The AI Readiness Checklist: Is Your Company Actually Ready?

There's a question I get asked at least once a week now, usually phrased with a slightly nervous laugh: "So... should we be doing something with AI?" It's the wrong question, but it's an honest one. The right question — the one almost nobody asks first — is quieter and far more useful: is this company actually ready to get value out of AI, or would we just be lighting money on fire to look modern?

I've spent years now sitting across the table from owners and managers of mid-sized companies — manufacturers, wholesalers, professional firms, regional service businesses with fifty to three hundred people. The ones who succeed with AI almost never start with the technology. They start by being honest about the state of their own house: their data, their processes, their people, and how decisions actually get made when nobody's watching. The technology is the easy part. Readiness is the part that decides whether you get a result or a regret.

So this is the checklist I run, more or less, before I'd recommend a company spend a single euro on building anything. It's not a maturity model with twelve dimensions and a colourful radar chart. It's a set of plain questions you can answer over coffee, grouped into five things that genuinely predict whether AI will work for you — or quietly embarrass you in a steering meeting six months from now.

Why "readiness" matters more than the model

Here's the uncomfortable truth the vendors won't lead with: the AI model is rarely what makes a project fail. The models are astonishingly good now, and they get better every few months whether you do anything or not. What kills projects is everything around the model — the messy data it's fed, the process it's bolted onto, the people who were never asked, the unclear owner, the fuzzy definition of success.

I've watched companies spend serious money on a clever pilot that worked beautifully in the demo and then died on contact with reality, because the data it needed lived in three incompatible systems and one person's head. The pilot wasn't wrong. The company just wasn't ready for it. Readiness isn't a gate that keeps you out of AI — it's the thing that determines whether your investment compounds or evaporates.

The good news is that readiness is fixable, and most of the fixes are unglamorous and cheap. You don't need a data lake or a chief AI officer. You need to know honestly where you stand on five fronts, and to close the one or two gaps that would otherwise sink the first project. Let's go through them.

Pillar 1: Your data — is it good enough to build on?

AI is only as good as what you feed it, and this is where mid-sized companies are most often surprised. Not because their data is terrible — usually it's fine for running the business — but because "fine for humans" and "fine for a machine" are different bars. A person can read a slightly inconsistent customer record and know that "Müller GmbH" and "Mueller G.m.b.H." are the same firm. A system needs you to have decided that already.

You don't need perfect data. Nobody has perfect data, and waiting for it is just procrastination in a suit. What you need is data that's accessible, reasonably consistent, and connected to the process you want to improve. Ask yourself the plain questions:

• Can you actually get at the data you'd need, or is it trapped in a system only one vendor can export from?
• Is the same thing recorded the same way most of the time — customers, products, statuses, dates?
• Do your core systems know about each other, or does someone retype between them every day?
• Is there enough history to learn from, where the task needs history at all?
• Would you be comfortable if a regulator asked where a given number came from?

One more thing people underestimate: data ownership. If a critical chunk of what you'd feed the model lives inside a third-party tool whose contract forbids exporting it, that's not a technical problem you can engineer around — it's a commercial one you have to solve first. Better to find that out now, on paper, than three months into a build.

Pillar 2: Your processes — are they stable enough to automate?

There's an old line in our world: if you automate a mess, you get a faster mess. It's a cliché because it's relentlessly true. AI dropped onto a chaotic process doesn't fix the chaos — it scales it, and now you're confused at speed. Before you can put intelligence on top of a process, the process has to be stable enough that you could describe it to a new hire on their first day.

That doesn't mean it has to be elegant or fully documented. It means it has to be real and repeatable. The test is simple: can you draw it? If you sat three people who do the job in a room and asked them to sketch the steps, would they draw roughly the same picture — or three different ones, each insisting theirs is how it's "actually" done?

Rules versus judgement

It also helps to know what kind of process you're looking at. If it follows fixed rules — move this data there, send that reminder, flag anything over a threshold — you may not need AI at all; plain automation is cheaper and more reliable. AI earns its keep where the work is genuinely messy and language- or judgement-shaped: reading a free-text email and pulling out the order, drafting a first-pass reply, sorting documents nobody wants to file. Being honest about which one you have saves you from buying a sledgehammer for a thumbtack.

