AI readiness assessment for founder-led companies: 12 questions about structure, not tools

Every AI readiness assessment I have been sent was written by someone who sells AI.

That is not a conspiracy. It is a catalogue. A vendor's assessment scores your data, infrastructure and skills because those are the three things an invoice can fix. Whatever you score, the answer was decided before you opened the form.

I run a different kind of AI readiness assessment with the founders I work with, and it asks almost nothing about tools. For a company of 20 to 200 people the constraint is not whether your data is clean enough. It is whether your org chart, your handoffs and your decision rights can absorb what AI does to them. This page is that assessment: 12 questions, four groups of three, no scoring formula.

Why vendor AI readiness assessments measure the wrong thing

Open a vendor's AI readiness assessment checklist and you'll find the same four columns: data, infrastructure, skills, governance.

Those columns describe a company's ability to install a system, not its ability to change shape once the system is installed. And the shape is what fails. MIT's Project NANDA studied enterprise generative-AI initiatives and found that around 95% of the organisations it studied showed no measurable return, and that the blocker was workflow integration, not model quality. The tools passed. The organisations didn't.

The long version of that mechanism is in why AI rollouts fail. The short version: you can score green on every vendor column and still buy licences and watch them rot... because what you need to be ready for is a restructure, and a restructure is not what they sell.

What this assessment measures instead

Four groups, in the order they break. Structure: whether your company still runs on a middle layer that exists to coordinate work AI now does. That is the Hourglass Collapse: once AI absorbs the bottom layer's routine work and the routing around it, the middle that coordinated it has nothing left to route. Work: whether anyone has sorted your workflows into Replace / Augment / Refuse, the 3-Zone Co-Pilot call. People: who already uses frontier tools, whether the five functions of ORBIT, the framework I use for this, have names attached, and whether juniors can still earn judgment. Decision: who can redraw the chart.

Answer them out loud, ideally with your ops lead in the room disagreeing with you.

The 12 questions

The not-ready answers are almost always longer.

Structure

1. Who routes information, and who decides?

Why it matters. Every org has people whose week is spent moving information between people who could talk directly. That routing is what the middle layer has left once AI absorbs the work it coordinated, so you need to know how much of your chart is made of it.

Ready sounds like: "These four people decide. These two route, and we know it, and we know what they'd do instead."

Not ready sounds like: "Everyone kind of does both."

2. Is there a narrow middle?

Why it matters. Draw the company as it actually runs, not as it's filed. If it pinches in the middle, a wide founder-and-senior layer above and a wide operator layer below, the pinch is the layer AI collapses.

Ready sounds like: "Yes, it's these roles, and we've started asking what each of them does that isn't coordination."

Not ready sounds like: "We're pretty flat." From a founder whose team leads spend Monday collecting status.

3. What fraction of attention goes to coordination rather than creation?

Why it matters. Per role, roughly how much of the week is chasing, summarising, aligning and reporting, versus producing something a customer would pay for? The coordination share is your Replace column waiting to be named.

Ready sounds like: "We estimated it per role last quarter. Ops was mostly coordination, delivery mostly creation."

Not ready sounds like: "Hard to say, it depends on the week."

Work

4. Has any workflow been sorted into Replace, Augment or Refuse?

Why it matters. The sort is the decision. Without it every workflow lands in the same undeclared middle: a human still in the loop, a tool bolted alongside, nobody able to say which owns the outcome.

Ready sounds like: "Ten workflows, sorted, with a name and a date on each. Three in Refuse, and we argued about two of them."

Not ready sounds like: "We use AI across the board."

5. Is "done" defined per workflow?

Why it matters. AI produces output that looks finished. If "done" lives in someone's head, that person becomes the quality gate, the queue backs up behind them, and they end up rubber-stamping. A written definition of done is what lets a Replace workflow run without a permanent human bottleneck.

Ready sounds like: "For the proposal draft, done means these five fields come from the CRM and the pricing matches the rate card. Anyone can check it."

Not ready sounds like: "Our senior people know it when they see it." They do. That's the problem.

6. Where does work stall?

Why it matters. Every company has a few places where work waits: an approval, a person who holds context, a handoff nobody owns. AI dissolves a routing stall and worsens a judgment stall, because it feeds the queue faster than the one person holding the decision can clear it. Know which before you automate anything upstream of it.

