Nobody Is AI-Ready. Here's What to Do Instead of Waiting.

There's a race going on, and a lot of businesses are waiting at the start line. Not because the technology is out of reach. Most of it isn't, and a fair amount of it is already sitting inside licences they pay for every month. They're stuck earlier than that. Where does this fit? Which tool? Is our data good enough? Who's supposed to own it? Nobody has a confident answer, so the decision gets pushed back — politely, sensibly, to next quarter. The trouble is that the readiness people are waiting for doesn't turn up on its own. Strategy agreed, staff trained, data tidied, rules written, systems set up, then begin: that's a description of the finish line, not the starting gun. Businesses arrive at that state because they started, not before. The opposite mistake is just as common, mind you. Signing off a batch of Copilot licences and waiting for transformation to happen isn't a strategy either. It's a subscription. Somebody uses it to tidy up emails, the numbers don't move, and eighteen months later the conclusion is "we tried AI, it didn't do much for us." To be clear, none of this is an argument for being reckless. Security and compliance still matter, and nothing here is permission to paste your customer list into a free chatbot and see what happens. Keep the risk small. Start where the stakes are low, with material that isn't sensitive, and with someone checking the output before it goes anywhere. But do it. That's the part that gets lost. Careful is not the same as stationary, and a lot of businesses have quietly turned the first into the second. If the caution never produces a decision, it isn't caution any more — it's just being stuck, with better vocabulary. So here's the order of work I'd use instead. None of it requires you to be ready. Start with a problem, not a tool The worst version of this conversation begins with "we want to use AI in marketing." The useful version begins with a task somebody does every week and quietly resents. Go and ask them. Sit with whoever handles the most repetitive part of the week and ask what they'd hand over if they could. You'll usually get the same handful of answers: building quotes and proposals from the same source material, first drafts of reports nobody reads closely, sifting a shared inbox, pulling the same numbers into the same monthly pack. Write two or three of them down, along with how long each takes and how often it happens. That's not a strategy document, it's half a page. But it turns "we should do something with AI" into a shortlist you can actually test, and it means you'll recognise an improvement when you see one. This is the same discipline behind process mapping: name the work before you name the technology. Pick the boring ones first. Something repeatable, low-risk, and done often enough that a small saving adds up. Resist the urge to start with the thing that would be most impressive in a board pack. Find out what you're already paying for A surprising number of businesses don't have a clear picture of what they already own from Microsoft. Which plan each person is on, what that plan includes, what's been bought and forgotten, and which AI features are already sitting there switched off. Get someone to produce that list. Licences by user, what each one covers, and what an upgrade would actually cost. One warning worth taking seriously: Microsoft changes its prices and what's included in each plan fairly often, and it has done so more than once in the past couple of years. I wouldn't trust a blog post on it, including this one. Check where things stand today with Microsoft directly, or with whoever looks after your IT. It's a short conversation and it occasionally saves a five-figure decision. Fix the plumbing before you open the doors AI is only as good as the material it can reach, and most businesses have ten years of documents piled up in shared folders. Four versions of the same policy. A 2019 price list nobody ever took down. Three folders that all look like the one you're meant to be using. You don't have to fix all of it. That's the mistake that turns this into a two-year project and then no project at all. Fix the part your first pilot needs. One site, one folder, one set of documents: duplicates deleted, everything up to date, everything clearly named. If that sounds familiar, it's the constraint I wrote about in data quality. Who can see what deserves its own look, and this is the bit people skip. Sharing settings go wrong in two directions. Either the tool can't reach the material that would make it useful, or it can reach material nobody should ever see — and the moment it starts answering questions, it will hand that material to whoever asked. The test is quick. Think of a document a new starter shouldn't be able to read: a salary review, a commercially sensitive contract, an HR file. Then check whether search finds it for someone at that level. If it does, you've learned something important, and you've learned it before the tool made it obvious to everybody at once. Give it an owner, and give people somewhere to ask Nothing moves without a name against it. That doesn't mean hiring anyone or inventing a title. It means one person who has this in their week, who knows what's being trialled where, and who stops three departments quietly buying three different tools. Training matters more than people expect, and not the training they expect. An hour on what these tools get wrong is worth more than a day on what they can do. People need to know that a confident answer can be entirely invented, that outputs need checking against something real, and where judgement still has to be human. Learning how to word a request is the easy part, and the least important. The rules can fit on one page. Which tools are approved, what information must never be pasted into a public one, who to ask when someone wants to try something new. That's it. The point is to let people try things without holding their breath. A policy that only says no gets worked around, usually on people's personal accounts, which is exactly what it was written to prevent — the shadow AI problem in one sentence. Decide what "better" looks like before you start Usage figures are not results. A dashboard showing that eighty per cent of staff opened Copilot last month tells you nothing about whether the business improved. So write down the current state before you begin. This task takes roughly four hours a week. We send out about twelve of these a month. We get this many corrections. Whatever the measure is, write it down while it's still unflattering, because nobody remembers it honestly once the tool is in. Then look again at six or eight weeks. Time freed up, better quality, fewer mistakes, a bit of breathing room in a team that was drowning. If something genuinely improved, you have a case for the next one. If nothing moved, change the approach or stop — and either of those is a legitimate result, as long as somebody makes the call rather than letting it drift. What you get for doing this The second pilot is easier than the first. That's the part worth holding onto. You'll have sorted out who can see what, once. You'll have one clean set of documents rather than a swamp. Someone owns it. People know enough to spot a wrong answer instead of forwarding it. And you'll have developed the thing you can't buy — a feel for where this works in your business and where it plainly doesn't, which is just as valuable and considerably rarer. That's the real gap opening up between businesses right now. Not who picked the better tool. Who's been learning for a year while everyone else waited for conditions to be right. You don't have to be ready to start. You do have to start getting ready. Carefully, in small steps, with the sensitive material locked down and somebody checking the results — but started. And the first move is an afternoon with the person who does your most repetitive job, and a piece of paper. FAQ What does AI readiness actually mean for a small business? It means having a named problem worth solving, a person responsible for the work, documents that are up to date and visible to the right people, and a way of telling whether anything improved. It doesn't mean a finished strategy or a rebuilt IT system. Most businesses that describe themselves as "not ready" already have enough to run a first pilot. Do we need to fix all our data before we use AI? No, and trying to is how these projects die. Fix the material your first pilot actually depends on: one site or folder, duplicates deleted, everything up to date, and visible only to the right people. The wider clean-up can follow once you've proved something works. Should we buy Copilot licences first? Not first. Establish what your existing licences already include before you buy anything, and bear in mind that what's included in Microsoft's plans changes regularly, so check where things stand today with Microsoft or with whoever looks after your IT. Licences are the easy part. Whether they get used depends on the groundwork around them. Who should own AI in a small business? One named person with time allocated to it, not a committee and not a new hire. Their job is to keep the different efforts joined up, stop two departments buying the same thing twice, and make sure someone is checking whether the pilots actually delivered anything. Seniority matters less than having the authority to make decisions and the time to follow through. If you'd find it useful to talk through where AI actually fits in your organisation, I'm happy to have that conversation. A working session, not a sales pitch — get in touch.

This article was written by Peter Lowe. The ideas and opinions are his own; AI was used to assist with drafting and editing.