An AI problem statement template is a one-page form a small business fills in before talking to any AI supplier. Its six lines cover who has the problem, the job and how often it happens. They also cover the cost today in hours and pounds, the number that counts as fixed, and the limits. It names no tool, and every supplier gets the same page.
What is an AI problem statement?
An AI problem statement is a short written description of one business problem. It records the cost today and the result that would count as solved, before anyone chooses a tool. That keeps the answer open, so the project never starts from a product.
RAND interviewed 65 experienced data scientists and engineers for its study of why AI projects fail. The first root cause it lists is that people misunderstand or miscommunicate what problem AI is meant to solve.
The UK government gives public buyers the same advice. Its Guidelines for AI procurement tell them to provide a clear problem statement “rather than detailed specifications for a solution”.
The guidance is written for public bodies, but a 12-person firm buying its first AI tool has the same risk.
What goes in an AI problem statement template?
An AI problem statement template needs six lines, and each line must hold a fact you can check. Fill them in this order.
- Who has the problem. A role, such as “the office manager”, never “the team”.
- The job. One task, in the words a new starter would use, such as “turning survey notes into a written quote”.
- How often. Volume per week or month, counted from last month’s records.
- The number. Hours per week the job takes today, and what those hours cost in pounds.
- What fixed looks like. The target number, such as minutes per quote or days to reply.
- The limits. The spending ceiling, the systems already in use, and the data that must stay in-house.
Lines one to three describe the work. Line four prices it. Line five is the test any supplier’s answer must pass. Line six stops an answer you could never afford or use.
If line four is a guess, log the job for a week first. Our AI implementation plan template explains the five-day log, and the same baseline carries straight into that plan later.
Which pages did we compare, and what do they leave out?
We read five problem statement templates that rank for this search: CTO Academy, Geekyants’ AI consulting guide, Almbok’s use case template, MIT OpenCourseWare’s D.I.S. method and Lucid. All five ask who has the problem, what it is and what better looks like.
None of the five turns the cost of the problem into a spending limit. None of them says what to do with the statement once suppliers start replying.
Those two gaps matter most to an owner, because they decide how much to spend and who to buy from.
How do you turn the problem into a spending limit?
You turn the problem into a spending limit by pricing the hours the fix would save, then halving the yearly figure. Halving it means any spend inside the limit pays back within six months, the rule the AI ROI calculator guide uses.
Take an imaginary 12-person kitchen fitting firm. This is an illustration built for this article, not a client case. Assume the office manager writes 25 quotes a week from surveyors’ notes, at 35 minutes each. That is about 14.6 hours a week.
Price the hour at £25.14, the loaded cost per hour worked that the ROI guide builds from ONS median pay.
At that rate the quoting job costs about £367 a week. Over 46 working weeks, the figure left after statutory holiday, that is about £16,865 a year.
Now set the target. Assume the firm would call it fixed at 15 minutes per quote. Then 25 quotes take 6.25 hours, which saves about 8.3 hours a week.
At £25.14 an hour, the ONS-based rate, over 46 weeks the saving is about £9,637 a year. Half of that is about £4,818. So the firm’s year-one spending limit, set-up included, is roughly £4,800.
Any proposal above that limit cannot pay back in six months, even if it hits the target in full.
What does a filled-in AI problem statement look like?
A filled-in AI problem statement for the imaginary kitchen fitting firm fits in one table and names no product. Every figure comes from the worked sum above.
| Line | Entry for the kitchen fitting firm |
|---|---|
| Who has the problem | The office manager |
| The job | Turning surveyors’ notes into a written quote |
| How often | 25 quotes a week, counted from last month |
| The number | 35 minutes a quote, about 14.6 hours and £367 a week (at £25.14 an hour) |
| What fixed looks like | 15 minutes a quote, same accuracy, quote sent the day after the survey |
| The limits | Year-one spend under £4,800 including set-up; prices stay in the existing quoting spreadsheet; no customer data leaves the UK without a written contract |
The “what fixed looks like” line includes accuracy on purpose. A quote written in five minutes with the wrong worktop price costs more than it saves.
Which words should an AI problem statement never contain?
An AI problem statement should never contain a product name, a type of tool or the word “automate”. Words like “chatbot”, “agent”, “Copilot” or “AI-powered” answer the question before it is asked.
Run a quick check before you send the page. Read each line and cross out any word that describes a solution. If a line makes no sense once the word is gone, it was a wish, and the job needs describing again.
The government’s Service Manual on discovery gives its teams the same instruction. When handed a ready-made solution, they should reframe it “as a problem to be solved” before doing research.
Bedrock AI’s view is that a problem statement with a tool’s name in it is a shopping list. The owner has already decided, and every supplier will simply agree.
How do you use the problem statement with suppliers?
You use the problem statement by sending the same page to two or three suppliers and judging every reply against lines five and six. Identical briefs make the replies comparable, which a sales call never does.
Read each reply with four questions.
- Does it restate your number? A supplier who quotes your 14.6 hours back has read the page.
- Does it stay under the limit? Count set-up, training and the first year of fees together.
- Does it say how it reaches the target? “Saves time” is not an answer to “15 minutes a quote”.
- Does it mention a cheaper fix? A better spreadsheet template might solve some of the problem. A supplier who says so is worth keeping on the list.
A reply that ignores the page and pitches a product is useful too. It tells you that supplier will not fit their work to yours.
Before signing anything, run the shortlist through the AI vendor due diligence checklist. The problem statement tells you whether the answer fits. The checklist tells you whether the supplier is safe.
How is a problem statement different from an implementation plan?
A problem statement describes the problem before any tool is chosen, and an implementation plan runs the trial after one is chosen. The statement goes to suppliers. The plan stays inside the business.
The two share a baseline, so the work is not wasted. Line four of the statement becomes the baseline line of the one-page implementation plan, and line five becomes its stop rule.
If you cannot decide which job to write up first, score your workflows on volume and cost of delay. The highest score is usually the right first page.
FAQ
Can ChatGPT write an AI problem statement for me? ChatGPT can tidy the wording of an AI problem statement, but it cannot supply the numbers. MIT OpenCourseWare’s D.I.S. course says a generative tool can check how a statement is structured but not whether it is correct. Count the volume and time yourself.
How long should an AI problem statement be? An AI problem statement for a small business should fit on one page, in six lines. MIT OpenCourseWare’s method aims for 100 to 150 words. If yours runs longer, it probably covers more than one job and should be split.
Should I put a budget in an AI problem statement? Yes, an AI problem statement should include a spending limit worked out from the cost of the problem. Without one, suppliers price to what they think you will pay. Halving the yearly saving gives a ceiling that pays back within six months.
What if the problem is not worth solving with AI? A problem statement that shows a small number has done its job. If the job costs a few hundred pounds a year, no supplier can fix it inside the limit. Keep the page and write one for a bigger job.
Get ahead with AI by getting The Business Brain | Would rather talk it through? Book a free 15-minute call