You're not ready yet if you can't name the job you want AI to do, the work lives in one person's head, your customer records disagree with each other, nobody holds the admin logins, or nobody has two hours a week to run a trial. Fix the first of those that applies before you pay for anything.
Not ready rarely means "wait a year". Most of the fixes below take a week or two, several take an afternoon, and you can keep using a business AI account for your own low-risk drafting while you make them. The signs matter because they're cheap to fix now and expensive to discover halfway through a paid project, when someone is billing for the wait.
1. Nobody can say which job AI should take on
How it shows up: the plan is "we should be using AI", the tool was chosen before the task, and different people have different ideas of what it's for. Usage starts high and fades within a month.
In practice: an illustrative tutoring agency bought four ChatGPT Business seats for its coordinators, $100 a month on monthly billing. For the first fortnight everyone experimented. By week four, two people used it occasionally for emails and two had stopped. Nothing was wrong with the tool; nobody had decided which of the agency's jobs it was meant to shorten.
Fix first: log where the week goes for five working days, ten minutes a day, then pick the one repeated writing or admin task that takes the most hours. That task, named in one sentence with its weekly hours, is your starting point. Auditing your workflows for AI opportunities turns the log into a ranked list if you want more than one candidate.
2. The process only exists in one person's head
How it shows up: "ask Sam, she does that". Holidays cause chaos, nobody else can describe the steps, and exceptions are handled from memory.
In practice: at an illustrative independent optician, the practice manager ran recall reminders entirely from memory: which patients were due, which were on different recall intervals, which letter wording suited which patient, and who never to chase. An AI drafting tool could have written the letters, but nobody could tell it the rules, because the rules had never been written down.
A chat assistant can do the tidying once you have the raw notes. A prompt that keeps it honest:
Below are my notes from our practice manager talking through five
recall reminders she sent last week. Turn them into numbered steps
anyone on the front desk could follow. List every exception
separately, and mark anything that sounds like a judgement call
with [DECISION]. Don't add steps she didn't describe.
[paste the notes, with patients' names removed]An illustrative reply, abridged:
"1. Each Monday, run the recalls-due report for the next 14 days. 2. Remove anyone who already has an appointment booked. 3. Send the standard reminder by email, or by letter if no email is held. Exception: patients on a shorter recall interval get the annual-check wording. [DECISION] Patients who haven't attended for over two years: decide whether to send the reminder. 4. If there's no reply after 10 days, send a text."
What you'd fix: step 4 invents a text follow-up she never mentioned, so delete it. The [DECISION] flag is the valuable part, because it marks exactly the rule that lived only in her head and shows where a person must stay involved. Check the finished steps with her before anyone relies on them.
Fix first: have the person talk through five real, recent cases while someone records or takes notes, which takes about 20 minutes. Turn the notes into numbered steps with the exceptions listed separately. Documenting your processes before adding AI covers the method; the output also protects you the next time that person is off.
3. Your customer records disagree with each other
How it shows up: the same customer appears twice with different emails, the till's customer list doesn't match the email list, and nobody trusts the numbers from either.
In practice: an illustrative two-shop pet store kept loyalty members in its till system, newsletter subscribers in an email tool and online buyers in its web shop. One regular appeared as three different people. An AI-written "we miss you" campaign built from the web shop list would have gone to customers who visit every week in person.
Fix first: take a sample of 50 records from the list the AI would use and count duplicates, blanks and out-of-date entries. If more than a handful are wrong, choose one list as the master and merge the others into it before any AI touches customer data.
4. Staff already paste work into personal AI accounts
How it shows up: someone mentions "I asked ChatGPT" in passing, there's no company AI account, and nobody knows what's been pasted where.
In practice: teachers at an illustrative language school were pasting students' written work into free chatbots to draft feedback. It saved them time, and it meant students' work sat in personal accounts whose model-training settings nobody had checked. Consumer plans rely on each user switching off the model-training setting themselves; business plans don't train on business content by default.
Fix first: don't punish it, because it shows you where the demand is. Write a one-page rule on what may and may not go into AI tools, move the work into one business account, and ask each person what they've been using it for. Shadow AI in a small business has the conversation and the questions.
The rule can be short. The opening of the language school's version:
Using AI at [school name]
1. Use only the school's AI workspace for school work, never a
personal account.
2. Never paste a student's full name, date of birth, contact
details or anything about their welfare into an AI tool.
3. Student writing may be pasted for feedback with the name removed.
4. A teacher reads and edits all AI-drafted feedback before a
student sees it.
5. Not sure? Ask [name] before you paste.5. Nobody in the business holds the admin logins
How it shows up: "the web designer set that up", "my nephew has the password", or an admin account in the name of someone who left. Switching on any new feature needs a favour from outside.
