How to run an AI opportunity review in your own business 

An AI opportunity review is a short, structured look at how work flows through your business today, so you can find the two or three places where AI or automation would save the most time or money for the least effort and risk. It ends with a ranked list and not a tool recommendation.

What is an AI opportunity review?

It's a way of choosing where AI goes in your business, based on how your work really happens and not on what a product page promises.

Most owners are told they should use AI but very few are told where. A review answers that question for your own business, and it starts from your work rather than from a tool. In my opinion the order is the whole game, because a tool you pick first will go looking for a job to do and it usually finds the wrong one.

You can run the method below yourself if you have an owner or manager who can make decisions and at least one person who does the daily work. Set aside a few focused sessions and keep real examples in front of you rather than memory.

1. List the work that repeats every week

You start by writing down every task that happens again and again in your business. Ask each person what they did last week that they'll do again next week, and include the small tasks too because small tasks add up.

A worked example from a small trades business might read like this - answer calls, book jobs, send quotes, chase unpaid invoices, enter supplier bills and send the weekly summary to the owner. Next to each task you note who does it and roughly how often it happens. Do not try to measure it precisely yet.

2. Follow one enquiry and one invoice from start to finish

Pick one real enquiry and one real invoice. Trace each one from the moment it arrives to the moment it's finished, and write down every system it touches and every person who handles it. This shows you how work actually flows, which is often not how the procedure says it flows.

Use something from the last few weeks so your people still remember the details. My rule of thumb is to ask the person involved to show you rather than describe it, because you'll spot steps that they no longer see themselves.

3. Mark every re-key, wait and chase

On each traced job you mark three kinds of waste. A re-key is any place where someone types information that already exists somewhere else. A wait is any place where work sits until a person gets to it, and a chase is any place where someone has to remind another person to act.

• A re-key is a job booked in one tool and typed again into an invoice.
• A wait is an enquiry sitting in an inbox until the owner is free.
• A chase is someone calling a customer about an unpaid invoice.

These marks are your candidate list, and each one is a place where a task might be automated or where AI might help. Many will turn out to need simple automation and no AI at all, and I count that as a good result.

4. Rank by value and effort

Score each candidate on two things - value and effort. Value is how much time or money it saves you, and effort is how hard it is to build and keep running. Use a simple low, medium and high scale and don't pretend you're more precise than you are.

The top left box is where you start, because these are quick wins that build your confidence. The bottom right box is where projects go to die, so you leave those out however exciting they sound.

5. Check the data and the risk

For each top candidate you ask two questions, and the first is whether your data is good enough. Is it complete and readable and sitting in one place? Poor data will sink a project that looked easy. The second question is about risk, so think about what a wrong result would cost you and who would see it.

If client data is involved then check where it would go, who could read it and which tools would handle it. Any legal questions about privacy stay with your own advisers. A candidate with poor data or high risk drops down your list until those problems are solved.

6. Pick two or three and define done

You choose the top two or three, and I do mean two or three and not ten. For each one you write a single sentence that says what will be true when the work succeeds, who will check it and what you will measure.

For example, every after-hours enquiry gets a reply and a record in the CRM, the owner checks urgent ones each morning and the team counts missed enquiries before and after. A sentence like that turns a vague wish into something you can build and judge.

What does a full example look like?

Take the same small trades business, where steps one to three gave you a candidate list of five items. Here's how I'd expect them to rank, though the ratings are not data and only show the method.

The owner picks the first two, since neither needs a large budget, both remove daily work and both are easy to measure. The last item sounds exciting but it would take a lot of effort for a result nobody could check, and I'd drop it without a second thought.

Who should take part in the review?

You include the person who decides and the people who do the work. The owner knows what matters and the people doing the work know where it goes wrong. If you leave the second group out then you'll rank the wrong items first.

Some staff worry that a review is really a search for jobs to cut, so make it safe for people to speak and say clearly that the aim is to remove repetitive admin. Then ask them which tasks they'd happily hand over.

What traps should you avoid?

• Starting from a tool, when you should start from the work and only then look for a tool.
• Reviewing everything at once, when one area such as sales or finance is enough.
• Counting wishes as savings, when your own records are the only figures you can trust.
• Ignoring the people, because a good plan still fails if the users were never asked.
• Stopping at the document, because a review that does not lead to a build is only a report.

Keep the output short and share it with the people who took part, because they'll want to see what happened to their input. You repeat the review when your business changes, such as after a new service, system or change of staff. The first round teaches you the method and later rounds go quicker.

What does a good review produce?

Your review should produce a short document that fits on a few pages, and I'd want anyone in your business to be able to read and understand it.

• A list of your repeating tasks, with who does each one.
• Your traced jobs, with every re-key, wait and chase marked on them.
• A ranked list of candidates with their value and effort scores.
• Notes on data quality and risk for the items at the top.
• Two or three chosen projects, each with a one-sentence definition of done.

It shouldn't lead with a product recommendation, because tools come after you know the task. It should also say plainly where AI is the wrong answer and what you can do instead. Sometimes a fix to a spreadsheet beats an AI project, and I'd tell you so without blinking.

What happens after the review?

Take your chosen projects and scope each one in detail. You decide who builds it, which tools it connects to, who checks the output and when you'll measure the result. Then build the first one small and look at your numbers before you start the second, because that habit saves you from the expensive mistakes.

Frequently asked

People ask me how long a review takes, whether they need a consultant and what to do afterwards, so the answers are below. My short version is to keep it brief, use real work and act on your top two or three items. Hope this helps!