Making Work Flow
One person using AI well makes one person faster. I build the workflow underneath — the enquiry classified, checked against your records and drafted before anyone's opened the inbox. Someone reads it and presses send. That happens whether or not your best person is in today.
£2,500 + VAT
Fixed price
3 workflows
Examined end to end
~3 hours
Of your business's time
10 days
Payment to review call
Somebody has already been in to talk to you about AI. A deck, some phrases about transformation, a pilot that ran for a few weeks and then stopped being mentioned. Or a monthly retainer with no end date and nothing you could point at when it finished.
The common thread is that nobody stayed long enough to find out whether anything changed. I look at three workflows, cost them using your figures, and come back at week three to check whether people are still doing it the new way.
A common objection
Good. So do I, all day. Here's the difference.
Using ChatGPT makes you faster at your job. It doesn't change how an enquiry gets from your inbox to a quote, how a report gets built every month or what happens when the person who worked out the trick is on holiday. That's a process. Prompting doesn't fix a process.
One or two capable people are getting a lot out of it and nobody else is — the gain never reaches the P&L. It works well until the person who worked out how to do it leaves. And somebody is pasting customer information into a personal account, and nobody has decided whether that's alright.
The difference is that I build these systems, not just prompt them.
A live trading system running on my own money, where the decisions that cannot be wrong are hard-coded and the AI works around them. An editing system that checks writing against a defined voice, because "it doesn't sound like us" is the real reason AI drafts get thrown away. A delivery forecasting tool that other people install and run on their own machines, not just mine.
Knowing where a model must not be allowed to decide is most of the job. That part doesn't come from prompting.
This is about the business, not the individual.
Track record
3 months → 2 weeks
Production deployment lead time
Hargreaves Lansdown
90% reduction in lead times · 80% fewer escaped defects · annual to weekly releases
Delivery across a department of 80
Flashtalking
£270m programme
Delivery lead, Customs Declaration Service — the systems handling UK imports and exports after Brexit
HMRC
None of that was AI. All of it was the same question — where does the work stall and what happens if I take the stall out?
The mechanism
Two examples. Neither of them is prompting.
Now
It sits in the shared inbox until somebody opens it. They read it and work out what it is. They decide who should handle it. They check whether this customer has bought before. They dig out a similar quote from three months ago. They write a reply. Two days later somebody asks whether anyone ever came back to the bloke from Halifax.
After
Another example
Month-end reporting. Now: someone spends a day and a half pulling numbers out of three systems into a spreadsheet, formatting it and writing the commentary. It goes out late and nobody reads all of it. After: the numbers pull themselves on the first of the month and a draft commentary is written against last month's figures, flagging what moved and by how much. The person who used to lose a day and a half spends 40 minutes checking it and adding the things the numbers don't say — which is the part they were always good at.
It runs when the email arrives, not when someone remembers.
It already has the customer history, the last three quotes and the price list.
It happens the same way on a Tuesday in August with half the office off.
It leaves a record — so you can count what came in, what went out and how long it took.
It stops at a person. Every time.
Prompting is invisible to management. You can't count it, so you can't prove it worked or decide whether to do more of it.
What you get
Guarantee
Before I write the report, you sign off the figures it rests on — your volumes, your hours, your rates. If those signed figures can't support at least five hours a week worth fixing, you get the fee back.
Your numbers, in writing, before anything is built on them. Neither of us has to argue about whether a saving is real.
Three shapes of work. Most of what I find falls into one of them.
Something arrives: a form, an email, a message taken over the phone. It gets classified, checked against your records, and a reply drafted in your language, waiting for someone to approve and send.
The same document rebuilt from scratch every time, usually by the person who can least afford the afternoon. Instead: assembled from your previous work, your pricing and the specifics of this job, as a draft for you to correct.
The monthly pack that eats a day and a half. The numbers assemble themselves and a first-draft commentary flags what moved and by how much. The person who used to build it spends forty minutes checking it.
Every one of them stops at a person. Nothing sends, quotes or commits on its own.
Who it's for
The assessment is for businesses of roughly 15 to 100 people, anywhere in the UK, in any sector. Under ten people, you probably do not need me — you need an afternoon and a free trial. Above a hundred, there will be procurement, an IT policy and information governance; those are real, and they need someone with more hours than I have.
If there is uncertainty about whether it is worth doing, the free 20-minute call is the place to find out.
I take four to eight clients a year. That's what one person can do properly, and it's why the scope is fixed rather than open-ended.
Limits
Equality Act exposure. Insurance almost certainly will not cover a discrimination claim.
PECR problems. Burns domain reputation. Small return.
A wrong estimate loses money on real work. The method comes attached.
I've built systems where a wrong number costs real money. That's exactly why I won't build you one that prices your jobs.
Who's behind it
I've spent 30 years building and delivering software — agencies, then government. That's where the assessment comes from: knowing what actually takes effort to build, what sounds simple and isn't, and where AI genuinely saves time rather than adding noise.
Read the full background
Two minutes
Four questions. An honest answer — including if the answer is no.
Sample result
6 people · Handling incoming enquiries · Mostly spreadsheets · One or two people, informally
A business your size usually doesn't have enough repeated process to justify £2,500, and you're probably doing most of it yourself. Spend an afternoon and twenty pounds on a paid AI subscription first. Come back when the same job is being done by three different people in three different ways.
That is a real result the questionnaire gives. It turns away more people than it qualifies.