Two client jobs and six worked examples, with the numbers.
I'd rather show you numbers than adjectives. Two of these are client jobs; six are composites built from work that keeps recurring. Names and details are changed throughout, and composite numbers are typical shapes, not claims.
Every order was being typed in twice.
A busy print shop was taking orders through its website, then having a member of staff re-key every one of them into the accounting system by hand. Roughly 18 minutes an order, all day, and the odd typo slipped through and caused a mismatch later.
I connected the store straight to the accounts, and each order now flows through on its own, correctly, the moment it's placed.
Re-keying dropped from ~18 minutes to under 3, and the transcription errors stopped.
Read the full case study, with the numbers → See the step-by-step fix →The same questions, answered over and over.
A small team was spending a chunk of every day answering the same handful of routine questions by hand, the kind of thing where the answer barely changes but someone still has to type it.
I built a simple assistant that handles the everyday queries and passes anything tricky straight to a person. A human stays in the loop for everything it touches.
Freed up hours every week for the work that actually needs a human.
See the step-by-step fix →Quotes written on the drive home.
Site visits ended with scribbled measurements and a quote written late that evening, or a few days later once it had already been forgotten twice.
Photos and a voice note now go to an AI step that drafts the quote in the firm's own format, priced from their rate card, ready to check and send.
Typical shape: quotes out same day instead of two evenings later, and first back wins work.
See the step-by-step fix →Every enquiry answered while it's still warm.
Tenant and landlord emails landed in one shared inbox and sat there. First replies took a day or two, and some never came at all.
AI now sorts each message, drafts the reply for a human to approve, and puts viewings straight into the diary.
Typical shape: first response in under twenty minutes, not a day, with a person approving every send.
See the step-by-step fix →Reminders that stopped the no-shows.
Bookings were confirmed on the phone and then forgotten by both sides, and an empty ramp is money you never get back.
Confirmations, day-before reminders and MOT-due nudges now run themselves from the booking diary, each one written to sound like the garage rather than a robot.
Typical shape: no-shows roughly halved, with MOT reminders bringing old customers back.
See the step-by-step fix →The month-end pack that builds itself.
Sales, stock and margin figures were copied from three systems into one spreadsheet every month: a full day of careful, boring work.
The systems now feed one small dashboard, and AI writes the month's summary in plain English: what moved, what stalled, what needs an eye.
Typical shape: month-end in about ten minutes, not a day, ready on the 1st.
See the step-by-step fix →Certificates and follow-ups on autopilot.
Every finished course meant certificates, feedback asks and renewal reminders assembled by hand, and renewals quietly slipped through the cracks.
Finishing a course now triggers the lot, with AI personalising each message and flagging who's due to rebook.
Typical shape: course admin in minutes, not an afternoon, and renewals stop leaking.
See the step-by-step fix →Tender drafts from their own past bids.
Tenders and PQQs meant a week of evenings rewriting the same answers from old documents scattered across the server.
Their past bids became a library an AI step draws on, producing a first draft in their own voice for the estimator to knock into shape.
Typical shape: a first draft in an afternoon, not a week of evenings, so you bid for more.
See the step-by-step fix →Jobs I see in almost every small business.
Beyond the stories above, these are worked examples: the fixes that turn up over and over in businesses under twenty people, with the typical shape of the numbers. Pick one, or let them rotate; if one sounds like your week, the free review will put a real price on it.
Late invoices, chased without you
Unpaid invoices get chased when somebody remembers, which is usually late and always awkward.
I'd connect your accounts package to a watcher that knows who's paid, has AI draft each polite chase in your tone, and sends it at the right moment. You'd get a short aged-debtors note every Monday.
Typical shape: a day of chasing each month → none
See the step-by-step fix →An inbox that drafts its own replies
Four people share info@ and nobody owns it, so first replies take days and things slip down the gap.
AI reads each enquiry as it lands, sorts it, and drafts a reply for a human to approve. Nothing goes out on its own; it just stops the queue building.
Typical shape: first reply in minutes, not days
See the step-by-step fix →Quotes that follow themselves up
A quote goes out, nobody nudges, and the job dies quietly after you'd already done the hard part.
Each quote gets a polite follow-up drafted from what you actually quoted, a few days later, then once more. You see everything still open at a glance.
Typical shape: every quote followed up, every time
See the step-by-step fix →Job write-ups that write themselves
An evening a week of typing up what you did on site, long after you did it.
Photos and a voice note from the job go in; AI drafts the report or the quote in your own template, ready for you to check and send.
Typical shape: an evening of write-ups → checked over a cuppa
See the step-by-step fix →The monthly numbers, assembled for you
Sales, bank and job figures copied into one spreadsheet by hand, every single month.
The systems feed one small dashboard instead, and AI writes a plain-English note of what changed and what needs an eye on it.
Typical shape: a Sunday with a spreadsheet → a two-minute read
See the step-by-step fix →Meetings that write their own actions
Decisions evaporate on the drive home, and nobody quite remembers who agreed to what.
Record the call, and AI turns it into a summary and a task list in whatever tool you already use, minutes after you hang up.
Typical shape: who said they'd do what, written down every time
See the step-by-step fix →The review ask that never gets forgotten
Happy customers would leave five stars if anyone asked at the right moment. Nobody asks.
When a job is marked done, a personal-sounding ask goes out with a direct link, and a QR card on the van or the counter does the same in person.
Typical shape: a review ask after every finished job
See the step-by-step fix →One customer, one record
The website, the diary and the accounts each hold their own copy of every customer, and the copies drift apart.
I'd connect them so a change in one place lands in all of them, with AI flagging duplicates and oddities as they appear.
Typical shape: three copies of every customer → one
See the step-by-step fix →Website orders straight into the books
Orders arrive online, then someone types each one into the accounts again by hand.
The store talks to the accounts directly: order in, invoice raised, nothing re-keyed. The oldest trick in my book, and still one of the best paybacks.
Typical shape: minutes of typing per order → none
See the step-by-step fix →Your version of this is where we'd start.
Most small businesses have at least one of these hiding in plain sight: a job done by hand that shouldn't be, two systems that don't talk, a question answered fifty times a week. The free review is where we find yours and put a number on fixing it.
See what it costs →