AI for the right reasons
Three in four UK businesses using AI say it hasn't changed their revenue yet. Usually nobody pointed it at what was holding the business back. I only recommend AI when it moves that one thing, or protects what you sell. We measure where you start, and the work runs in phases with a decision at the end of each one.
Jump to: The two questions · How the work runs · What people ask first · What you get
AI has to be the right answer, aimed at the ONE THING that will make a difference.
In two minutes it shows you whether AI could replace what you sell, and whether it should be a priority at all.
Every business should answer two questions before it spends anything on AI.
Can AI already produce what you sell?
What is holding the business back?
1. The AI Exposure Test
2. The Constraint Locator
The share of your gross profit AI can already reproduce, the things it can't, and roughly how long you have to react. If that share is large, start with what you sell before worrying about how you run things.
Finds the one thing holding your business back. There are eight places it can sit, and I start with your market and work inwards. If AI can't move that one thing, an AI project won't move the business.
Once you have both answers, we can decide whether AI belongs in your plan at all.
Most AI advice skips both questions, which is why so much of it ends in a programme that never pays back.
What the evidence says
Most UK businesses using AI are getting something from it. Three in four say their people are more productive. But 77% say it hasn't changed their revenue yet, and only 12% have seen revenue go up.¹
Bigger companies hit the same wall further along. In one study this year, fewer than a third had got AI past the pilot stage, and only 13% were on track with their plans.²
So the tools work, and people use them. What's missing is the link to the numbers that matter, and that link starts with knowing what's holding the business back.
¹ UK Government, AI Adoption Research, February 2026. A survey of 3,500 UK businesses.
² BearingPoint study, October 2026.
I won't sell you an AI programme.
I only recommend AI when it's aimed at a real problem: the thing holding you back, or the thing that stops you being replaced.
Three ways businesses lose money on AI
None of them is a technology problem. All three happen before anyone opens a laptop.
One
THE solution CHOOSES THE PROBLEM
Someone buys a tool, then looks for a use for it. They usually pick whatever's easiest to automate, which is rarely what's holding the business back.
Two
THE WRONG THING GETS FASTER
Someone automates a task they don't fully understand, and the same mistakes happen faster and more often. Or they automate the part customers valued, and save time but lose the reason customers chose them.
Three
NOBODY MEASURES BEFORE OR AFTER
If nobody measured how things were before, nobody can show what changed. Someone reports hours "saved", nobody can say what they were worth, and AI becomes another cost on the P&L.
How I run the work, starting in the right place
Six phases, one area at a time
I scope each phase on its own, at a fixed fee agreed once we know the work, with a defined end. Each finishes at a decision point where you can stop and keep everything produced so far.
The four questions we ask of every change that touches a customer.
1. Does it take away from the experience they came for?
2. Is it faster, but feels less cared for?
3. Does it remove a person where trust matters?
4. Does it make the business feel colder?
A restaurant can take orders faster and lose the welcome people came for.
Before we start
A short call to check we're a good fit. If we are, I send you the Evidence Request and an engagement letter that says exactly what happens to your data. You decide: whether to go ahead.
Phase by phase
You make the call at the end of every phase.
2. Assess
I look at how ready the business is across eight areas, from data and processes to skills and governance, and rank the places where AI could add value. You get it all in the Key Findings Report.
You decide: which areas deserve a closer look.
4. Design
We redesign the process and work out the data it needs. You get a plan for the people affected, a brief for what the AI has to do, and how we'll know it worked, with someone in your business responsible for each measure.
You decide: whether to invest in building it.
6. Sustain
We check the results at 30, 90 and 180 days against the numbers we measured at the start, then move on to the next area. At least once a year we re-run the AI Exposure Test and the Constraint Locator, because what AI can reproduce moves and so does the one thing.
You decide: what comes next, if anything.
