The sales call ended on a high. The copilot drafted a perfect customer reply in seconds. Someone said “this will transform the office.” Seats were almost ordered before the kettle boiled.

Then Monday arrived.

The inbox was not the clean demo mailbox. It was half-finished threads, job notes with typos, and an invoice scan rescued from a van. The person who “owned” the pilot was on leave. Nobody had booked time to check outputs. By Wednesday the tool was still on, the metric was “we’ll know it when we see it,” and nobody could say what success looked like without sounding ungrateful.

That scene is common in UK SMBs. The gap is rarely “we need more AI.” It is organisational readiness: a named sponsor, a plan before the shopping list, people’s calendar time, one focused trial, and the nerve to score honestly.

This is not a vendor roast. Tools can help — drafting routine emails, summarising job notes, pulling fields from invoices. The operational question is simpler and harder: are we ready to try this one job safely?

Why honesty beats optimism theatre

Vendor promises and business facts are different kinds of claim. A polished demo is a vendor claim. Your messy data, shared passwords, and already-full diaries are facts you have to face.

Optimism theatre sounds like: “We’re an AI-ready company now.” It usually means seats were bought and a channel was renamed. Honest readiness sounds like: “We don’t know where prompts go yet — score that zero before we paste customer names.”

If you are unsure, score 0. Guessing high hides the gaps that bite in week two. A low honest score is useful. A fake high score is expensive theatre.

This is not legal advice. It is operational hygiene for people who still have to ship work on Monday.

Management buy-in is a person with a name (and a budget for time)

“The directors are supportive” is not sponsorship. Sponsorship is a named person who can say yes or no to spend and to people’s time.

Without that name, pilots drift. Nobody protects the diary slots or stops the trial when kill criteria are hit. IT might switch a feature on; they cannot own whether enquiry drafts are actually better.

Write the name. Even if it is you. Pair it with a process owner who lives in the work — invoices, enquiries, scheduling — not “whoever installed the tool.”

Buy-in without calendar authority is applause. Buy-in with a named sponsor and booked hours is a pilot.

Preplanning and assessment before the shopping list

The shopping list is the easy bit. The hard bit is deciding what you are testing before anyone demos you into a decision.

Before you compare tools, write one sentence for the job. Good: “Draft first replies to routine customer email enquiries so the office team spends less time typing.” Bad: “Be more productive with AI.”

Then separate promise from proof. What would you see in your files in week one that proves this works? What will still be done by a human — review, approval, customer contact? What would make you stop (kill criteria): too many factual fixes, no time saved after four weeks, staff will not use it, unexpected personal-data sharing?

That preplanning is assessment, not bureaucracy. It stops the classic UK SMB trap: turning on a helper in software you already pay for, then discovering nobody agreed what “good” looked like.

Do not skip the boring gates. If you cannot list the data the job needs, or whether you are allowed to use it for this purpose, you are not ready to shop. Foundations first; shortlists later.

Assess honestly, book the time, run a bounded pilot.

Resources = calendar time, owner, deputy — not just seats

Licences are visible costs. Hours are the costs that kill pilots.

A 90-day sketch should include setup, training, and weekly hours for people who will review AI output. If those hours are not in the diary, you allocated hope, not resources.

Name an outcome owner who can spot a wrong or made-up answer in this job’s content. Name a deputy for holidays and illness. A pilot only one person understands dies the week they are out — or runs unchecked.

Write a half-page “do not paste” rule: passwords, card data, full customer exports, sensitive HR notes. Verbal once is not a control.

Seats without time is how you get an expensive unread feature and a team that quietly goes back to the old way of working.

One focused trial beats ten vague AI projects

Scatter feels strategic. It usually means five half-started ideas and no measurable result.

Pick one job. Time-box it — often two to six weeks. Set one measurable metric: minutes per invoice, rework rate on quotes, handling time on sample emails. Put a success threshold and reassess date in the diary when you start, not after.

Kill criteria are not pessimism; they protect time and money. Willingness to stop if payback is still unclear is a strength. Sunk-cost bias is how weak tools become permanent habits.

A focused trial asks: did this one job get better enough, safely enough, to keep going? Ten vague “AI projects” answer nothing and consume everything.

What GO / WAIT / NO really means

You do not need a forty-question ritual to use the idea. You do need the posture.

GO means a scoped pilot is justified for this job — not “transform the business.” You have a named owner, a metric, and no hard blockers on data or the ability to leave or fall back.

WAIT means the job is promising but conditions are missing. Fix named gaps. Re-score in 30–90 days. Do not expand scope unpaid. Soft money maths never upgrades you past a continuity problem.

NO means starting would create avoidable risk or theatre. Fix foundations — logins, owners, data permission and minimisation — before any shopping list.

Bands describe this job, not your forever AI fate. You can fail one job’s readiness and still be sensible about the next.

Next Monday checklist

• Write one sentence for the job and who benefits.
• Name the sponsor, outcome owner, and deputy — real names, not “the office.”
• Book calendar time for review and a week-two check-in; list seats and hours.
• Set a time box, one metric, and two or three kill criteria.
• Where you are unsure (data, where prompts go, who can turn it off), score 0 and fix that before you buy.

The demo can still look great. Monday gets better when readiness is a person, a plan, booked time, one job, and the courage to be honest.

Decision aid thinking only — not legal advice, not a security audit, and not a vendor recommendation. Distinguishes operational facts businesses must face from vendor promises.

Drafted with assistant help; Mark Braddock owns the published wording. SovereignBase · Manchester, UK.