In practice: an illustrative picture framer's Microsoft 365 had been set up by a former part-timer, and the only administrator account was theirs. When the owner wanted to try Copilot Chat, which comes with Microsoft 365 business plans, there was nobody who could change a single setting.
Fix first: recover admin access through the vendor's account-recovery route, then make sure at least two accounts the business controls hold admin rights, both protected by two-step sign-in. Microsoft itself recommends keeping two emergency-access administrator accounts for exactly the day the usual admin is locked out or gone. This is often a half-day job, and every later AI step depends on it. A quick test of whether this sign applies: can you, today, open the admin page of your email, your website and your main business software yourself, without asking anyone?
6. You don't know how long anything takes now
How it shows up: "it takes ages", "it's a nightmare", but no number. Any claim that AI saved time will be a feeling, and feelings don't survive a renewal date.
In practice: a solo music teacher with about 45 pupils was sure termly invoicing "took a whole day". Timing it showed that producing the invoices took about three and a half hours. The real drain was chasing late payers afterwards: around 40 messages spread over three weeks. The AI plan changed from generating invoices to drafting polite, staged payment reminders, which was the job actually eating the time.
| Part of the job | What the teacher believed | What timing showed |
|---|---|---|
| Producing 45 termly invoices | "A whole day" | About 3.5 hours |
| Chasing late payers | "A few reminders" | About 40 messages over three weeks, roughly 4 hours |
Fix first: count and time the task for two normal weeks: volume, minutes per item, how often it's redone. Write the figures down with the date. Ten timed examples beat a month of impressions, and the split between parts of a job often matters more than the total.
7. The process is broken, not just slow
How it shows up: the same mistake happens every week, work gets redone, and customers complain about the same thing. Staff say "it's always like this".
In practice: the tutoring agency from sign 1 had lessons recorded in three places: the coordinators' calendar, tutors' own diaries and a spreadsheet. Double bookings happened weekly. An AI booking assistant would have made bookings faster, into whichever of the three it could reach, and so produced double bookings faster.
Fix first: agree one source of truth for the process and run it by hand for a fortnight until the repeat errors stop. Then automate the working version. Automating a broken process explains why this order matters and how to spot which kind of problem you have.
8. There's no slack for learning in the next two months
How it shows up: a peak season, a new till or booking system going in, a move, or the one organised person about to go on leave. Everyone is already stretched.
In practice: the pet store planned an AI rollout for its busiest trading weeks of the year, on the logic that it would help most when things were hectic. The staff who needed to learn it had no time to, the owner had no time to check the drafts, and the pilot was abandoned in its second week with nothing learned.
Fix first: schedule the pilot for the first quiet eight weeks you can see, and use the busy period for the paperwork fixes from this list, such as the written rule, the admin logins and the counting. Those need an hour here and there, not sustained attention.
9. You're hoping AI will fix a people or product problem
How it shows up: the pain is poor reviews, long waits, staff turnover or a product customers don't like, and the proposed fix is an AI tool that talks to customers faster.
In practice: the optician from sign 2 considered AI-drafted replies to online reviews. Reading the reviews showed that most of the complaints were about waiting weeks for glasses to come back from the lab. Faster, politer replies wouldn't change the wait, and a stream of well-written apologies for the same delay would have read worse over time.
Fix first: read the last 30 complaints or bad reviews and tally the root causes. Fix the top cause directly. AI can help later with the honest parts, such as proactive status updates when an order is delayed, once there's something true and useful to say.
10. The money for a three-month trial isn't there
How it shows up: you'd cancel after the first month's bill, or the trial would come out of money set aside for something essential.
In practice: the language school priced six Claude Team Standard seats at $25 each on monthly billing: $150 a month, or $450 for a three-month trial, plus the staff hours. In a low-enrolment term that was money it didn't have, and a trial run on a budget you resent tends to be judged in week two, not week twelve.
Fix first: start with AI already included in the software you run. Google Workspace plans now include Gemini, and Microsoft 365 business plans include Copilot Chat at no extra cost. One owner's individual plan, around $20 a month, is often enough to prove a single job before anyone else gets a seat.
11. Nobody has checked the rules that apply to your customers
How it shows up: your customers include children, patients or people in sensitive situations, or you'd be recording conversations, and nobody has asked what that means for an AI tool.
In practice: the music teacher wanted to record lessons and have AI write the summaries for parents. Most pupils were under 18. Recording children raises questions about their parents' consent, where recordings are stored and for how long, and what the note-taking tool does with the audio. None of that had been thought through, and it was the part most likely to cause real harm if handled carelessly.