3. Map
We pick one area, the one that moves your one thing, and measure how it runs today before anything changes. Then we trace what your customers value back through the teams and workflows that deliver it. By the end you have a business case: what changing it would be worth.
You decide: whether the business case justifies designing it.
5. Deliver
We choose how to build it and who with, bring the partner on board, then build it and pilot it on part of the work while the rest carries on as before. Once it goes live we embed it properly: new procedures, training, and the old process switched off.
You decide: whether it's ready to go live, and later whether to accept the handover.
If you stop at any decision point, you keep everything produced so far, and you can take it to anyone.
You can stop at any point
Four possible answers
At the first decision point, once you've completed the AI Exposure Test and the Constraint Locator, the answer is one of these. Two of them come down to "AI isn't a priority yet".
1. AI can already produce a large share of what generates your gross profit. If so, the priority is to redesign your offer, price and position so they hold up against AI. The work goes into the value you deliver to customers rather than your back office.
2. AI can move what's holding your business back, so we go on to a deeper assessment.
3. Fix it without AI. The one thing is real, but AI isn't the cheapest way to move it. You get the plan, and AI waits.
4. Your team already use AI well, and what you need is a policy that protects your IP and customer data. You get a short policy, an audit of the tools in use and clear rules on data. It takes days.
Time saved is not a benefit. It becomes one when it turns into cash out of the business, capacity you sell, or quality you can measure. Until then it is a feeling.
“Mark's challenge sharpened the proposition and saved me from a few expensive wrong turns.”
Dr Mick O’Connor, CEO, HAELO
1. Diagnose
I interview your key people, then run a Decision Day or The Fix with your leadership team working through the AI Exposure Test and the Constraint Locator. By the end you know how much of what you sell AI could already reproduce, the one thing holding the business back, and whether AI should be a priority at all.
You decide: what to do next.
Value first, process second, technology third
Before we automate anything, we work out what it's for.
Start from the customer. What your customers receive, and how it leaves them feeling: confident, looked after, sure of the value received. Then which teams deliver each of those, and which processes they depend on. We stop work that serves neither, rather than automating it.
Redesign, then automate. Copying today's process into software keeps all of its friction. The people closest to the work usually know the better version. We ask them, and we design that.
Human-led. Every AI workflow has a named person who is accountable for what it does, and written limits on what it may do without approval. A human somewhere in the loop is not enough. Someone has to own the outcome.
Six levels, chosen step by step. Each step of a workflow gets its own level. Most land in the middle. Letting AI do more on its own usually adds cost, and rarely adds value.
-
0 Stays human. Judgement, empathy, trust.
-
1 Stop doing it. The cheapest automation there is.
-
2 Simple rules. No AI needed.
-
3 AI drafts, a person decides.
-
4 AI acts, a person approves.
-
5 AI acts within limits, a person monitors.
THE RETURN
We measure every gain against where you started
Six things are in place before you spend any money on building. Together they are what lets anyone understand the result for themselves.
SIX THINGS WE PUT IN PLACE FIRST
-
Baseline first. Volumes, how long each job takes, cost per job, error rates and win rate, measured before anything changes, with the definitions agreed.
-
What the saved time turns into. Every claimed benefit says whether it becomes cash out, capacity you sell, or measurable quality. If we can't say, it doesn't count.
-
When we'd stop, agreed in advance. We write down, before any money goes out, what would make us stop.
-
Tested on real work. We check the AI against a set of your real cases before go-live, then pilot it on part of the work and compare that with the part left alone.
-
A named owner for every number. Someone in your business, not me, owns each benefit and reports it.
-
30, 90 and 180 days. Results reported against the baseline, including the ones that disappoint.
What people ask first
-
We already use ChatGPT or Copilot. Is this for us? Probably. Using AI tools and getting a return from them are different things. If your people already use them, the first useful step may be the fourth answer above, safe use only: clear rules on what goes into which tool and how you use the output.
-
Do you build the AI? No. I design what needs building, help you choose who builds it, and stay on your side while they do. Often the right answer is to switch on something you already pay for.
