The AI readiness assessments that rank on Google were built for companies with an IT department. Search the term and the tools that come back are from Microsoft, Cisco, Avanade and Eide Bailly, and they’re good tools. Microsoft’s AI Readiness Wizard asks ten questions across five value drivers, and Google’s AI Overview sums up the whole category as five pillars: strategy, data, infrastructure, governance, and talent and culture. If you run a big company with a transformation budget, start there.
If you run a shop with an accounting package, a CRM, a quoting tool and a spreadsheet somebody updates every Friday, those tools will score you on things you don’t have a department for. And they’ll skip the one question that decides whether AI does anything useful in your building this year.
An AI readiness assessment for a small manufacturer should answer one question: could AI answer the questions you ask about your business today, from the systems you already own, without anyone downloading, retyping, or asking the one person who knows? The 7-Question AI Test on this page measures that in about 15 minutes, with a pencil, and gives you a number out of seven.
The whole test is below, open, no email required. It won’t tell you to buy anything, and it doesn’t need access to a single system. The rest of this page explains what your number means and what to do about it.
What does an AI readiness assessment check, and what do the big ones skip?
Most AI readiness assessments score your organization across a handful of broad dimensions, usually strategy, data, technology, governance and people, and place you on a maturity scale. Google’s AI Overview defines it as “a structured evaluation that measures an organization’s preparedness to adopt, scale, and sustain artificial intelligence initiatives.” That’s a fair definition for the companies those tools were built for.
To be fair to the enterprise tools, “data foundation” is one of the five pillars they score. But it sits next to culture, governance and cloud infrastructure as one score among five. In a small manufacturer, whether your systems can hand over an answer is the thing every other pillar depends on, and the fastest way to find out is to ask them a question and watch what has to happen before the answer shows up.
That’s what the 7-Question AI Test does. It skips strategy and culture, because in a shop where the owner still makes the growth decisions, strategy is a conversation with yourself. It goes straight to whether your systems could hand an AI an answer.
The 7-Question AI Test: could AI answer these seven questions about your shop today?
The test is seven questions an owner would ask out loud on a normal Tuesday. For each one you write down who would have to get the answer, where it lives, and how long it takes. At the bottom is one number: how many of the seven AI could answer today.
All seven are about the same three things: pipeline, revenue and margin. That’s deliberate. They’re the questions that decide whether you’re making money, and they’re the ones a connected set of business systems can answer. We left out questions about the shop floor, such as which jobs keep coming back for rework. They matter, but they’re about machines and process, and they need a different kind of help than the one this test measures.
Use these, or cross any of them out and write your own. They should be your questions, about your business, and the test only works if you answer them honestly rather than the way you’d like them to be.
The seven questions
- Which customers haven’t reordered in 90 days?
- What did we actually make on the last job for our biggest customer?
- Which open quotes are older than 30 days, and who owns each one?
- What’s our on-time delivery this month?
- How much work is booked for next quarter?
- Where did our last five new customers come from?
- What will cash look like in 60 days?
What to write next to each question
For each question, jot down three short answers. “I don’t know” is a legitimate answer, and it’s usually the most useful one on the page.
- Who has to get it? Anyone in the office, or one particular person?
- Where does the answer live? Name the system or systems. If the answer is “two systems and a spreadsheet,” write that.
- How long would it take today? Two minutes, an afternoon, or a few days?
How to score it
A question counts as AI-ready only if anyone could get the answer without retyping or exporting anything, from systems that already hold it. Everything else is a No, however quickly the person who knows could get it for you.
That rule is stricter than it looks, and it’s strict on purpose. If the answer to “which customers haven’t reordered” exists, but only after your office manager exports two reports and lines them up in a spreadsheet, then AI can’t answer it either. AI doesn’t have her login, and it doesn’t know which two reports she’d pull or how she lines them up. Whatever a person has to do by hand before the answer exists, an AI would have to do too, and right now nobody has built it a way to.
Count the questions that pass. Write it in the box:
AI could answer ___ of 7 today.
What your score means
- 5 to 7: AI will earn its keep here now. Start with the questions you ask most often.
- 3 to 4: The answers exist, but they get retyped or run through one person. Fix the most-asked question first.
- 0 to 2: It’s not an AI problem yet. Nothing can read what your systems can’t hand over, and that’s where the work is.
Want someone to go through it with you?
If you finish it and want someone to go through your answers with you, that’s what the 25-minute call is for. Bring your answers. We’ll skip the part where I ask how things work today, because you’ll have already written it down.
Why does AI give generic answers about your business?
AI gives generic answers about your business because nothing connects it to the systems that hold your business, so every session starts from a blank slate and anything good you get out of it had to be hand-fed. It happens with every model. Any model, writing about a customer relationship it can’t see, is going to produce something generic, and switching vendors doesn’t change that.
