An AI implementation plan template for a small business fits on one page. It names one job, one owner and today’s baseline number. It also lists the data the tool touches, the rules for staff, a checking step, a trial with a stop rule, and a review date. If you cannot fill in the baseline, you have not diagnosed the job yet.
Why do most AI implementation plan templates fail a small business?
Most AI implementation plan templates fail a small business because they begin with phases and tools, and never with a measured problem. They run from assessment to pilot to scaling over six to twelve months. A 12-person firm has neither the staff nor the time.
The result is a plan that looks tidy and answers nothing. Nobody wrote down what the job costs today, so nobody can say whether the tool worked.
The UK government’s AI Adoption Research shows how loose planning is. Among businesses planning to adopt AI in the next year, 47% had no specific budget set. Only 34% of those planners said they felt ready.
Bedrock AI’s view is that a plan is a list of things you have already measured. Phases come later and take one paragraph, which is the last section below.
What goes into a one-page AI implementation plan?
A one-page AI implementation plan has eight lines, and each line must hold a fact or a decision, never an intention. Fill them in this order, because each line depends on the one before.
- The job. One task, named so a new starter would recognise it, such as “chasing invoices more than 14 days overdue”.
- The owner. One named person, who uses the tool themselves and reports at the review. A committee does not count.
- The baseline. Hours per week and cost per hour for that job today, taken from a week of logging, not a guess.
- The data. Every kind of information the tool will see, such as client names, bank details or staff records.
- The staff rules. Which tool and which account staff use, and what must never be pasted in.
- The check. Who reads the output before it leaves the building, and how long that takes.
- The trial. Start date, length (four weeks is enough for most office jobs), and the stop rule written in numbers.
- The review. A diary date and the one figure that gets compared with the baseline.
Lines one to three are the diagnosis. Lines four to six are the safeguards. Lines seven and eight are the decision. The order is the point: nobody should pick a tool before line three is filled.
How do you fill in the baseline when nobody has timed the job?
You fill in the baseline by asking the person who does the job to log it for five working days. A simple note of start time, finish time and what was done is enough. Do this before any tool is discussed.
Staff estimates run wrong in both directions. People remember the annoying minutes and forget the quick ones, so a logged week beats a remembered one.
The baseline also settles the skip question early. If the whole job takes 40 minutes a week, no tool will repay a trial. Our guide to scoring your workflows ranks jobs by volume and cost of delay so you pick the right one first.
For turning hours into pounds, use the method in the AI ROI calculator guide. It prices an hour of staff time properly and discounts the saving honestly.
What does a filled-in AI implementation plan look like?
A filled-in AI implementation plan for an imaginary 11-person surveying firm shows a modest saving and a clear decision. This is an illustration built for this article, not a client case, and the figures are round to keep the sum visible.
| Plan line | Entry for the surveying firm |
|---|---|
| Job | Chasing client invoices more than 14 days overdue |
| Owner | The office manager |
| Baseline | 3 hours a week, costed at £25 an hour worked |
| Data | Client names, invoice amounts, no bank details |
| Staff rules | One company-managed account, no bank details pasted in |
| Check | Office manager reads every draft reminder, about 20 minutes a week |
| Trial | Four weeks from 1 December, stop if net saving is under 1 hour a week |
| Review | 5 January, compare weekly hours with the baseline of 3 |
After 20 minutes of weekly checking, the firm plans on a net saving of 1.5 hours a week.
At £25 an hour, using the costing method in the ROI guide, 1.5 hours is £37.50 a week. Over 46 working weeks that comes to about £1,725 a year.
Assume a tool seat at an illustrative £15 a month, which is £180 a year, plus six hours of set-up at £25, which is £150. Year-one cost is £330, so the net saving is about £1,395.
Payback comes in a little over a month. The plan also shows the limit of the job. A saving under £2,000 a year is worth taking because it is cheap and quick, but it is not a transformation.
What should the stop rule say?
The stop rule says, in numbers, the result at which the trial ends. Write it before the trial starts, because nobody votes to stop a tool they have already learned.
A good stop rule has two parts. The first is a minimum saving, such as the one hour a week above. The second is a failure condition, such as any reminder sent without being read, or any client data pasted outside the agreed account.
The stop rule protects the owner from sunk-cost thinking. A tool that saves 20 minutes a week and costs £15 a month is a poor trade, however friendly it feels.
If the trial passes, the next step is a three-week test on your own work. The vendor proof of concept framework sets out what each week must prove before you sign anything longer.
Who needs to sign off the plan before the trial starts?
The owner of the business signs off the plan, and the person running the job signs that they can do it. Two signatures are enough for a firm of this size. Anything heavier invites the slow, vague planning that sinks small projects.
Lines four and five carry the risk. If the tool will see client records or staff data, run the AI risk assessment template first and put its result on line five.
Staff also need to know the rule in writing. A short acceptable use policy turns line five into something they can follow without asking.
If nobody will read the output each week, line six is empty and the plan is incomplete.
How long should an AI implementation take in a small business?
A first AI implementation takes about eight weeks in a small business, from baseline week to review. One week of logging, one week of set-up, four weeks of trial and a review week cover it. Longer timelines usually mean the job was too big.
Start with one job. When the first plan passes its review, copy the page, change the job, and start again. A second plan costs an hour because lines two to six are mostly reused.
Phased roadmaps run assessment, pilot, rollout and scaling. This plan covers the first two: lines one to three are the assessment and lines seven and eight are the pilot. Rollout waits for a passed review.
Resist adding a second job during the trial. Two changes at once make the baseline comparison meaningless.
If the review fails, that is a useful result. The plan cost a few hours and prevented a year of subscription. If you want your business context ready for the trial, The Business Brain is a £27 download built for that.
FAQ
What is an AI implementation plan? An AI implementation plan is a written page that names the job AI will do and who owns it. It also records the baseline cost today, the data involved, the staff rules, the checking step and the trial stop rule. It exists so the result can be compared with a measured starting point.
How long should an AI implementation plan be? An AI implementation plan for a business of 5 to 20 people should be one page of eight lines. A longer document suggests the job has not been narrowed enough. Phased roadmaps of many pages suit larger organisations with a dedicated team.
Do I need a budget before I write the plan? You need a rough ceiling, and the baseline gives you one. If the job costs £4,000 a year in staff time, spending £4,000 on a tool makes no sense. Write the ceiling on line three, then let the trial show whether a smaller spend works.
Who should own an AI implementation in a small firm? The person who does the job day to day should own it, with the business owner signing off. Owners who delegate the plan to a supplier lose the baseline, and the baseline is the part nobody else can write.
What should I do if the trial fails the stop rule? Stop the tool, cancel the seat and keep the page. Note which line was wrong, usually the job or the baseline, and choose a different job. A failed trial that cost four weeks and one seat is a cheap answer.
Get ahead with AI by getting The Business Brain | Would rather talk it through? Book a free 15-minute call