Hands-on AI for small businesses · UK

Put AI to work
in your
studio.

You know AI could help. You don't know where. I work that out with you, set up the right tools, and build the ones that don't exist. Costs shown up front. You own everything.

What I’d set up and build for a business like yours

Examples and case studies. Much of this is showing you tools you haven't tried, and what the ones you already pay for can do once AI is connected to them: meeting notes that file themselves, a CRM that fills itself in, invoices chased for you. Each one is scoped and priced on the call. The case studies are already built.

Case studyClaude · ChatGPT · your files

ChatGPT or Claude set up to do things your way

For
Anyone who has ChatGPT, Claude or Copilot and can't get it to behave the same way twice.
What it does
A folder of plain-English files it reads every time: how you work, each job as a recipe with its checks, what it must never do, what was decided. The quote reply comes out in your voice. The weekly report is the same shape every week, and checked before you see it.
Your part
You correct it when it's wrong. I show you where the correction goes, so it stays fixed.

Case study · How this consultancy and the services at a SaaS startup are run.

Claude · ChatGPT · your filesCase study
  1. readhow we work, before anything else
  2. runthe job, from its written recipe
  3. checkagainst its definition of done
  4. log"refunds counted as sales", as an issue
  5. fixthe job, plus a check. Right next week
You correct it when it's wrong. I show you where the correction goes, so it stays fixed.

Pick a card. 1 / 16

What we do

First we work out where AI helps. Then I set it up and build what’s missing.
Most of the value is in the first two steps.

01

Understand

A half-day workshop, or a walk through your business on a call: what AI can and can't do for you, the tools you haven't heard of, what the ones you already pay for can do once AI is connected, how to prompt them for your work, and where the hours go.

A shortlist, a cost per item, and what to do first.

02

Set up and integrate

Claude, ChatGPT, Copilot or Gemini, whichever you already pay for, set up properly and connected to Shopify, your CRM, Gmail, Slack, your accounts. Plus the tools you haven't met: Granola or Quill for meeting notes, Fastlane for short videos made, scheduled and posted. Training for the team, and your Playbook: the repeatable tasks written down once, so they run the same way every time.

You leave with your Playbook: how you work, the jobs written down, an always-and-never list. Every correction you make goes in, so it gets better instead of fizzling.

03

Build

When nothing off the shelf does it: a bespoke service (an agent) that runs the job, or a small internal tool built the same way, a dashboard, a stock view, a booking form, a website. Weeks, not months.

Working in about two weeks, on your real data. Nothing it drafts goes out until you approve it.

04

Run

Every build comes with its monthly running cost shown before we start, handover notes you can follow, and 30 days of support while it beds in.

You own the code. No platform fee, no lock-in.

Or tell me about your annoying manual process. If it repeats and has a clear input and output, it can probably run itself. Tell me about it

Tools, by the kind of business

Pick the one closest to yours. These are the tools I set up, connect and build on for a business like it. It isn’t the whole list. The ones that would make the biggest difference to you are the ones we talk about on the call.

Pick a kind of business

The tools nearly every small business already has, and the handful worth adding.

Everyday AI

  • ChatGPT
  • Claude
  • Microsoft Copilot

Enquiries and bookings

  • Calendly
  • Tally

Documents and admin

  • Gmail
  • Outlook
  • Google Drive
  • Microsoft 365
  • Google Sheets
  • Notion

Money

  • Xero
  • QuickBooks
  • FreeAgent
  • Stripe
  • GoCardless

Meetings and notes

  • Slack
  • WhatsApp
  • Zoom
  • Granola
  • Quill

Running the business

  • Zapier
  • You probably already pay for it. I connect it.
  • Worth trying. Most people haven’t.

Getting the most out of ChatGPT, Claude or Copilot

The difference isn't the tool.
It's the structure and rules around it.

Most people open ChatGPT or Claude, type what they want, and get something nearly right. Then they type it all again tomorrow. Everything it knows is what is in front of it, so the fix is structure: a folder of plain-English files it reads every time, with the checks it runs before you see the result. Set up properly, it does the job your way, and every correction stays made.

  1. 01

    How we work

    One short page on your business: what you sell, who to, how you sound, what matters, and the words you use for things. Read first, every time.

