AI Transformation Roadmap

Bramforth Interiors

Prepared for: Gary Whitlock
Business: Bramforth Interiors
Sprint conversation: 12 August 2026
Sent: 19 August 2026
Prepared by: Alistair Anthony, Dux
Note: Illustrative. Not client data.

A demo asset. Bramforth Interiors is a fictional business, and everything in this document, prices included, is invented for illustration. The prices are indicative only. Nothing in it is an offer or advice, and none of it binds Dux or any reader.

What this document is

Bramforth Interiors is a commercial interior fit-out contractor working across the Thames Valley. When we spoke you said the thing you actually want is pricing everything you are sent without another estimator on the payroll. That is the test everything below has to pass.

The findings below draw on the systems (Microsoft 365, Xero, Dropbox, Bluebeam Revu and the shared Excel programme sheets), the process walkthrough and the questionnaire answers. Each mark has its reason set out beside it, so you can argue with the ranking where you know better.

This document ranks the options; it does not design the builds. Each entry sets out what the work involves and what has to be true before it starts. The detail can wait. The decision now is which of these is worth doing at all.

This column is the judgement. The wider column is your business, in your own figures.

How priorities are marked

Two things decide where an item sits. The first is what it would change here if it worked. The second is what it would take to get there, including how much has to be true before it can start.

[First] means the change is worth having and the route to it is short, or that other items depend on it being done first. [Soon] means worth doing once what it leans on exists. [Later] is a statement about order, not worth: some of the best ideas below are better done once something else is in place.

[First] does not mean urgent, and it does not mean do all of them.

Gold marks the [First] items.

Room on the page follows the ranking. The items worth moving on first get the most of it.

The map

The whole judgement on one page: 6 items ranked, 3 refused.

The gold marks sit in different parts of the business. Unrelated work, so both start at once.

The right half is not worse work. It depends on habits that are not in place yet.

What it returns

700%

year-1 return on the 2 builds, on gross margin

£84,000

gross profit added a year, at today's win rate

33 days

of the office manager's year returned

Price every enquiry at today's win rate and the wins double.

Sources: Xero and the Sprint conversation (notes 1 and 2).

Your last 12 months in Xero show 21 completed contracts at an average value of £148,000, on a blended gross margin of about 9.5%.¹ Price all 60 enquiries and today's 1 in 5 adds about 6 wins a year on top of the book you win now: roughly £890,000 of new work, carrying about £84,000 of gross profit.² The pilot that starts it costs £4,500.

Gold here is what the [First] build buys: the 6 wins a year lost because half the book never gets priced.

Nothing here assumes a better win rate. It assumes today's rate on twice the book.

The only new figure is the contract value, and it is yours, from Xero.

One tender: 5.5 hours to 1.5. One handover: 2 days to about half.

Sources: walkthrough notes (notes 3 and 4).

Checking a schedule is about 1.5 hours where building one was 5.5,³ so the same evenings price twice the book. A handover drops from 2 days of assembly⁴ to about half a day of chasing gaps; across the year that is about 33 of the 44 days the manuals cost. Invoice matching, when its turn comes, starts from a full day a week of the bookkeeper's time.¹³ On the 2-in-5 exception rate the walkthrough found,¹¹ about half of that day comes back.

Both [First] builds, £10,500 together, are covered by the margin on 1 extra contract. Over the year that is a return of about 700% on gross margin: £84,000 back on £10,500 spent.

Checking replaces building. The person still makes the judgement.

Opportunities

Where one item depends on another, that dependency is written beside it.

Estimating and tendering

The scope schedule pulled out of the drawing pack

Every tender starts the same way. Someone reads the drawing pack and the specification and turns them into a room by room schedule of what has to be built, and only then does anyone put rates against it. The reading and the listing are the slow part, and that is the part a tool can now do well enough to check rather than trust blind. A tool can work the drawing set into your estimating template with the source page noted against every line, so an estimator checks a schedule rather than builds one.

