The AI Readiness Assessment: 12 Questions to Ask Before You Spend a Dollar
Twelve questions I'd want answered before a team spends anything on AI, what a good answer sounds like, the red flags, and what to do with your score.
By some estimates, more than 80% of AI projects fail, twice the rate of IT projects that don't involve AI (RAND). In my experience, very few of them fail because the AI wasn't clever enough. They fail because the team bought something before they could answer a few plain questions.
I have a five-question version for a quick gut check. This is the longer one: the twelve questions I work through with a founder before we build anything. Each has why it matters, what "ready" looks like, and the red flag that tells you to stop.
Answer them for one specific piece of work, not for the company. "Is our company ready for AI?" has no useful answer. "Is our weekly client report ready for AI?" does.
The work
1. Can you name the one piece of work you'd start with?
Not "marketing" or "operations". One job, with a start and a finish, that happens on a schedule.
Teams that start with a category end up with a tool and no use for it. Teams that start with a job end up with something running.
Ready looks like: "The weekly status email to each client" or "turning sales call notes into the CRM update". It happens at least weekly, someone does it well, and most people would happily never do it again.
Red flag: the answer is a list of ten things, or "we want to see what AI can do for us first."
2. Could a new hire do it from what's written down?
This is the question that decides most projects. AI has no memory of how your team does things and no sense of what you meant. It does what's on the page.
If the process lives in one person's head, the AI will produce something almost right, which is the same as useless. I wrote more about this in AI can only do work that's been defined.
Ready looks like: a page with the trigger, the inputs, the steps in order and the judgement calls ("if the numbers dropped, say so in the first line"). A colleague who's never done it has followed the page and got an acceptable result.
Red flag: "Ask Sarah, she knows how it works." Or a document from two years ago that nobody follows.
3. Do you have examples of the work done well?
Rules tell the AI what to do. Examples show it what good looks like, and they settle arguments about tone, length and detail faster than any instruction.
Ready looks like: three recent, real examples of finished work you were happy with, plus one you weren't, with a note on why.
Red flag: nobody can agree on what a good one looks like. That's a disagreement to settle among the team before any AI gets involved.
The data
4. Can the AI reach the information the person uses?
Whoever does the job today opens a spreadsheet, a CRM, an inbox and probably a chat thread. The AI needs the same. If it can't reach an input, someone has to copy and paste it in, and that someone will stop doing it within a month.
Ready looks like: every input for this one job lives in a tool with a way to connect to it, and you know who has the access to set that up.
Red flag: a key input only exists in someone's personal notes, on paper, or in a system nobody has admin rights to.
5. When two sources disagree, do you know which one wins?
The CRM says one thing, the spreadsheet says another. A person resolves that without thinking. The AI will pick one, confidently, and won't tell you it had to choose.
Gartner predicts that through 2026, organisations will abandon 60% of AI projects that aren't supported by AI-ready data (Gartner). For a small team, AI-ready rarely means a data warehouse. It means knowing which source is true.
Ready looks like: a written rule for each input: "the CRM is right about contacts, the sheet is right about budgets."
Red flag: people fix the same data by hand every week and nobody has decided which version is correct.
6. Is there anything in this work that shouldn't go into an AI tool?
Client contracts, health information, salaries, anything under an NDA. Most AI tools have business plans that handle data properly, but you need to know what's in the work before you choose.
Ready looks like: you've checked what the job touches, you know what your contracts allow, and you're using a business account rather than people's personal ones.
Red flag: half the team is already pasting client data into free personal accounts and nobody has asked about it.
The people
7. Will the person who does it best give you a few hours?
The person who does the job best is the one who knows the judgement calls. Their time is what turns a vague process into a defined one. It's usually hours per process, not weeks: one session watching them do it on real work, and one test where someone else follows the page.
Ready looks like: that person is named, willing, and their manager has cleared the time.
Red flag: "They're too busy, just build it and show them." You'll build the wrong thing.
8. Who owns it once it's running?
AI processes need small, regular care. The work changes, a tool updates, a new client has an odd case. Somebody has to notice and adjust.
Ready looks like: one named owner, and at least one other person who can read and change the rules. Ideally the rules are in plain language so the owner doesn't need to code.