  2. 02

    The jobs, written down

    Each repeatable task as a recipe with a definition of done: what goes in, what comes out, and the checks it runs before you see the result.

  3. 03

    Always and never

    The rules. What needs your approval before it goes out, what it never says, what it never touches, and what it must ask about first.

  4. 04

    Issues

    When it gets something wrong it is logged like a fault report: what was asked, what came out, what was expected. The fix goes back into the job.

  5. 05

    The decision log

    What you tried, what you decided, why. So next month, or the next person, doesn't start from zero.

Your Playbook, as a folder

  • your-playbook/
  • how-we-work
  • jobs/
  • weekly-report
  • quote-reply
  • newsletter
  • always-and-never
  • issues
  • decisions

The bit nobody tells you

When it gets something wrong, the correction goes in the file, not in your head. Logged once, fixed in the job, checked from then on. That is the difference between progress and the same frustration every morning.

  1. you"The weekly report counted refunds as sales."
  2. loggedissues: what was asked, what came out, what was expected
  3. fixed injobs/weekly-report, plus a check: sales never include refunds
  4. next weekright, and checked before you see it

This is the Playbook. It is what you leave the workshop with, set up in whichever tool you already use.

How to work with it, day to day

The files are half of it. The other half is method: how good software teams work with AI, translated for a business that has never written a line of code.

01

Brief before build

A one-page brief you approve before any work starts: what, for whom, and what done looks like. Bigger jobs are broken into small tickets, each signed off on its own.

02

Have it interview you first

Before the brief, it asks the awkward questions: what happens when the data is missing, who signs off, what must never happen. Five minutes of questions saves an hour of fixing.

03

Define done, then check it

Every job carries a short list of checks it runs before you see the result: the figures add up, every row is filled, nothing goes to the wrong person. A job with no checks isn't finished.

04

Keep a known-good example

One past run with the answer you were happy with. When the instructions change, run it again. If it still matches, nothing broke.

05

Log problems like faults

What you asked, what came out, what you expected, in a list it can read. Fixes go into the job, and the list shows you what keeps going wrong.

06

Small steps, tight loops

The faster you can check, the faster it improves. One step, looked at, then the next. Big asks get big mistakes.

07

Give it what it needs, not everything

Everything it knows is what is in front of it. A short front page it reads every time, and the detail in each job's own file, read only when that job runs. More isn't better.

08

Examples, not adjectives

Three real emails you were happy with beat 'friendly but professional'. It copies what it can see.

09

A note at the end

What changed, what was decided, what to do next time, written where it will read it tomorrow. That is how it gets better week on week.

Habits I run every day and coach my clients on.

Who this is for

For people who run a small business
and still do the slow, manual work themselves.

A good fit if

  • You run a small business, agency or practice, and you're still the one doing the long, manual jobs: the emails, the chasing, the reports nobody has time to pull together. You'd like AI to take some of it, and you'd like to be the one who knows how.
  • Something in the business repeats every week, and you'd rather it ran itself.
  • There's a report or a piece of analysis you know you should be doing, sales by month, what the reviews are saying, whether your shop turns up when people ask ChatGPT, and it never gets done because it takes a day.
  • Every call ends with notes nobody files, and every week ends with a catch-up nobody writes.
  • You know AI and better tools could help, but nobody has shown you where, and there's no time to work out which ones actually matter for your business.
  • You've used ChatGPT and got something out of it, but it's still a tab you open, not something the business runs on. You want it built in, and you want to get properly good at it yourself.
  • Something technical goes wrong now and then, an order that fails, an alert nobody understands, and you'd like a first diagnosis waiting for you rather than a morning lost to it.

Probably not if

  • You want a chatbot on your website. That's a different product and there are cheaper ways to get one.
  • You want a strategy document. I build things, then show you what they cost.
  • You need enterprise procurement, SLAs and a team of five.

How it works

A free call, a one-page plan,
then it’s built and handed over.

  1. I

    A 30-minute call

    You describe the business and the bits that annoy you. I tell you straight what AI will and won't do for you, roughly what it costs, and whether I'm the right person.

    Free. If it's not a fit, I'll say so.

  2. II

    A plan you can read

    A one-page shortlist: the tools, what each costs a month, how much work to connect it, and what to do first. For a team, we do this live in the workshop.