Why it matters here. You price most tenders yourself and it costs you 2 evenings a week.⁵ In the walkthrough, one live tender took 5.5 hours to price, and 4 of those hours went on reading and listing before a rate went in.³ You said that with 2 estimators you should be checking numbers rather than producing them. This is the part that produces them, and it does not need you.

What it would take. 2 to 3 weeks to map your template and tune the extraction, then a spell running it beside the manual take-off until the estimators trust it. Scoped as a fixed piece of work: a pilot on 10 past tenders, about £4,500. Packs where you sign an NDA on the drawings need a tool that agrees not to learn from your drawings, which rules out the free tiers.

What it would free up. Enough capacity to price more of the 60 tenders a year you are sent instead of about half of them. At 1 in 5, the 30 that go unpriced are 6 jobs a year never bid for.² That is the shortest route to money on this list, and it does not need a hire.

[First].

The drawings already sit in Dropbox and Bluebeam Revu, so the route is short, and every other estimating item gets easier once this exists.

Which enquiries get priced at all

A triage step reads each enquiry as it lands and writes one line into a list: client, size, deadline, type of work, and whether it looks like the work you usually win. Nothing is answered or declined automatically.

Why it matters here. You price about half of the 60 tenders a year you receive, and which half is decided by how your week is going rather than by which ones you would win. 3 of the 10 most recent enquiries in the inbox had passed their deadline unopened.⁶

What it would take. About 1 week. It needs the scope schedule in place first: a list of what you are turning down only helps once you have the capacity to price more of it.

[Soon].

Gated on the scope schedule. Triage sorts the queue; the schedule is what shrinks it.

Running the jobs

Site visit notes turned into the actions they carry

Your project managers walk sites and the record of it lives in a notebook or in the WhatsApp groups for each site. A voice note on the drive back can be turned into a short written record with the actions separated out: who, what, and by when. The manager corrects it and sends it.

Why it matters here. You said nothing from a site meeting gets written up properly unless something has already gone wrong, and that how a job is run lives in the project managers' heads. One site's WhatsApp group took 41 messages on a single Friday; none of them reached a job file.⁷ That is knowledge the business cannot spread or cover when someone is away.

What it would take. 2 weeks, most of it spent agreeing with your 3 project managers what a good record looks like. Start on one site.

[Soon].

Short route, but the two [First] items come first, and the board below has nothing to show until this exists.

One board that shows where every job actually is

The programme for every live job is a shared Excel sheet, kept up to date by whoever last remembered. A board would hold one view of where each job has got to, built from the site records.

Why it matters here. With 22 projects a year overlapping,⁴ the question you ask most often is where something has got to, and answering it costs a phone call and an afternoon. The shared programme sheet had last been saved 9 days before the walkthrough.⁸

What it would take. Longer than it looks, because a board is only as honest as what feeds it. It needs the site records above to be a habit first.

[Later].

Waits on the site records habit. Built now it would be a tidier version of the shared Excel sheet with the same holes in it.

Handover and the back office

The handover manual assembled instead of rebuilt

Every handover produces the same document, and most of what goes in it exists by the time the job finishes. A tool can pull the warranties, data sheets, as-built drawings and commissioning certificates out of the job folder, lay them out to a standard contents page, and flag the gaps while there is still time to chase them. That leaves the office manager chasing gaps instead of laying out pages.

Why it matters here. The O&M manual takes the best part of 2 days per project, and at 22 projects a year that is the largest block of time the office loses to rebuilding a document that already exists.⁴ The walkthrough opened the folders for 4 live jobs and found 4 different structures, with the same warranty filed 3 ways.⁹ The missing-items flag is worth as much as the assembly, because chasing a subcontractor for a certificate is easy in week 10 and hard in week 15.

What it would take. 3 to 4 weeks, most of it on the folder structure, which has to be consistent before anything can read it. Scoped as a fixed piece of work: about £6,000. Some of this sits inside what you own already: you pay for Microsoft 365 Copilot on 3 seats and barely touch it.¹⁰

What it would free up. About 33 of the 44 days a year the manuals cost the office manager, on your own figures, and handovers that arrive on time.

