You want to do something with AI this year. You have a budget in mind and a vendor or two in your inbox. What you do not have is a way to tell whether your business is actually ready, or whether you are about to pay for a system that will sit on top of a filing cabinet and a process that lives in three people’s heads.
The AI readiness checklist below is the one we run before agreeing to build anything. It has twenty points in seven areas. None of them are technical. Most businesses pass about twelve on the first read, and the eight they fail are exactly where the project would have died.
Go through it honestly. Where you fail a point, the fix is usually a week of management work, not a purchase, and doing it first is the cheapest thing you will do all year.
What this actually is
AI readiness is not about technology. A model can be connected to almost anything in an afternoon. Readiness is whether the material the AI needs exists in a usable form, whether the process it is meant to automate is real, whether someone inside the business will own it, whether the rules for what it may do are written down, and whether you will know if it worked. Those are ordinary business preconditions. AI just exposes them faster than most changes do, because a model given a messy process produces messy output at speed.
The analogy is bringing on a very capable temp on a Monday. Before they arrive you would want their desk to have the files they need, a written procedure, a person to report to, a clear list of what they may decide alone and what they must ask about, a budget for their time, and a way to tell at the end of the month whether they helped. The checklist is that, for software.
The AI readiness checklist, all 20 points
1. Your data is in one place and means what it says
- Point 1: the records the AI would need (customers, jobs, orders, documents, prices) live in identifiable systems, not in personal spreadsheets and email threads.
- Point 2: the fields in those systems are used consistently (a “status” column does not mean five different things depending on who typed it).
- Point 3: you can get the data out. The system has an export, an API (a way for other software to read it), or at least a scheduled report.
This is the point most businesses fail, and it is the one that decides everything after it. An AI system that answers from your documents needs the documents in one searchable place. A quote drafter needs the price list to be the price list, not one of four versions. If the honest answer is “it is in people’s heads and a shared drive,” the first project is consolidation, not AI, and it is worth doing regardless.
2. The process you want to automate is written down
- Point 4: there is a one-page description of how the task is done today, including the exceptions and who handles them.
- Point 5: the people who do the task agree that the page is accurate.
- Point 6: you can point to the input (the email, the form, the call) and the output (the quote, the booking, the record) of the task.
You cannot automate what you cannot describe. When a builder asks how quoting works and gets four answers, the project has already failed; the software will encode one person’s version and the others will ignore it. Writing the page is a management exercise, it takes an afternoon, and it often reveals that the process itself needs fixing before any software touches it. That is a good outcome. Do the fix first.
3. Someone inside the business owns it
- Point 7: a named person is responsible for the AI system’s outcome, and it is in their job, not a favor.
- Point 8: that person has time protected for it, a rough rule being two hours a week during build and one hour a week after.
- Point 9: that person will read the logs, keep the material current, and be the one staff go to when the system does something odd.
Vendors and builders come and go. The person inside the business is the reason the system is still working in a year. In our experience the projects that stick all have this person, and the ones that fade all lacked one. It does not have to be the owner. It usually should not be, because the owner will not have the hours. An operations manager, a practice manager, a senior admin: someone close to the work.
4. The rules for what the AI may and may not do are written
- Point 10: there is a written list of what the system may do on its own (classify, draft, summarize, book a routine slot) and what it must never do without a person (send money, quote a price it was not given, promise a date, answer anything legal or medical, reply to anyone angry).
- Point 11: there is a written rule for what happens when the AI is unsure: who it hands off to, through what channel, with what summary.
- Point 12: you have decided what customer or staff data may reach the model, and read the provider’s terms on training and retention.
These three points are the difference between an AI system that can be trusted with customers and one that cannot. They are decisions, not code, and they have to come from you, because they encode your appetite for risk. Write them before the builder starts and they become the specification. Leave them until after and the builder guesses. AI Agent Permissions: Least Privilege for Business AI and AI Data Privacy for Business, What Happens to Customer Data give the detail for points 10 and 12.
5. There is a budget, and it includes running costs
- Point 13: there is a number for the build or the subscription, agreed before anyone starts.
- Point 14: there is a monthly running-cost line (model usage, platform, server, phone minutes) estimated from the current pricing pages.
- Point 15: there is a maintenance line: a few hours a month of a capable person, or an arrangement with the builder.
The first number everyone has. The second most businesses can estimate with the arithmetic in AI Cost for Small Business, What It Really Costs to Run in 2026. The third almost nobody budgets, and it is why systems decay. Providers retire model versions, upstream tools change their export formats, knowledge bases go stale. If nobody is paid to notice, nobody notices.
