Four hours of checking became fifteen minutes. Nine times the applications, handled by three people.
For six years I've worked with teams in big companies on the same problems you have: too much admin, too few people, and customers you can't afford to lose. We found the jobs people did by hand and got software to do them, first the ordinary kind, now AI. Here's what that looked like, and what it means for a business your size.
15 min
a morning's checking, down from four hours
9x
the applications three people could handle
3 mths
the most it took that automation to pay for itself
100%
of customers kept through a full rebuild
Five jobs people used to do by hand
Each one says what it was like before, what we did, and what the same idea looks like in a business your size.
The four-hour morning job that now does itself
- Before
- Every morning an analyst spent four hours working out which documents had arrived overnight and updating a spreadsheet by hand.
- How we solved it
- We set AI up to read, sort and file every document the moment it arrives, and keep the spreadsheet up to date on its own.
- The result
- Four hours a day down to fifteen minutes. The team starts each morning knowing exactly who to chase, and the work runs around the clock.
In your business
Think of every order, invoice or supplier email someone in your team sorts and logs by hand.
Under the bonnet: Automated ingestion, classification and filing of due diligence documents, with a live tracker.
Rebuilding the whole product without losing a single customer
- Before
- Hundreds of businesses had built their own software on top of ours. Rebuilding ours meant every one of them had to change theirs, and any of them could have left.
- How we solved it
- We made the new version so much simpler to use that switching was worth their effort, then moved everyone across in four months.
- The result
- 100% of customers kept.
In your business
A website rebuild is the same job at a smaller scale: the new one has to be better, and nobody who already buys from you can be lost on the way.
Under the bonnet: Full API replatform and redesign, carrying hundreds of thousands of card transactions a month.
Growing without hiring
- Before
- Every new customer had to be checked: can they afford it, is their credit sound, are they who they say they are. As the company grew, that took nearly ten people.
- How we solved it
- We automated the whole check, from application to decision.
- The result
- Three people handled nine times as many applications, and it paid for itself in under three months.
In your business
Like handling nine times as many enquiries or orders without taking on more staff.
Under the bonnet: End-to-end credit decisioning, affordability and identity checks in an FCA-regulated lender.
Spotting risky customers months earlier
- Before
- Twice a year, analysts checked whether each tenant could still pay its rent. Between checks, a warning sign could sit unnoticed for months.
- How we solved it
- We built an AI that reads the news, company results and the client's own records, and flags any tenant whose risk changes.
- The result
- Every tenant watched all the time, problems caught early enough to act, and hundreds of hours of analysts' time back.
In your business
Anyone who owes you money, whether trade accounts, deposits or half-paid jobs. You'd hear they were in trouble before the invoice went overdue, not after.
Under the bonnet: Signals from earnings calls, news and internal data feeding a self-improving credit-rating model.
A website you change by asking
- Before
- A website is never finished, and every change used to mean a developer.
- How we solved it
- Both sites are built so I ask for a change in plain English, check it, and it's live the same day. Automatic checks stop anything wrong from going live.
- The result
- Changes go live in an afternoon, not weeks.
In your business
That's how I'd build yours.
Under the bonnet: Next.js with all the copy in one file. AI agents write, test and ship changes under my review, with checks that block an invented price or promise.
Where I've done it
2026 to now
Founder, TaxPost
A UK company that files small companies' accounts, run on the automations I build for clients.
2025 to 202618 months
Head of Solutions, then product manager, Fifth Dimension
An AI company for commercial property, backed by a $26m investment round. I led the team that went into clients' offices to make the AI work.
2022 to 20253 years
Senior product manager to Head of Product, Fidel API (now Astrada)
The data behind business payment cards and expense apps.
2020 to 20222 years
Product manager to senior product manager, Wayhome
Helping people buy a home without a mortgage.
2019 to 20201 year
Product manager, Sixley
A referral hiring platform used by FTSE 250 companies.
2016 to 2019
BSc Mathematics, University of Bristol
How I work
1
Sit with the work
I watch how the job really gets done, or how an enquiry really reaches you, because the real bottleneck is rarely the one you'd guess.
2
Pick what pays
A short list of what's worth fixing first, website or admin, with the time or money each one is worth.
3
Build it in
Set up properly for your whole team, with sensible rules on data, not one person's side project.
4
Make it stick
I train the people who'll use it and measure the difference.
Every one of these started with someone doing a job by hand that a machine could do better.
What's yours? One to three days of work, and a 30-minute call first. No pitch.
Book a call