    Most people find two things on it they can do this week.

  3. III

    Set up, built, handed over

    I connect the tools, train you, and build anything bespoke. You see it working on your real data within days, not at the end. Then handover notes, a handover call, and 30 days of support.

    It asks before it acts: it drafts, you approve.

An example, in plain words

Every Monday morning it checks what's running low across the store, works out what to reorder from the last two months of sales, and drafts an email to each supplier. I read them with a coffee and press send. That's the whole job.
Rory, on a restocking service. An example build, scoped and priced on the call.

Bespoke services & agents

Built for real businesses.
Running today.

Every service here runs on a schedule, on a known monthly cost, and asks a person before it acts. The status is real: if one is paused or not built yet, it says so.

1 / 13

  • Live

    Every job, every day

    The Playbook: the files every service here runs from

    A folder of plain-English files read before every job: how we work, each job as a recipe, what it must never do, what was decided. Every service below runs from it.

    Reads
    How we work, the job's recipe, the always-and-never list.
    Keeps
    A log of what was decided and why, so nothing is re-argued.
    Guardrail
    Corrections go in the file, not in someone's head. Said once, fixed for good.
    For
    This consultancy and the services at a SaaS startup.

    Claude · Claude Code

  • Live

    Four sweeps a day

    Alert triage

    Groups the error alerts, investigates each group against the logs and the code, and posts a diagnosis. Nothing is filed until someone approves it.

    Runs
    Four sweeps a day, on a schedule.
    Reads
    The alerts channel, then the logs and the code for each cluster.
    Guardrail
    Proposed tickets wait for approval. If nobody approves one, it is dropped.
    For
    A SaaS startup, live production alerts.

    Discord · GitHub · Claude

  • Live

    07:00 daily

    Social post drafter and scheduler

    The day's posts and replies drafted and queued for the company account. A separate, approved step is the only thing that ever posts, each at its time.

    Runs
    07:00 daily for the posts, always-on for replies.
    Writes
    The day's posts and replies, in the company voice.
    Guardrail
    A separate poster is the only process allowed to write to X, and every post carries an explicit approval.
    For
    A SaaS startup's company account.

    X · Claude

  • Live

    Weekly

    Release notes

    Reads what shipped this week, drafts the what's-new post, and publishes at the weekend once someone has approved it.

    Runs
    Weekly: drafted on Friday, published at the weekend.
    Reads
    The week's code changes.
    Guardrail
    Nothing publishes until someone approves it. An unapproved draft is dropped.
    For
    A SaaS startup's customers and team.

    GitHub · Claude

  • Live

    Always on

    Ticket intake

    Watches the support channel, turns each real problem into a ticket with the details filled in, and skips the chatter.

    Runs
    Always on.
    Watches
    The support channel, one thread per report.
    Guardrail
    Only an approved list of actions. A reaction shows the status of each.
    For
    A SaaS startup, support channel.

    Discord · GitHub · Claude

  • Live

    07:30 daily

    One next action a day

    One nudge per founder each morning, with a memory of everything learned along the way. It never touches the platform itself.

    Runs
    07:30 daily, plus always-on.
    Keeps
    A memory of everything learned along the way.
    Guardrail
    It never posts anywhere itself. A person does every action.
    For
    A SaaS startup's founders.

    Discord · Claude

  • Live

    Weekly

    Meeting notes and a weekly team update

    Quill records and writes up every call. Once a week a small service reads the notes, pulls out the decisions and action points, and sends the team one update.

    Runs
    Once a week, after the week's calls.
    Reads
    The notes Quill wrote, one per call.
    Guardrail
    Read-only. It writes one update to the team and nothing else.
    For
    A SaaS startup.

    Quill · Claude

  • Live

    Mondays at 09:00

    Business miner

    Pulls founder interviews from YouTube, writes a note per episode on what worked and the tools used, and builds a weekly digest.

    Runs
    Mondays at 09:00, unattended, on a schedule.
    So far
    26 founder interviews read and written up, one note each.
    Guardrail
    Read-only. It only writes notes, a tools list and a digest.
    For
    One Less Hat, and the free report to come.

    YouTube · Claude Code

  • Live

    Relaunched April 2026

    Prints & ceramics store

    Shopify theme rebuilt and relaunched, back-in-stock and abandoned-cart emails added, and a full sales report on what changed and what didn't.