6. You know what success looks like and how you will measure it
- Point 16: there is one measurable target, agreed in writing, with a date (“missed calls under 5 percent within 90 days,” “quotes out within two hours for most requests,” “support inbox cleared daily by one person instead of two”).
- Point 17: you know the current number for that measure, today, before anything is built.
- Point 18: there is a review date, in the calendar, at which you decide to keep, change, or stop.
“Explore AI for sales” is not a target and cannot succeed or fail. A number with a date can. The current number matters as much as the target, because without it you will never know what changed. And the review date is what stops a three-month trial from becoming an unexamined permanent line item. AI Roadmap for Small Business: 90 Days Without a Developer shows how to sequence these into a first quarter.
7. The basics of ownership and access are in place
- Point 19: every account the system will use (model provider, automation platform, server, domain, phone number) can be registered to the business, billed to the business, with the builder as a removable user.
- Point 20: the staff who will work alongside the system have been told what it does, what it does not do, and how to flag a problem.
Point 19 is the rule that saves businesses from losing a system when a vendor or developer relationship ends. Insist on it before the first line of work. Point 20 is the one that decides adoption. Staff who were not told what the system does treat it as a threat or ignore it. Staff who were told, and who have a way to say “it got this one wrong,” become the system’s best source of corrections. Five minutes in a team meeting is enough.
Picture a business like this one
The business below is a composite of the kind of company that writes to us, not a client. The numbers describe the shape of the problem, not a case study.
Picture a business like this one: a commercial cleaning company with 60 staff, two office admins, and a client roster of about 150 sites. The owner wants “AI for scheduling and client communication.” Running the AI readiness checklist, the company passes on ownership (the operations manager is keen and has time), on budget, and on access. It fails on data: schedules live in one admin’s spreadsheet, client contact details are in a second spreadsheet and in email signatures, and site instructions are in a shared drive of Word documents with no naming convention. It also fails on process: the two admins handle client requests differently, and neither version is written down.
What gets done first is not AI. Over three weeks, the operations manager consolidates schedules and client details into one system with an export, agrees a single one-page process for client requests with both admins, and writes the allowed-actions list: the system may draft replies and log requests, it may propose schedule changes, it may never confirm a change or a price without an admin. A target is set: client requests acknowledged within 30 minutes during business hours, measured from the current average of about four hours.
What gets built, after that:
- An inbox step that classifies incoming client emails, logs each request against the right site, and drafts an acknowledgment for the admin to approve.
- A schedule-change proposer that reads the request and the consolidated schedule and suggests a change, which an admin confirms or edits.
- A log the operations manager reviews every Friday, with a monthly list of the requests the system could not classify.
What changes is that acknowledgments go out in minutes, both admins handle requests the same way because the way is now written down, and the owner has a number on a page each month. The three weeks of preparation were what made the software work.
What it costs to run
For the cleaning company, the tooling is modest: model usage for classifying and drafting a few hundred emails a month is single-digit to low double-digit dollars; an automation platform tier or a self-hosted n8n on a $15 to $30 server; the consolidated scheduling system, if a hosted product, at whatever its per-user pricing page says. Total tooling likely under $100 a month. The larger cost is the operations manager’s time: three weeks of part-time preparation, then an hour a week. That time is also the reason the money spent on tooling is not wasted.
The mistakes we see most
- Treating the checklist as a formality. Ticking “data in one place” because there is a system, when half the records live outside it.
- Making the owner the owner. The business owner rarely has the weekly hour. Name someone closer to the work.
- Skipping the written rules. Leaving “what may it do alone” to the builder, who will guess, and guess more permissively than you would.
- Setting a target without a starting number. You cannot show improvement from a baseline you never measured.
- Passing on paper, failing in practice. The process page exists but the people who do the job have never read it.
When to bring in help
Every point on this checklist is something an owner or manager can do without a developer, and most of them should be done before a developer is involved. Consolidating data, writing the process page, naming an owner, writing the rules, setting the target: this is management work, and it is where the value of an AI project is actually created.
A developer is needed after the checklist passes, to build the parts that are software: the integration with your consolidated system, the classification and drafting steps, the approval screen, the log, the alerting. A good builder will run an AI readiness checklist like this one on the first call and will tell you if you are not ready yet. That is a sign of a good builder, not a stalling one. How to Hire an AI Consultant Without Getting Burned covers the rest of what to ask.
Levelbrook runs this checklist with every business before scoping a build, and will say plainly if the answer is “consolidate first.” When the checklist passes, we build on a fixed price from a written scope, in accounts you own, with the owner on your side written into the plan. If you want to go through the twenty points with someone, the form below is where to start.