Fix first: list what's unusual about your customers and your data, then take the specific questions to your data-protection adviser before switching anything on. If you sell to customers in the EU and plan a chatbot, the EU AI Act's transparency duties already apply: people must be told when they're talking to an AI.
Which fix to make first when several signs apply
Most businesses find three or four signs apply. Order the fixes so each one makes the next easier:
- Access first (sign 5). Nothing else can be set up without it.
- Rules next (signs 4 and 11). They stop the current risk while you work on the rest.
- Time and timing (sign 8). Book the weeks and the person before designing anything.
- The job and its numbers (signs 1 and 6). Pick the task, then measure it.
- The process and the data behind that job (signs 2, 3 and 7). Fix only what the chosen job depends on, not everything.
Signs 9 and 10 are different: they're go or no-go questions. If the real problem isn't one AI can touch, or the money isn't there, the right move is to fix the business problem or wait for a better quarter, and that's a sound decision rather than a failure.
Here's how that ordering played out for the pet store, which found signs 1, 3, 5 and 8 applied:
Week 1 (busy season, one hour a week available)
- Recover the web shop admin login from the web designer.
- Add a second admin account; switch on two-step sign-in.
Week 2
- Write the one-page AI rule; share it at the staff meeting.
Weeks 3-4 (still busy)
- Five-day time log for the owner and one shop manager.
First quiet week
- Pick the job: weekly stock reorder emails to three suppliers
(about 3 hours a week).
- Time ten reorders; note how often an order is corrected.
Next two weeks
- Merge the supplier contact list (the only data this job uses).
- Re-score readiness; start a four-week pilot.Notice what's missing: the customer-list clean-up from sign 3. The chosen job didn't use customer data, so that fix could wait. Fixing only what the first job needs keeps groundwork from becoming a project of its own. If you'd like a single number to track progress, the 20-minute AI readiness checklist gives you one to re-score after each round of fixes.
What you can safely do with AI while you're not ready
Not being ready for an AI project doesn't mean avoiding AI. With a business account and a person checking the results, these are low-risk while you fix the rest:
- Drafting internal documents: a staff rota note, a job advert, a supplier email.
- Turning your own rough notes into a tidy summary or checklist.
- Rewriting a policy or a price list description in plainer words.
- Brainstorming headlines or product descriptions that you then edit.
- Explaining a contract clause or a software setting to you in plain English, with the original kept alongside.
Keep customer, patient and pupil details out of it, keep a person between the AI and anyone outside the business, and keep it to jobs where a wrong draft costs a minute to fix. That's also good practice: by the time the groundwork is done, whoever will own the first pilot will already know what the tool does well and where it guesses. Ownership is its own question, and deciding who should own AI in a small business is worth settling before the pilot starts.
Further reads
- How to Set a Baseline Before You Introduce AI — Get the before numbers that sign 6 says you're missing.
- Is Your Business Data Ready for AI? A Clean-Up Checklist — A 50-record test for the data problems behind sign 3.
- How to Clean Up Customer Records Before You Add AI — Merge duplicate customers before any AI writes to them.
- How to Write an AI Usage Policy for Your Small Business — The one-page rule that clears sign 4 in an afternoon.
- Do I Need AI in My Business? A Decision Guide for Owners — Check whether AI is the answer before readiness matters.
- Is My Business Too Small for AI? What Works for Teams of 1 to 10 — What still works for teams of one to ten.
- AI Readiness Checklist for Accountants, Solicitors, Consultants — Twenty checks, grouped and scored, that tell an accountancy, law or consulting firm whether it is ready to pilot AI or has gaps to fix first.
- Signs Your Business Is Ready to Automate Sales Follow-Ups With AI — A 16-point readiness checklist for automating sales follow-ups, each item with how to verify it, plus a filled-in example and a scoring table.
- What Is a Fractional Chief AI Officer and Do You Need One? — What a fractional chief AI officer does each month, what providers publish as prices, and cheaper set-ups that suit most small businesses.
- What Is an AI Readiness Assessment and What Does It Involve? — What an AI readiness assessment examines, how it runs, a scored example for a homeware brand, and what vendor readiness reports leave out.
- How Much Does an AI Readiness Assessment Cost? — Free, vendor-funded and paid readiness assessments compared, where the days go, your staff's share of the cost, and a way to compare three quotes fairly.
- AI Audit vs AI Readiness Assessment: What's the Difference? — How an AI audit differs from a readiness assessment, the four things sold as an 'AI audit', worked barber and dry cleaner examples, and a DIY version of each.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI, Anthropic and Google plan pages (business seat prices, training defaults, included Gemini features); Microsoft 365 business plan pages (Copilot Chat included); Microsoft Learn (security defaults and emergency access accounts); EU AI Act Article 50 transparency duties.