    Live since
    The April 2026 relaunch.
    Result
    +28% average order value and +13% revenue, same window year on year.
    Found
    £3.2k of abandoned checkouts with nothing recovered, which became the next build.
    Honest bit
    Conversion dipped, from 1.3% to 1.1%, and the report said so.

    Shopify · Email flows

  • Live

    Set up, in use

    Short videos made, scheduled and posted

    Granola and Fastlane set up for an online store: meeting notes taken for them, and short videos made from products and trending clips, scheduled and posted.

    Status
    Set up and in use at an online store.
    Does
    Short videos made, scheduled and posted to TikTok, Reels and Shorts.
    Guardrail
    The owner picks which videos go out. Nothing posts without a yes.
    Saves
    The hours that went on making, scheduling and posting content.

    Fastlane · Granola

  • In build

    September 2026

    Corporate finance boutique website

    A new site for a London corporate finance adviser: three styled concepts to choose between, built with AI in the loop in days.

    Status
    In build, September 2026.
    Scope
    A new site in three styled concepts, for the partners to choose between.
    Built with
    AI in the loop, in days rather than weeks.
    Next
    The partners pick one, then it goes live.

    Claude Code · GitHub

  • In build

    Weekly, in build

    Competitor watch

    Reads the competitors' sites, changelogs and pricing every week, reports what shipped and what it means, and raises a ticket for anything worth building. Tickets wait for approval.

    Status
    In build for a SaaS startup, September 2026.
    Reads
    Competitors' sites, changelogs and pricing pages, once a week.
    Writes
    A weekly report on what shipped and what it means, and a ticket per thing worth building.
    Guardrail
    Read-only on the web. Every ticket waits for approval, and unapproved ones are dropped.

    Competitor sites · GitHub · Claude

  • Research

    Next up

    Free AI opportunities report

    Describe your business and get three specific automations with real costs, grounded in what similar businesses actually run. Not built yet.

    Status
    Research. Not built yet.
    Shape
    Three specific automations for your workflow, each with a tool, a monthly cost and a real example.
    Fed by
    What the miner learns each Monday.
    Next
    Built once the notes cover enough kinds of business.

    The miner's notes · Claude

The numbers so far

Real numbers,
from real reports.

+28%

average order value

Shopify relaunch

Rory Collins, smiling, in glasses and a blue shirt

Who's building it

I'm Rory.

Full-stack software engineer, now AI engineer · UK

I'm a full-stack software engineer who moved into AI. I've built and run live products end to end: the screens people use every day, the systems and data pipelines behind them, and the unglamorous parts that keep them working, from testing and security to hosting and costs.

Then I built a SaaS startup, a trading journal, end to end. It's also where the AI side took over: rather than add people, the routine work went to services I built, which watch the alerts, draft the posts, write the release notes and file the tickets, each one asking a human before it acts.

AI is moving fast. New tools come out every week, the good ones genuinely save hours, win sales and help a small team do more, and keeping up with it all is a job in itself. That's the hat I wear so you don't have to. I follow what's new, try it on real work, and know what these models can do this month and what to keep them away from.

That's the useful bit for you. I've shipped software that businesses run on, and I run this consultancy on tools I built myself. So when I look at your business I'm not selling AI. I'm looking for the hours you'd get back.

  • Built a SaaS startup
  • Full-stack → AI engineer
  • UK · remote

Pricing

Three ways to work with me,
each with the cost up front.

Most people start with the workshop. Setting up and integrating tools is scoped on the call; most of it is a day or two. Every price comes with the running cost, shown before we start.

01

Start here

Workshop

Where most people start

£750–1,500

half a day or a full day, in person or remote

What AI can and can't do for your business, hands-on, and you leave with your Playbook set up in ChatGPT, Claude or Copilot. Half a day finds the wins and writes the first pages. A full day also builds your first bespoke service (agent) with you in the room.

  • Built around your own tasks, in your own accounts
  • Your Playbook: how you work, your first three jobs written down, and the always-and-never list
  • Full day: a first bespoke service, built live
  • Credited against a bigger build if you go ahead

02

Bespoke build

Fixed price

£3,000–5,000

one job, one price

Something built for the one job nothing off the shelf covers: a service that runs it, or a small internal tool. Working in about two weeks.