Ready sounds like: "Client work stalls at the account lead's review and at finance sign-off. The first is routing, the second is a real decision."

Not ready sounds like: "Things get stuck sometimes, it's usually a people thing."

People

7. Who already uses frontier tools on real work?

Why it matters. Microsoft and LinkedIn's 2024 Work Trend Index found 75% of knowledge workers using generative AI at work, and 78% of those bringing their own tools. Some of those people are in your company. The question is whether you can name them.

Ready sounds like: "Three names. One in ops, one in delivery, one in sales. We know what each uses it for and we've started copying."

Not ready sounds like: "We rolled out licences to everyone." Licences describe access, not use.

8. Who owns each of the five functions?

Why it matters. ORBIT names five functions every AI-augmented team has to have covered by a named human: Outcome Orchestrator (owns the outcome, routes human versus AI work), Operator (runs the systems day to day and catches silent failures), Innovator / Builder (designs the workflows and judges the output), AI Champion (carries the change at peer level), and Stakeholder Liaison / Translator (connects the team to users and stakeholders, both directions). An unowned function is where the rollout leaks.

Ready sounds like: "Here's the map. Two functions share a person, which we know is a risk, and the Champion is the person peers go to, not the loudest one."

Not ready sounds like: "I guess that's mostly me." A founder carrying four of five functions is the whole company's ceiling.

9. Is there a path for juniors to earn judgment?

Why it matters. The work juniors used to learn on is the work AI takes first. Remove it with no replacement and in a few years you'll have no bench of senior judgment. I have not seen a vendor assessment that asks this, because the answer doesn't produce a purchase.

Ready sounds like: "We've redesigned the first year. Juniors supervise agent output and own a quality gate by month three."

Not ready sounds like: "We'll cross that bridge." You're on it.

Decision

10. Who can redraw the chart?

Why it matters. If the only person with authority to change the chart also has the fullest calendar, the change waits for a gap that never comes. And if the people running the rollout sit in the layer it makes redundant, the chart will not be redrawn from inside.

Ready sounds like: "I can, and I've named someone outside the affected layer to propose it."

Not ready sounds like: "We'd take it to the leadership team." Which is drawn from the layer in question.

11. What was the last AI purchase, and who chose it?

Why it matters. If IT or engineering chose the last tool, you ran Tool before People, and nobody did anything wrong: choosing tools is their remit. The problem is the seat, not the person, and the seat tells you whether the company treats AI as a systems decision or as a decision about who does what.

Ready sounds like: "The founder and the ops lead chose it, after the sort, and we can tell you which Replace workflow it runs."

Not ready sounds like: "Engineering picked it."

12. What happens when the tool works and the company doesn't?

Why it matters. MIT's finding in one sentence: the tool worked, the person got faster, the company didn't. Whether you have a next move for that outcome is the sharpest readiness signal there is, because it tests whether anyone has understood that the tool was never the project.

Ready sounds like: "We'd treat it as a structure finding. Someone's job is now mostly waiting on a machine, so we'd redraw."

Not ready sounds like: "Then we'd try a different tool." That is the second rollout.

How to read your answers

There is no score. A percentage would tell you that you're "64% ready," and you'd have no idea what to do on Monday. This page has no score because I have no sourced basis for what a readiness percentage would predict.

Instead, a reading rule. Count the not-ready answers in each of the four groups. The group with the most is where the engagement starts... whether you run it yourself or bring someone in. If two groups tie, Decision first, because nothing moves if nobody can redraw the chart; then Structure, then People, then Work. That is the order the work runs in: people before process before tool.

What to do with a "not ready" in each group

Structure. Draw the real org, not the filed one: who routes, who decides, where work stalls. The output is a diagram that will not match the chart on the wall, and the gap between the two is the actual risk register for any AI rollout. The Hourglass Diagnosis, a free download on the resources page, finds the middle layer AI removes first; it is the worksheet for this group.

Work. Take the ten workflows that consume the most attention and sort each into Replace, Augment or Refuse, with a named person and a date on every call. Then write "done" for every Replace workflow.

People. Put the people already using frontier tools on real work in the room. Fill in the five ORBIT functions by name; an unowned one is the first hire or reassignment, ahead of any tool. And write down how a junior gets to senior when the junior work is automated.

