. How many AI tools are in use, counting the ones people picked themselves? In one organization we assessed, departments had each built their own assistant on a pile of uploaded documents, with no shared knowledge model between them.
3. Who decides which AI tools people may use with company information? "Nobody has decided" and "Each person decides" are the same answer in different words. In a ten-person firm we interviewed, the accountant had her own paid subscription, and the owner did not know whether it touched the books. Trust stood in for a policy.
4. How sure are you that nobody pastes customer data, contracts or financials into a public AI tool? The honest middle answer is "We have a rule, and I am not sure they follow it." With the two questions above it gives the shadow AI finding. Shadow AI has its own guide.
What does AI do in the work today? Seven questions
5. Which of these happen in your company today? This one question places the stage, because a stage is read from what already runs: steps without a person touching them mean Supervised, a shared assistant or skill means Collaborative, a chat tool or a built-in feature means Assisted.
6. What usually happens to a draft or an answer that AI produces? "It is used as it is" means there is no verification gate, and the first visible mistake will be found by a customer or a manager. "Most of it is rewritten" and "People avoid AI for anything that matters" mean AI has stayed a personal tool.
7. What is the biggest reason people hold back? Made-up answers, a possible data leak, not knowing how to get a good result, no time, work that is too specific. Each is a different gap, and a rule and a check remove the first two.
8. What has AI changed so far? The most common honest answer is "Some tasks take less time, but we have not measured it." Without a baseline the gain never appears in the numbers, and the next investment rests on impressions.
9. Which of these is true about the AI tools you rely on most? Prompts and settings that live inside the tool, or one vendor holding most of the setup, mean the durable value sits in definitions that are not yours. One buyer we worked with refused to build connectors into chat tools for that reason: user habits would lock the company in. Assistants built around your own documents are a strength.
10. What happened to the last AI experiment that worked? A pilot that stayed with the person who ran it is the warning sign our blog named: the project relies on one enthusiastic person. Rolled out, with an owner, it is a strength.
11. Is there a written plan for AI, and who owns it? Without a written plan with an owner and a budget, the strategy comes first, as the last section explains.
Knowledge and process: four questions that reveal what AI would copy
Two reasons explain most pilots that worked in a demo and failed in production: weakly defined processes and missing specific knowledge. Four questions find them.
12. When someone needs a rule, a price or a past decision, where do they get it? A colleague, a chat history or a shared folder people hope is current are the same finding. Behind the folder sits a known spiral: documents are written, not found, not updated, and people stop trusting what they find.
13. If a key person were away for a month, what would happen to their work? "Part of the work would stop" is a single point of failure in plain words. We have seen a workshop where quality control, stock, machine setup and delivery all ran through one foreman.
14. Two people handle the same case. How similar is the result? "It depends on who handled it" means the work follows habit, and people fill the gaps from experience without noticing. AI fills the same gaps by guessing. Why written procedures alone do not fix this is a guide of its own.
15. Think of the last unusual case. How was it resolved? "It went up to the manager" is a wall for automation: an agent can only hand over what a rule tells it to hand over. Knowledge that lives only in people has its own guide too.
Systems and alignment: four questions
16. How does information move between the steps of the work? Where work moves by email, retyping and exports, AI cannot act where the work happens, and automating one step only moves the bottleneck to the next handoff.
17. Has anyone ever pulled data out of your main systems other than through the screen? "Nobody has tried" leaves the connectivity question open. It is also where companies find systems they forgot to count: one inventory compiled from memory missed the telephony platform and the messenger, the two systems through which the work actually entered.
18. If you asked three leaders what AI should do for the company, how many answers would you get? Without one answer, initiatives drift into side projects and lose support as fast as they gained it.
19. Has the company been through an improvement, documentation or automation attempt before? What is left of it? "It faded, and nothing is left" is a finding about the present. Whatever stopped the last attempt is still there, and the next one meets it unless it is named first.
Quality, control and capacity: five questions
20. Think of the last time a piece of work went out wrong. How did it come to light? A customer, a manager, a check before it went out, or the numbers. The first two mean control depends on attention. At the Supervised stage a mistake travels on without a person, and this answer decides how much of that stage the company can carry.
21. Which of these would you never let a system do on its own? Pay or move money, commit to a customer, sign a contract, anything that touches safety, decide about a person. "Not decided yet" is the gap: then either everything waits for approval, or nothing is protected. Human on the loop describes the control pattern that follows.
22. If your core work got 30 percent faster next quarter, how would you know? "We would feel it" and "We would not know" are the same finding. In one engineering organization time was tracked only for cost allocation, which left nothing to measure an AI gain against.
23. What has stopped or slowed change in the company before? Budget, systems that cannot change, data quality, policy, people who resisted, the person who left. The report sizes the first step to the constraint named.