  • Scoping call and a written plan
  • Code you own, handover notes and a walkthrough
  • 30 days of support
  • Run cost shown before we start

03

Partnership

Ongoing, month by month

£1,500–3,000

a month, depending on what you need. Cancel any time

For after a workshop or a build, when you want it to stick and keep growing. I stay on, keep what we built running, and keep finding the next hour to give back. Where in the range depends on how much of my time you need each month: how many sessions we do, and how many services I keep running for you.

  • Two working sessions a month, on your real work
  • Your services kept running: monitoring, fixes, small changes
  • New tasks added to your Playbook as they come up
  • Priority line between sessions, and new starters trained

Prices exclude VAT. Running costs are the real monthly bills for the AI tools and hosting, shown up front.

Questions

The usual ones,
answered straight.

I don't know what I need. Is that OK?+

That's most people, and that's what the call and the workshop are for. Describe the business and the annoying bits; I'll tell you what's worth doing, what it costs, and what to leave alone.

Do I need to be technical?+

No. I do the connecting and the setting up. You learn where the buttons are, what to glance at each week, and how to change the instructions when the business changes.

I use ChatGPT or Claude but can't get it to do what I want. Can you help?+

Yes, and it's the most common thing I fix. It's almost never the model. It's that nothing is written down, so every conversation starts from zero and every correction is lost by tomorrow. The workshop sets up your Playbook: who you are, how each job is done, what it must never do. After that, it reads it every time, and when you correct it, the fix stays.

We already use ChatGPT. What would be different?+

Using ChatGPT for the odd email is a good start, and most businesses stop there. The difference is making it part of how the business runs: connected to your inbox, your calendar and the tools you already pay for, with clear instructions for your jobs, and a named person on your side who owns it. That is what usually goes missing when someone sets up AI and it quietly stops getting used after a fortnight. So every build here comes with a plain-English playbook, an owner, and sessions until it's routine.

What does one of these actually cost to run?+

Typically £10–30 a month for the AI tools and hosting behind it, and sometimes nothing: a lot of the best tools are free and open source, and where a free one does the job, that's the one I'll use. I'll give you the number before we start, and the handover notes show you where to check it.

Do I need a developer to keep any bespoke-built agents running?+

No. The instructions are plain text files you can edit yourself, and the handover notes cover the routine things. You get 30 days of support after handover, and I am a message away after that.

What if it makes a mistake?+

Everything I build asks before it acts. It drafts, you approve. Nothing gets sent, posted or filed without a person saying yes, and if nobody says yes, it's dropped. Silence means no.

Do you do Shopify?+

Yes. I've rebuilt several stores and built the reporting behind them, so store digests, listing writers and customer email triage are the builds I know best.

What is the Playbook?+

Your repeatable tasks, written down once in plain English so the tools do them the same way every time: the quote email, the product description, the weekly report, the supplier chase. It lives in your own files, you can edit it, and the bespoke services work from the same Playbook.

Do you build websites and apps too?+

Yes, when they're the right answer. With AI in the loop, a dashboard, an internal app or a website rebuild takes days rather than months, so it's often the quickest way to give a business the tool it's missing. I don't take on big software projects, and I'll say so on the call if that's what you need.

What do you mean by tools I haven't tried?+

Two kinds. Tools you may not have met: Granola, or Quill which is free, to take your meeting notes for you; Fastlane, which makes short videos for a store and posts them; Claude or ChatGPT set up for your business with your Playbook, rather than a blank tab. And the tools you already pay for, connected to AI so they do more: your CRM kept up to date from your calls and emails, your meeting notes filed where the team can find them, your accounts package drafting the invoice chasers. Most businesses I meet are using a fraction of what their tools can already do.

Can it work with my CRM?+

Yes. HubSpot, Pipedrive, Zoho, Salesforce and most of the others let tools like this read and write to them, so notes, deal stages and follow-ups can be kept up to date without anyone typing them in. If your CRM is a spreadsheet, that works too.

Not sure where to start?
That's what the call is for.

Book a 30-minute call. I'll tell you straight what's worth doing, what it costs to run, and whether I'm the right person to do it.

Thirty minutes, free, nothing to prepare.