Red flag: the plan is for the consultant, or the one technical person, to own it forever. When they're busy or gone, it quietly stops.
9. Does the rest of the team want this off their plate?
A process the team didn't ask for gets worked around. A process that removes a job they dislike gets used. At a 25-person marketing agency I worked with, the team ended up writing most of their shared processes themselves, because they knew the work better than anyone.
If some of the team are wary, that's normal, and it's worth taking seriously rather than steamrolling. Part of my work is sitting with those people as a mindset coach, hearing out what worries them, and helping them see what the change gives them. People who've been heard are far more likely to give a new process a fair try.
Ready looks like: the people who'd use it have said, unprompted, that they'd like this job to go away, and they've seen what's being built.
Red flag: the first time the team hears about it is the launch email.
The tools
10. Do you know what you already pay for, and who uses it?
Most teams already pay for an AI tool, and often only a few people use it. A new tool rarely fixes that. A clear process usually runs fine on ChatGPT, Claude, Gemini or Copilot.
Ready looks like: a list of the AI and automation tools you pay for, who has seats, and which ones connect to the systems from question 4. You're willing to start there.
Red flag: the plan begins with buying a new platform, and nobody can say why the current one isn't enough.
Trust and results
11. What's the worst thing it could do, and who checks before that happens?
It will be wrong sometimes. The question is whether a wrong output reaches a customer before a person sees it. One bad email is enough to get a project switched off for good, along with everything useful it was doing.
My rule of thumb: let AI act on its own only where a mistake is cheap, easy to notice and easy to take back. In a sales pipeline I built for a software startup, the AI drafts every follow-up and sends none of them; the founder reads and sends each one herself. More in Drafts, never sends.
Ready looks like: you've written down what the AI must never do, and a named person reviews anything customer-facing, at least for the first month.
Red flag: "It'll be fine, we'll fix it if something goes wrong." Nobody will know something went wrong until a customer tells you.
12. How will you know in a month whether it's working?
Without a before, there's no after. This doesn't need a dashboard. It needs a clear statement of what's true today.
Ready looks like: a rough baseline you trust, such as how long the job takes now, how often it slips, or how many go out late. And a date to check: "in four weeks, is the team using it, and is the output good enough that they send it with light edits?"
Red flag: success is defined as "people like it" or "we're using AI now."
Scoring your answers
Count a yes only when you could show someone the evidence: the page, the owner's name, the examples. A "sort of" is a no.
| Yeses | What it means | What to do |
|---|---|---|
| 10 to 12 | Ready for this piece of work | Build it. Start with one process, run it for a month with a person reviewing the output, then pick the next. |
| 6 to 9 | Close, with specific gaps | Fix the gaps for this one process before spending on tools. It's usually questions 2, 5 and 8, and it's weeks of effort, not months. |
| 0 to 5 | Not yet, and that's normal | Don't buy anything. Write the process down (question 2) and name an owner (question 8). Those two are worth doing even if you never automate anything. |
Two rules override the score. A no on question 2 means you start there, whatever else you scored, because nothing else works without it. A no on question 11 means you don't launch anything customer-facing until it's a yes.
Run the assessment again for the next piece of work. Readiness isn't a company-wide grade; it's a property of each job.
If you want a second opinion on your answers, tell me about the work and I'll give you my honest read.
Common questions
What is an AI readiness assessment?
It's a check of whether a specific piece of work, and the team around it, can actually run on AI before you spend money. A useful one covers the work, the data, the people, the tools you already have, and how mistakes get caught.
How long does an AI readiness assessment take?
For one piece of work, an hour or two to answer honestly. Fixing the gaps it finds, usually writing the process down and naming an owner, takes days to a few weeks.
What is the most common reason teams aren't ready for AI?
The work was never written down clearly enough for someone new to follow it. AI can only do work that's been defined.
Do we need clean data across the company before using AI?
No. You need the data for the one process you're starting with to be reachable, and a rule for which source wins when two disagree.
Should we buy an AI tool before doing the assessment?
Usually not. Most teams already pay for a capable AI tool. Do the assessment first, start with what you have, and buy only when a defined process needs something your current tools can't do.