24. If you decided to change how the company works with AI, who would do the work of changing it? You in your spare time, a manager part-time, a team, or an outside partner. A roadmap moves at the pace of what the company can actually build.
The open question asks what you would change, within your control. In our interviews with owners of small firms the answer has been a person to carry the work or a system of record. An AI use case has not come up once.
What risks can an AI readiness assessment reveal?
A risk costs the business today whether AI goes further or not. A gap blocks the next stage of autonomy. Each is read from two or more answers and gets a severity from 0 to 3. The report shows those at 2 or more, at most five, after what already works: a rule people follow, an experiment with an owner, errors caught before they go out.
| Finding | What it leads to | The answers that point to it |
|---|---|---|
| Shadow AI (risk) | Company information goes into tools nobody registered | Unknown tool count; nobody decides; no data rule |
| The useful part belongs to the vendor (risk) | A change of price or features becomes your problem | Prompts live inside the tool; one vendor holds the setup |
| No verification gate (risk) | The first mistake is found by a customer or a manager | Output used as it is; mistakes noticed by customers |
| AI rests on one person (risk) | Everything holds while that person stays | One leader drives it; the last experiment stayed personal |
| No baseline (risk) | The gain cannot be shown, and the next investment is a guess | Effects not measured; a faster quarter would be felt |
| Knowledge that lives in people (gap) | A key absence stops part of the work | Rules come from a colleague; a month away stops work |
| Work that varies by person (gap) | AI would copy the inconsistency | The result depends on who handled it |
| No shared goal for AI (gap) | Initiatives drift into side projects | Three leaders, three answers; no written plan |
Seven more follow the same logic: tool sprawl, AI that stays a personal tool, exceptions only a manager can settle, work that moves between systems by hand, a faded attempt, no line between what AI may and may not do, no capacity to carry the change.
How is the result read? A stage and its state
A maturity score gives one number for the company, and one number says little about what to do next. The checklist gives a stage, and within the stage an early or later state read from the four checks. Early Assisted: AI arrived through people or a vendor feature, and nobody has decided which tools may see company information. Later Assisted: a rule or a shared experiment exists, and the value still depends on who uses it. Early Collaborative: shared assistants carry work in pockets, often inside one vendor's tool. Later Collaborative: the result is the same whoever handles it, and the rules sit in a system. Early Supervised: steps run without a person and control has not caught up, which is where an unchecked mistake costs the most. Later Supervised: the control holds, as far as a checklist can place a company.
Autonomous stays on the scale as the finish line. We never assign it from a checklist or sell it as a target. It follows from Supervised work that has held its standard long enough. Most companies sit between Manual and Collaborative on this scale, and a company can be uneven: Supervised where AI already runs, a chat tool everywhere else.
After the stage come the risks, the four checks and two or three next steps for the state: foundations first, each one something a person in the company can start this week. Then the offer we would start with, at most two, with the reasons in plain words.
AI readiness checklist or interview-led assessment: which do you need?
Both, in that order. The checklist is the quick first look: the whole company, 27 questions, seven minutes, a stage, the risks and the first steps. The interview-led assessment deepens the same picture with what no checklist can see: who holds each function, where the knowledge sits and the limits leadership sets. It is the readiness part of the current-state analysis inside an AI Organization Strategy, and the strategy then adds the organization model, the AI-native blueprint and the phased roadmap.
A small firm whose owner holds half the functions gets as much from the checklist as a business unit of a large company; the wording changes with the size you give. If you are comparing AI readiness assessment services or an AI readiness assessment tool, two questions sort them quickly. Does it ask what happened, or how you rate yourself? Does it end with what to prepare first, or with a score? A score alone starts no work.
How EIC runs the AI readiness checklist
The AI readiness checklist is free and needs no account. It asks the 27 questions above, one screen per area, in seven minutes. If you give us your website, we read what is public there, and the questions and the report speak about your own work and systems. Nothing from the site reaches the report unless you confirm it.
The report opens as soon as you finish, and a copy with a PDF arrives by email. Your answers reach only the engineer who prepares your call.
The next step is a free readiness call. We go through the report with you, settle the points the answers could not, and agree where to start. You talk with an engineer, not a salesperson. From there the path is the one the report named. Without a written plan for AI, the interview-led assessment and the strategy come first: that is the AI strategy consulting in AI Organization Strategy. Otherwise the first foundation: AI Process Reengineering for work that varies by person, AI Knowledge Base for knowledge that lives in people. Where AI output already reaches the work unchecked, AI Validation & Control comes beside it.
Short answers on what an AI readiness assessment is, how use cases are prioritized and what the first phase of a roadmap holds are in the FAQ on the AI Organization Strategy page.