The way it usually comes up is somebody telling me they already tried. They bought the subscription, had it write a few customer emails, and what came back was close. Close, but not close enough to send. So it sits there, the seats renew, and the honest summary is that it was a neat demo and nothing changed. We wrote about that pattern in We Pay for AI and Nobody Uses It, because it’s where a lot of AI subscriptions end up.
Manufacturers are stuck in that spot more than most. Grant Thornton’s 2026 AI Impact Survey found 48 percent of manufacturers are still piloting AI, against 34 percent across all industries. Plenty of shops have tried it. Far fewer have it running on anything that matters.
What the AI is missing is context: everything your business already knows that the model can’t see. What you quoted this customer last spring, what the last job for them made, who owns the account. You have all of it. It’s in your accounting package, your CRM, your email and the spreadsheets next to them. The AI has none of it, so when you ask it a question about your business, it answers like a stranger would.
That’s why the test asks where each answer lives. If your answer to question 1 is “the CRM has the orders, QuickBooks has the invoices, and a spreadsheet in the office ties them together,” you’ve just described why AI can’t answer it. The information exists. Nothing connects it, and the thing doing the connecting is a person.
The workarounds people reach for all have the same problem. You can export the spreadsheet, paste it into ChatGPT, and get a sharp answer. It works. Then next Tuesday you need the answer again, the export is a week old, and getting a fresh one takes longer than just asking whoever usually pulls it. So you don’t. If using the tool takes longer than doing the task, your team has already stopped using the tool.
The fix is to connect the systems so the AI can read them without anyone gathering inputs first.
Why did our AI pilot stall?
AI pilots stall when they’re built on hand-fed data: the demo works because someone gathered the inputs for it, and it stops working the week nobody has time to gather them again. The tool keeps working perfectly. It just never gets run.
The call usually opens one of two ways. Either “I know I need to get AI going, I don’t know where to start, and I need somebody to come build some skills for my team,” or “we already bought the subscription, we’re not getting anything out of it, and I need somebody to run a pilot.” Both are reasonable things to want.
A typical pilot goes like this. Somebody exports the spreadsheets, pulls the CSVs, forwards the emails, and hands all of it over. A skill gets pointed at it, and it produces a drafted email or a finished report that’s impressive in the room. Then it has to run next week without the consultant there. Your team member has to gather all that material again before they can start. It takes twenty minutes, they’re busy, and they do it twice and stop.
A pilot built on hand-fed data proves the model can write, which you already knew. It tells you nothing about whether it will run in your business, and that was the only question the pilot was supposed to answer.
People who work with manufacturers describe clients arriving after a pilot with the same complaint: the silos they expected to break down are still there, and the information they expected to start flowing still doesn’t. The test is a good way to see why that happened. Score your seven questions. If most of them need a download, a retype or a specific person, the pilot was always going to stall, because there was nothing for it to read without help.
Where does AI earn its keep first in a small manufacturer?
In a small manufacturer, AI earns its keep first on one named job, on one named system, where somebody gets time back, and quote intake is the clearest example. An agent on your quote intake can draft a new quote from your last 200, so the estimator edits a draft instead of starting from a blank page.
That job works as a first step because the inputs already exist in one place. Your past quotes sit in a quoting tool, and the person who benefits is easy to name. It also fails in a way you can see. If quote history lives in three places and half of it in an estimator’s inbox, the draft will be wrong, and you’ll know it the first time someone reads it.
Look at the jobs AI does well and you’ll see they map back to the test. A quote drafted from past quotes needs question 3 to pass. An alert when a job crosses its estimate needs question 2. A weekly look at who has gone quiet needs question 1. Asking your own data is all seven questions at once.
So readiness belongs to a job more than to a whole business. A business is ready for a specific AI job when the questions that job depends on pass the test. That’s good news, because you don’t have to fix everything before AI does something. You fix what one job needs.
One we’re starting now is the handoff from prospect to active client. Two of our clients are about to begin connecting their CRM to their operations system so a closed deal builds the project instead of someone retyping it. It’s the same fix we made inside our own company, and it’s the kind of job where AI can help once the connection exists.
What should you fix first, based on your score?
Fix the most-asked question that failed, and fix it at the lowest point where it breaks: whether the answer gets recorded, whether it moves between systems without a person, and only then whether anyone looks at it. Your notes next to each question tell you where that is.
Every failed question points somewhere on the nine-constraint map, the model we use for how information breaks in a small manufacturer. AI sits at the top of that map as constraint 9, starving tools: BI, automation and AI, bought and paid for, fed nothing. You can’t fix constraint 9 by working on constraint 9. A fix at the top inherits every problem underneath it, and it inherits them silently.
Read your worksheet against the map:
- If the answer was never recorded, that’s a capture problem. Nobody wrote it down anywhere it can be read.
- If people record it differently, that’s a definitions problem. “On time” means one thing to the shop and another to the customer.
- If it exists but a person has to carry it from one system to another, that’s a connection problem.
- If it takes one particular person to produce, you’ve found a name in the system.