Decision. Decide who can redraw the chart, and say it out loud. Move the AI purchasing decision from the systems seat to the who-does-what seat. And agree in advance what happens when the next tool works and the company doesn't, so the answer isn't "a different tool."

When to bring in outside help, and when not to

Don't, if your worst group is Work and the rest is mostly ready. A founder with a clear chart and a clear decision seat can run the zone sort with their ops lead in a fortnight.

Don't, if your company is under about 20 people. There isn't enough org to redesign yet. Sort the work, use the tools, and come back when there's a middle.

Do consider it when Structure or Decision is your worst group. Both involve redrawing the chart, and a chart is very hard to redraw from inside it, because the people with the most context on the coordination layer are the people the redraw affects. That is the structural reason a Fractional CAIO exists: senior enough to redraw the chart, external enough to say what the people inside it can't, and built to leave. Cost and scope are on that page.

The AI strategy essays carry the longer argument behind each question.

The binary

Take a vendor's AI readiness assessment, score well, buy the tool, and the post-mortem produces the usual reasons, none of which mention the chart. Or spend an afternoon on these 12 questions with your ops lead, find your worst group, and start there, with or without me, before any purchase.

The tools in both versions are identical. Only one of them was ever going to change the company.

FAQ

What is an AI readiness assessment?

An AI readiness assessment is a structured set of questions that tells a company whether it can absorb AI before it spends on tools. Vendor versions score data, infrastructure and skills. For a 20 to 200-person company the constraint is structural, so this assessment asks whether the org chart, the handoffs and the decision rights can change shape when AI removes the coordination layer.

What questions should an AI readiness assessment ask?

Twelve, in four groups. Structure: who routes and who decides, is there a narrow middle, how much attention goes to coordination. Work: have workflows been sorted into Replace, Augment or Refuse, is "done" defined, where does work stall. People: who uses frontier tools already, who owns the five ORBIT functions, how juniors earn judgment. Decision: who can redraw the chart, who chose the last purchase, what happens when the tool works and the company doesn't.

How do I know if my company is ready for AI?

Answer the 12 questions out loud with one other person in the room and listen to the sound of your answers. Ready answers are short and name people, dates and workflows. Not-ready answers are long, hedge, and use words like "everyone" and "it depends." Count the not-ready answers per group. The group with the most is where your work starts, before any purchase.

Do small businesses need an AI readiness assessment?

Below about 20 people, no. There isn't enough org to redesign; sort the work, use the tools, and revisit when a middle layer appears. From 20 to 200, yes, and the structural version matters more than the vendor version, because at that size the coordination layer is often informal, held in two or three heads, and invisible to any checklist that only measures data and infrastructure.

What is the difference between AI readiness and AI maturity?

Maturity measures how far along a company already is with AI: how many tools, how much usage, how developed the governance. Readiness measures whether the company can change shape to absorb the next step. A company can be high-maturity and not ready: plenty of licences, plenty of usage, and an org chart nobody is allowed to redraw. This assessment measures readiness in that sense.

Read this next: The Co-Pilot Zone: Replace, Augment, or Refuse... the sort behind questions 4 to 6, and the first thing to run after the assessment.

Sources

1. MIT Project NANDA: The GenAI Divide: State of AI in Business 2025 (July 2025). 300+ initiative reviews, 52 interviews, 153 survey responses; finding that ~95% of organizations studied saw no measurable P&L return, with the blocker identified as workflow integration rather than model quality. Widely mirrored PDF: https://cloudelligent.com/wp-content/uploads/2026/02/v0.1_State_of_AI_in_Business_2025_Report.pdf
2. Microsoft & LinkedIn: 2024 Work Trend Index Annual Report (8 May 2024). Survey of 31,000 people across 31 countries. 75% of knowledge workers using generative AI at work; 78% of those users bringing their own tools. https://news.microsoft.com/source/2024/05/08/microsoft-and-linkedin-release-the-2024-work-trend-index-on-the-state-of-ai-at-work/

Frameworks referenced are my own: the Hourglass Collapse, the ORBIT Framework, and the 3-Zone Co-Pilot (Replace / Augment / Refuse).

---

The Hourglass Diagnosis: the free ten-statement worksheet for the Structure group.

It scores the layer between leadership and the people doing the work, and tells you which of the three readings you got.

Download the Hourglass Diagnosis →

Or get the frameworks behind this page by email: subscribe

Or if you already know this is your situation: book a 30-minute scoping call

---