-
How long before we see a return? It depends on the area, and the business case says when before you commit any money. A small workflow on tools you already own can move within weeks. A bigger redesign takes months.
-
What if the answer is no AI? Then you get a plan for the one thing that doesn't involve AI, and you have saved the cost of finding out the hard way.
-
Is our data safe with you? Before any work starts you get a short note setting out what I collect, where it is stored and processed, which tools touch it, and when it is deleted. I record interviews for accuracy.
-
Who is this for? Owners, founders and business leaders of services businesses with £3m to £50m of revenue and 30 to 200 people, where the person who can make the decision is willing to be in the room.
WHAT YOU GET
What lands on your desk
Named documents, not "a report". I write each one to be handed to the accountable person, and each one is yours to keep.
-
The Verdict. Phase 1, four pages, within three days. What the room agreed, what's holding you back and how sure we are, how much of what you sell AI could replace, and the first ninety days with an owner against every line.
-
The Decision Paper. End of Phase 1. Which of the four routes, the evidence for it, and what happens next, including when to look again if the answer is "not now".
-
The Key Findings Report. End of Phase 2. Where you stand, what is already working, where the value is, the risks and blind spots, quick wins set against longer work, and the areas worth a closer look.
-
The Value Map and Workflow Map. Phase 3. What your customers value, which teams deliver it, and how the chosen workflow runs today and should run tomorrow, with a level set for every step.
-
The Business Case. Phase 3, then firmed up in Phase 4. Baseline, target, a range of benefit, cost, where the hour goes, kill criteria and the person who owns the result.
-
The Implementation Design Pack. Phase 4. Process, data, people, the scope of work for the AI, success measures and the rules it operates under. Detailed enough for a supplier to quote against.
-
The Partner Brief and Scorecard. Phase 5. What you need, written so the chosen supplier delivers against your requirements rather than their product, and a scorecard tested on your own cases.
-
The Benefits Report. Phase 6, at 30, 90 and 180 days. Each measure against its baseline and target, what changed, what didn't, and what to adjust.
If you stop at any decision point, you keep everything produced so far, and you can take it to anyone.
I don't build it. I make sure the supplier you choose builds it properly. I can do that because I don't work for them.
Most AI money goes to software companies and implementation partners. My job is to make sure what you buy matches what you need, is implemented the way you want and then does what it promised.
We start with what you already own
Plenty of AI already sits inside software you already pay for and don't use. We check that before anyone writes a proposal.
We test suppliers on your own work
Shortlisted suppliers demonstrate on a set of your own real cases, and are scored against the same criteria, written down in advance.
On your side of the table
Keeping you on the right side of the rules
During delivery I check progress against the design, the budget and the evidence, and I say so when they drift apart.
Data protection, the UK's updated rules on automated decisions, the EU AI Act if you sell into the EU, where your data is stored and processed, and consultation with staff whose roles change. I build all of it into the design work rather than bolting it on afterwards. I'm not a lawyer, and where it matters I'll tell you to ask one.
“Calm, straight and on your side.”
Phil Clements, CEO, Finspector
What this asks of you
This is what I'll need from you. If you can't give it, better we both know now.
A sponsor who decides
One senior person who owns the outcome and makes the call at each decision point. Not delegated.
Your people's time
Interviews, mapping sessions and, later, the pilot. The people closest to the work know how to make it better.
Honest numbers
Willingness to stop
A baseline means measuring things you may never have measured. If the numbers don't exist, we build them first.
Sometimes the right answer at a decision point is ‘not yet’. The method only works if you allow that answer.
And what I won't do
-
Start with a tool and look for a use for it.
-
Recommend AI when it can't move your one thing.
-
Automate something nobody has measured.
-
Automate away the part of the work your customers come back for.
-
Leave you with a system nobody in your business owns.
-
Mark my own homework. The results are your numbers, against your baseline.