That last one is the one I’d look at hardest, partly because I’m the name in it at my own company. I manage most of our client relationships, so when I agree a change with a client, the invoice only knows about it if I remember to pass it on. In a busy month I don’t always. We fixed it by connecting our accounting, project and time-tracking systems, so the system flags an invoice that doesn’t add up before it goes out. Nobody had to get more disciplined, me included.
A question can fail in more than one way, and that’s normal. When it does, the lowest constraint on the map is the one to fix first, because the ones above it are often symptoms. If question 2 takes days, lives in a spreadsheet and only your controller can produce it, start with the spreadsheet and the dependency on one person, because the slow report sits on top of both.
This is also why we won’t put AI on a business that can’t feed it. We wrote about that in We Won’t Put AI on a Business That Can’t Feed It, including why we don’t sell AI pilots on their own. You can’t pick the skill out of the project and leave the plumbing, because a skill with nothing feeding it won’t produce reliable results. When we do AI work, we build the connection first and put the skill on top of it, as one project.
Your first step: pick the question you’d ask most often if you could get the answer in two minutes. Find the lowest place it breaks, and that’s your first fix. One question. You don’t need to fix all seven.
Is it worth fixing the systems before buying more AI?
Yes. Fixing the systems is usually cheaper than what you were about to buy, and it’s what makes the AI you already pay for useful. The subscription is the cheap part of AI. It looks wasted because nobody funded the part that makes it work.
I’ll be straight about the tradeoff, because it’s the pushback I hear most. Doing this properly takes longer than a pilot and costs more up front. Our 90-day build is $9,000, and “I don’t have ninety days, why can’t you just write us the AI skill?” is a fair question. The honest answer is that we can write the skill in an afternoon. Writing a skill that works is close to free now. Whether it’s connected to the right systems, and whether those systems are clean and talking to each other, decides whether it’s still running in six months.
There is a narrower version. We can scope to one function instead of one department and finish sooner. Say the target is your inbox: we’d start with triage and nothing else, connected to the email system and only the email system. Starting narrow and widening is slower on paper, and it’s the version that’s still in use next year.
I hold my own company to the same rule. We’ve built an internal operating system that runs most of our business, and people ask why I don’t sell it. I don’t, because it fits the three experienced people who built it, and in a less technical team’s hands it would degrade in ways that are hard to spot. Something that works for one team isn’t automatically something that works for yours, and finding out what yours needs is most of the job.
What should you ask before hiring an AI consultant?
Before you hire anyone to put AI into your business, ask what they’ll connect it to, who maintains it after they leave, and what they’ll need access to. The answers tell you whether you’re buying a working tool or a demo.
AI is the area where the gap between the demo and the Tuesday after it is widest. These are the questions I’d ask, including of us:
- What will it read, and how does it get there? If the answer involves your team exporting files on a schedule, you’re buying the pilot that stalls.
- What happens the week it breaks? Connections drop. Someone has to notice and fix it, and you want to know who before you sign.
- Where will it live? Shared skills inside a tool your team already uses, a local file on one person’s laptop, or something in between. That’s a decision about your team’s technical range, and a good consultant asks about your team before answering.
- What access do you need, and is it read-only first? Our rule is read access before write access, always, with writes scoped and agreed before we make any. Your data isn’t used to train anything, and a person reviews anything an AI produces before it reaches a decision.
- Do you earn anything from the tools you recommend? We don’t sell or resell software. If a vendor ever offers us a commission, it comes to you as a credit on your invoice.
- What’s the one job, and who gets time back? If they can’t name both, you’re buying a strategy.
You may not need a consultant at all. If you have someone technical in-house, use them. The part that trips people up is knowing which systems hold what, in what shape, and what everyone in the building means by the words in them. The code is the easy part.
What happens after you take the test?
Take the test, count your number, and if you want a second opinion on your answers, book a free 25-minute call and bring your answers. The test is useful on its own, and plenty of people will take it and fix their first question themselves.
If you do book, the steps run in this order, and the prices are published because you’d ask:
- A 25-minute call. Free. Five questions about how your systems work today. No access to anything, nothing to prepare, and if you bring the worksheet we start from your answers.
- The Systems Snapshot. A short form covering where each function lives, what gets retyped and by whom, and which of your own numbers you can produce. Your test answers carry over, so you aren’t starting from zero.
- The systems audit, $1,200. The form tells us what you believe, and getting into the systems tells us what’s true. You leave with a 90-day plan you can implement yourself, or the $1,200 comes off the build.
- The 90-day build, $9,000. One thread, one scope, a defined end. By day 90 the systems and numbers the plan named are connected, they reconcile, and the report ships, or we keep working at no additional cost.
And we measure the change. On day one we write down the questions you most need answered and how long each takes, and on day 90 we ask the same questions again. If you took the 7-Question AI Test, you’ve already done the first half.
You’re doing the right thing by checking before you buy. Fifteen minutes with a pencil now is a lot cheaper than another round of seats nobody opens.
Book a 25-minute call. Free. Twenty-five minutes, five questions, nothing to prepare.