The AI opportunity workbook: find, score and shortlist

Tip: think of the task you dread, the one that causes complaints, or the one you'd hire someone tomorrow to take off your plate.

Step 3: map the worst one, step by step

Take the worst of the three and pick one outcome, such as “a customer gets a quote”. Write one sticky note per step: what happens, who does it, how long it takes. Use last Tuesday's version, not the one in the manual. Something you thought was five steps usually turns out to be fifteen.

Then put a red dot on every note that shows one of the six signs of drag:

• Typing it twice: the same details keyed into a second system.
• Chasing: payments, timesheets, signatures, replies.
• Hunting: “Where's the latest drawing? What did we charge last time?”
• Checking by hand: matching invoices to delivery notes, line by line.
• Slow to respond: the enquiry that waits until someone's free.
• Only one person knows: the spreadsheet only Sue understands, and Sue's on holiday.

Exercise 2: Score your top three 20 minutes

Before you spend a penny on AI, score each of your three processes from 1 to 5 on four questions. All four point the same way: high is always good. Add them up for a total out of 20. The higher the total, the sooner you do it.

Be honest on Feasibility. It's the score everyone flatters. Ask yourself: where does that data actually live right now?

How to read your result

• High everything (15 to 20): your quick win. Start here.
• High impact, low feasibility: worth doing, but fix the data or process first. Rushing this is where businesses waste money.
• Low impact, high ease: a nice-to-have. Fine to do, but don't let it distract you from the real prize.
• High impact, low ease, uncertain: the trap. Big, hard and unproven. Get expert eyes on it before you commit budget.

High impact with low feasibility isn't a no. It's a not yet.

A worked example: two ideas, scored honestly

• Chasing unpaid invoices: Impact 4 (it's cash flow), Frequency 5 (every week), Feasibility 4 (the data is already in your accounts system), Ease 4 (simple automation). 17 out of 20. Start here.
• Predicting which customers will leave: Impact 4, Frequency 3, Feasibility 2, Ease 2. 11 out of 20. Not yet. The data is scattered across systems and nothing is labelled. Sort the data first, then build.

Same business, two good ideas. The test tells you the order, and the order is where the money is.

Go deeper on your number one: the Five Whys

Take your highest-scoring process and ask “why is this a drag?” five times. Most of the time the real problem isn't the task itself. It's a broken handoff, missing data, or nobody owning it. If the root cause is a missing process or messy data, fix that first. AI on top of a mess just makes the mess faster.

Exercise 3: Put a number on it 10 minutes

Do this for your number one opportunity only. Rough numbers are fine: scale matters, not precision.

Also add anything else it costs: errors and rework, sales lost by being slow, staff you can't move onto better work.

A worked example, not a promise: a task done 5 times a week (260 times a year), taking 2 hours each time, by someone costing £30 an hour fully loaded, is £15,600 a year on one recurring task. If AI takes even half of it off your plate, that's about £7,800 a year back, every year, from one task.

Finish with one line describing the shape of the solution: “AI could ... so that my team ...”

Shelf or rewire: can I do this myself?

You'd happily put up a shelf. You wouldn't rewire the house. Not because you can't, but because the stakes are high and a specialist does it right, once, safely.

• The shelf, start it yourself: small and low-risk if it goes wrong, an off-the-shelf tool already does it, no connection to your core systems. Drafting, summarising, meeting notes, asking questions of your own documents. Go and do it on Monday.
• The rewire, get a specialist: business-critical, or touching money, customers or compliance. It needs building and connecting to your systems and data, and a wrong answer is expensive or hard to catch. Quote automation, customer-facing agents, anything with your live data. Scope it properly before you spend.

Two questions for any vendor: where does my data go, and how does your system fail safely? If they can't answer both, walk away. And on anything customer-facing, financial or compliance-related, keep a person in the loop.

Your next move

Not homework. Momentum, while it's fresh. Write down three things:

• This week: one thing you'll start yourself. Small, low-risk, off the shelf. Put it in the calendar or it won't happen.
• This month: one thing you'll explore further. It needs a conversation with your team, a look at the data, or a question answered by a supplier.
• When you're ready: one thing you'd want expert help with. The big number. Business-critical, connected, worth doing properly.

If your number is big, that's exactly where Discovery & Diagnosis starts. For a deeper look at scoring, read the Earns Its Keep test.