AI Opportunity Map

Ashcombe runs on five office flows: answering new work requests, quoting from the survey, booking and updating customers, keying supplier documents, and your Monday round-up. On a recorded 90-minute call you and your office manager took us through each one: the tools each flow passes through, roughly how many hours a week your people spend on it, and what goes wrong often enough to be worth mentioning. This report prices what you described.

The pattern repeats in all five. The work itself is fine. It is the handovers that lose things. Nobody owns the reply, so about one new request in four waits past the same day. Every quote is rebuilt from supplier PDFs and an old job that looked similar. Bookings are confirmed by nobody, so four times a week a technician arrives at an empty house. Supplier documents are retyped by hand into the accounting system. And the numbers that would show all of this get assembled on Monday morning, from four systems, by you.

Two of the six need no build at all: an off-the-shelf tool already does the job, and one of those two you pay for today with the feature switched off. The other four are rules and drafts inside systems you already own, not a rebuild. Three areas we recommend leaving to people are set out in section 7, with the reason for each.

One recorded 90-minute call with you and your office manager on 26 August 2026. You described each process as it actually runs, named the tools it passes through, estimated the hours your people spend on it, and said plainly where the current setup frustrates you. Eleven office, dispatch and finance processes were covered; the work in the vans was out of scope. Nothing here comes from watching the systems ourselves, which is why the volumes are your figures, and why the four worth checking are named below.

The funnel, once. About 1,200 new work requests reach you a year, 23 a week, by phone, web form and email. Plan-service bookings and callbacks are counted separately. Those 1,200 become about 530 quotes and about 265 won jobs. Add the 1,450 maintenance-plan visits and the repair work and the year comes to about 3,200 jobs, counting a two-day install as one, and about €2.9M invoiced. Every number below sits on that funnel.

How the numbers work. Time leaks are hours multiplied by a blended loaded cost of €40 across office, dispatch and surveyor time - our standard assumption for a trade office, not your payroll data. Most of those hours belong to your surveyors and to you (findings 3 and 6), priced at the same blended rate, which understates them; the rest is office time on supplier documents (finding 5). Findings 1, 2 and 4 carry no hours at all, because nobody put a duration on that work on the call: we price only what they leak, and an empty hours cell in the map means exactly that. We count 52 weeks because the office and the emergency line run year-round, and recoverable values assume only 70% of the affected work actually moves. Money leaks - jobs lost to the slow reply, wasted trips, missed credits, lapsed plans and unchased invoices - are sized separately and conservatively per finding, in contribution rather than revenue, each with its formula written out under "How we sized it", and never derived from the hours, so nothing is counted twice. We count only 50% of a money leak as recoverable. Each finding carries a confidence label: high where the count already sits in a system and can be pulled, medium where the volume is your estimate, low where no baseline exists yet.

The four numbers worth checking. Four figures here are your estimates from a single call, and each one can be settled from a system you already have: count last week's requests that got no same-day reply and how many of those never booked at all, from the phone log and the inbox; count the wasted trips in the dispatch log; list the credit notes and duplicate payments on last quarter's supplier statements; pull last year's lapsed maintenance plans from the CRM. The €800 contribution per won job is our assumption across installs and repairs, and the last twenty invoices will confirm or correct it. Each money figure is built from the formula printed under "How we sized it" on the finding itself, so a corrected count moves that one line and nothing else.

What this report deliberately leaves out. Our price for building any of it. The audit's job is to find the leaks, price them and name the fix, and all six findings carry the fix, the tool and the effort class. What a build costs depends on which findings you pick and who does the work, so a fixed quote comes per opportunity, once you have chosen.

One compliance note. None of the recommended workflows touch hiring, worker evaluation or eligibility decisions, no after-hours AI voice agent is recommended on this map, and the only AI-generated content is quote drafts a person reviews and sends. That keeps you outside the EU AI Act's transparency-labelling duties as of September 2026. One finding names an optional AI layer that would change the answer: a thread that talks to your customers has to say it is AI, which is part of why it is flagged as something to test later rather than recommended here.

Most steps are fine. The leaks sit at the handovers: the call nobody picks up, the quote that takes three evenings, the booking nobody confirms, the invoice nobody chases, and the Monday that starts with a spreadsheet.

Also seen, not ranked. Six smaller leaks came up on the call and did not make the top six: your inbox, about an hour a day; social posts written between jobs when someone remembers; outreach to property managers done by hand, twice a year; a CRM trial paid for and never finished; an automation between the web form and the spreadsheet that stopped working in March and was never noticed; and new-hire paperwork rebuilt from scratch for every apprentice. None of them is small enough to ignore forever and none is big enough to come before the five above. They wait for the second map.

Four moves you can make before any workflow is built, worth about five hours a week between them. None of these came up on the call: they are what we would start with in a business set up like yours. Two are also groundwork for the fixes above, the enquiry response and the quotes built by hand.

Not everything on the pain list should be automated. Three areas came up where the answer is a person, and the reason is different in each case. What they share is the shape of the answer, not the reason for it: in all three, automation still carries the paperwork around the decision. It just never makes the decision.

The order follows one rule: fix the constraint first. The phone is the front door the whole business queues behind - and the first three steps are rules and settings in systems you already have, so the one foundation build at the end is funded by measured results, not projections.

Two layers of assumption sit under every number. The four in the box below scale the whole report: change one and every figure moves with it, from the summary tiles to each finding. Underneath those sit the volumes and rates inside each money leak - how many requests a year, what a won job is worth, how often a document carries an error. Those are printed under "How we sized it" on the finding they belong to, and collected at the end of this section. Correct one of those and only that line moves. The printed copy uses our conservative defaults throughout.