How do you know your business is not ready for AI?
Six signs, and none of them are about technology. If nobody can describe a workflow the same way twice, if the data lives in someone’s head, if no one has hours to review outputs, if the real problem is people, if the budget is already strained, or if the whole case is that a competitor said “AI” — the software will not fix what is actually wrong.
It costs us something to say this, because we sell the fix: audits at $3,500–$8,500 and implementation sprints at $12,000–$45,000, on a published price list. But a project built on any of the six signs below fails, and a failed project costs us more in reputation than the sale was worth. So here is the test we would run on your business before taking your money. Run it on yourself first, for free. Its longer companion is the 20-point readiness checklist; this guide is the short version with the exits marked.
Can two people describe the workflow the same way?
If you ask two people to walk you through the same process and get two different processes, you are not ready to automate it. AI built on an undocumented workflow does not fix the chaos — it runs the chaos faster.
Automation is standardization enforced by software. Before a system can draft the proposal, chase the invoice, or file the intake, someone has to decide what “correctly” means, and in most small businesses nobody ever has — the workflow lives as habit, and every person’s habit is slightly different. The test takes an hour: have the two people who touch a workflow each write it down on one page, then put the pages side by side. Every place they disagree is a decision no one has made yet. Those decisions cost nothing to make now and a great deal to discover mid-project, when a build priced in the $12,000–$45,000 range is standing on them.
Where does the data actually live?
If the answer is “in Maria’s head” or “in a spreadsheet on Dave’s desktop,” you are not ready. A system can only work from facts it can reach, and a personal spreadsheet is not a system of record — it is a single point of failure with formatting.
This is the most common sign we see in professional-services firms: client history scattered across four inboxes, pricing logic in a partner’s memory, project status in whichever tool each manager happened to like. None of that is fatal, and the fix does not require new software. It requires a rule: one system of record per fact. A customer’s status lives in exactly one place; a price lives in exactly one place; when the spreadsheet and the CRM disagree, the CRM wins. Until that rule exists, any AI you buy will be confidently wrong, because it will be reading whichever copy of the truth it happened to find.
Who has two hours a week to review the outputs?
If nobody on the team can spare two to three hours a week to review what the system produces, do not build it. Every AI system worth having keeps a person in the approval seat, and an empty approval seat is where these projects go to die quietly.
The biggest system we have published — MARCUS, 14 AI agents across 7 departments of an SBA lender — is built so that nothing sends, files, posts, or pays without human approval. That design is not decoration; it is the reason the system is trusted. But it only works because named people actually show up and approve things. If your team is already at capacity, the reviews will slip, a backlog of unreviewed drafts will pile up, and a backlog of drafts is worse than no drafts — it is work someone now feels guilty about. Approval is the bottleneck. Budget the human hours before you budget the software, and if the hours do not exist, that is your answer for this quarter.
Are you solving a people problem with software?
If the workflow keeps breaking because of turnover, unclear roles, or two managers who disagree about who owns it, AI will not fix it. Software is very good at executing a decision and completely unable to make one for you.
The tell is the sentence before the AI idea. “We can never keep that seat filled, so maybe AI could…” — that is a hiring or management problem wearing a technology costume. So is “nobody really owns follow-up,” and so is the deferred conversation with the employee whose work everyone quietly redoes. Automating around a people problem does not remove it; it pours concrete over it, and the ambiguity you laminated into the system is now expensive to change. Settle who owns the workflow and what good performance looks like first. Sometimes that settles the whole question, and no software was ever needed.
Is the money actually there?
If a $3,500 audit — the bottom of our published range — would genuinely strain the budget, wait. Not because the audit is the number that matters, but because the running costs behind any working system are typically $50–$500 a month, indefinitely, and a strained budget cuts that line first.
A system whose subscriptions get cancelled in month five is worth less than no system: you paid for the build, disrupted the routine, and kept neither. The full arithmetic is on the pricing page, deliberately, so you can make this call without sitting through a meeting. If the numbers do not fit this year, skip the meetings entirely — that is not failure, it is sequencing. The 90-day plan below costs about $25 a month, and it moves you forward anyway.
Do you want AI, or did a competitor mention it?
If the entire case for the project is that a competitor put “AI-powered” in a newsletter, you are not ready. Envy is not a workflow.
A competitor’s press release tells you nothing about their operations — most announcements describe pilots, and most pilots quietly end. The Ampersand essay How to Smell the Hype is a field guide to exactly this genre of claim. The right trigger for an AI project is boring and specific: a named workflow that eats a known number of hours every week. If you cannot name the workflow, the project has no subject — and a vendor will happily supply one that fits their product rather than your business.
What should you do for the next 90 days instead?
Run this five-step plan. It costs about $25 a month plus a few afternoons, and it clears most of the six signs without a consultant in the room.
- Write down your top three workflows — one page each. What comes in, who touches it, what goes out. Where two people’s pages disagree, that disagreement is the finding. This one step does more for readiness than any tool.
- Pick one system of record per fact. Every fact that matters — customer status, price, job stage — gets exactly one home, and the home wins every argument with every copy.
- Take the free AI-Ready Score. Fourteen questions, about 7 minutes, a 0–100 score, no sales call. It puts a number on where you stand today so the retest in step five means something.
- Read two essays. What the Machine Cannot Do and How to Smell the Hype — roughly twenty minutes that will save you from most vendor meetings.
- Run one $25-a-month AI subscription with one named owner. One person, one tool, one job — drafting responses, summarizing calls, whatever fits. Ninety days. If it sticks, you have learned where the value is; if it does not, that is real information and it cost you $75.
Then re-run this list. Ninety days of writing things down beats ninety days of vendor demos every time. The plan has a useful side effect, too: it makes you a harder customer. A business that shows up with documented workflows, one source of truth, and a named reviewer gets sharper quotes, shorter projects, and fewer change orders — from any firm it eventually hires, including this one.
How do you know when you have become ready?
The six signs flip. The workflows are on paper and two people describe them the same way; the facts live in systems; a named person has review hours; the people questions are settled; the budget covers the run rate; and you can name the workflow you want back.
At that point, take the full readiness checklist as a final pass, then take a meeting if you want one. Ours is a free 30-minute assessment, and it works as a readiness read in its own right: a senior advisor walks your workflows and tells you what we see — including “wait,” when that is the true answer, because it sometimes is. If the read is “ready,” the next step is an AI Readiness Audit at $3,500–$8,500 over 2–4 weeks, which replaces all of this self-assessment with measured numbers. And if the read is “wait,” you keep your money and leave with the same 90-day plan you just read. We reply within 24 hours either way.