Pillar 3: Your people — will they actually use it?

This is the pillar that gets the least attention and breaks the most projects. You can have clean data and a tidy process and still end up with a tool nobody touches, because the people whose work it changed were treated as an afterthought. A model that's ignored has an ROI of exactly zero, no matter how impressive it looked in the pitch.

Readiness here is partly about skills, but mostly about trust and involvement. Were the people who do the work asked what slows them down, or were they handed a solution to a problem they didn't agree they had? Does someone in the building feel a sense of ownership over making this succeed, or is it "the consultant's project"? Is there at least one curious person on the team who'll champion it rather than quietly route around it?

You don't need data scientists on staff. For most mid-sized companies, the right partner brings the technical depth. What you can't outsource is the willingness to change how a few daily tasks get done — and that willingness is built by involving people early, not by announcing the new system at an all-hands once it's already bought.

Pillar 4: Strategy and budget — is there a real owner and a real number?

AI initiatives that float free of the business, with no clear owner and no defined success, tend to drift until someone quietly defunds them. Readiness on this front is unglamorous but decisive: a named person who owns the outcome, a specific problem worth solving, and a budget that matches the ambition — including the part everyone forgets, which is keeping the thing running after launch.

You should be able to finish this sentence before you start: "This project succeeds if, within three months, [specific measurable thing] happens." "We become more innovative" is not a finish line; it's a press release. "Our team stops spending six hours a week retyping orders" is a finish line. The narrower and more measurable, the better — and the easier it is to know whether you actually won.

On budget specifically: the build is usually the smaller half. A model needs monitoring, the occasional retune, and an owner who notices when reality drifts from what it was trained on. Companies that budget only for the build are setting up a tool that works brilliantly for a quarter and then slowly rots because no one was paid to look after it. Plan for the boring afterlife of the project, not just its exciting birth.

Pillar 5: Risk and governance — the part you can't skip in Europe

For a mid-sized European company, this pillar isn't optional, and pretending otherwise is how you end up with a clever tool you legally can't deploy. You don't need a compliance department, but you do need to have asked the basic questions before you build, not after. Mostly it comes down to knowing what data the system touches, where it's processed, and who's accountable when it gets something wrong.

The questions are less scary than they sound. Does the tool process personal data, and if so, on whose servers and under what agreement? If the AI suggests something — a price, a reply, a decision — is a human reviewing it where it matters, or is it acting alone? Can you explain, in plain language, how it reached a given output if a customer or auditor asks? Have you thought about what happens on the day it's confidently wrong, because that day will come?

Putting it together: scoring yourself honestly

You don't need a spreadsheet for this, though you can build one. Go through the five pillars and rate each one red, amber or green — not aspirationally, but as it really is today. The point isn't to score straight green; almost no company does, and you don't need to. The point is to find the one or two reds that would sink a first project, and deal with those before you build.

1. 1
   Rate each pillar red, amber, green

   Data, processes, people, strategy, governance. Be honest — an optimistic self-assessment just moves the disappointment to later.
2. 2
   Find your blocking reds

   A red on the exact pillar your first project depends on is a stop sign. A red somewhere unrelated can wait. Context matters more than the total.
3. 3
   Fix the cheap gaps first

   Most readiness gaps are closed with unglamorous, low-cost work — making data reachable, writing the process down, getting the team involved early.
4. 4
   Pick the smallest real project

   Choose one stable, frequent, well-owned process where you're green enough to win. A small finished win funds the next one — in trust, not just money.
5. 5
   Then, and only then, build

   With reds cleared on the pillars that matter and a measurable goal written down, you're ready. Not perfect — ready. Those are different, and only one of them ships.

Notice that most of this work has nothing to do with AI. Reachable data, a process you can draw, a team that's bought in, a clear owner, a human in the loop — these make your company better even if you never deploy a single model. That's the quiet bonus of taking readiness seriously: the preparation is valuable on its own. AI just gives you a reason to finally do it.

Common questions