[First].

The item that fits the only capacity you named: the office manager's half day a week.

Subcontractor invoices matched against the purchase orders

Your bookkeeper spends a day a week matching subcontractor invoices to purchase orders.¹³ That work is mechanical: match each invoice to its order, pass the clean ones through, and put the rest in front of a person with the difference already worked out. Nothing is approved or paid without the bookkeeper.

Why it matters here. It is a full day of a part-time role, and the errors it catches surface late, usually once the final account is being argued about. 2 of the 5 invoices checked in the walkthrough had no purchase order to match.¹¹

What it would take. Purchase orders would have to be raised consistently, which you said they are not. That is a change in habit before it is a change in software. Once the habit holds, the build is 2 to 3 weeks.

[Later].

A real saving, gated on the purchase-order habit. The day those orders become consistent, this moves up.

Where AI stays out

These are the parts where AI would waste your money.

The rates and the final tender number.

You already know why. You asked for pricing help, the numbers came back looking right and were not, and you stopped. That instinct is correct. The scope schedule above stops deliberately where the rates begin. It lists what has to be built; pricing it stays with your estimators.

Risk assessments and method statements.

Your insurer expects these signed off by a named person and main contractors audit it, so the signature has to mean somebody read it. You also said it invented a British Standard reference in a method statement, and that you caught it only because you knew the number. A method statement has to be right about the standard it cites. A tool that can sound certain and still be wrong is the wrong tool for that job.

The conversation when a job has slipped.

Nothing here should draft the message you send when the programme has gone. That conversation is the relationship.

The missing marks are deliberate. These 3 are out permanently, not parked for a later phase.

The order to take these in

The scope schedule comes first. Until it is settled, sorting your enquiries better would only tell you in more detail what you are turning down. The handover manual is the natural second move. It barely depends on the first, so the two can run side by side. It is also the one most likely to bring the rest of the team round, because the time it frees up belongs to someone other than you.

Enquiry triage follows, gated on the scope schedule. Site notes follow once the two [First] items have bedded in.

The board waits on the site records becoming a habit. Invoice matching waits on purchase orders being raised consistently, and that fix is yours to make.

2 of these done properly beats all 6 started.

The first 90 days

Dates assume a September start and the two [First] builds running side by side.

Weeks 1 to 3: The scope schedule pilot: 10 past tenders through the extraction, checked against the schedules your estimators built by hand. Your time is checking output, about 2 hours a week. Decision at week 3: either the pilot prices a live tender to your standard, or it stops.
Weeks 2 to 6: One folder structure agreed for every live job, then the handover assembler runs on the next project to finish. The office manager's half day a week is the only capacity it needs.
Weeks 4 to 8: The schedule runs beside the manual take-off on live tenders. This phase ends when the estimators sign off, not on a fixed date.
Week 8: Enquiry triage switches on behind the scope schedule. From then on every enquiry gets a line in the list, priced or not.
Week 12: Review against the test this document opened with: tenders priced per month, against the 2 or 3 a month you price today.

The £15,000 you named was against the first build.¹² Taken as the ceiling for both [First] builds together, they land under it at £10,500. Neither commits you to anything further down this list.

If the dates slip, the order holds.

Site notes start around month 4. The board and the invoice matching start once their habits are in place.

If you want help afterwards

Nothing here needs Dux to act on. If you already work with someone who could build it, this document should be enough for them to start from.

If you would rather not do it alone, there are two ways Dux works after a Sprint. Implementation is scoped and priced separately, against whichever item on this list you pick. AI Concierge is ongoing access as questions come up, for a run of decisions rather than a single build.

Either way it starts as a conversation. If the scope schedule is where you want to begin, reply to the email this arrived with and name a morning. The pilot comes back scoped and priced in writing the same week.

Sources

Every figure in this document is one of three things: yours, an artefact opened during the Sprint, or arithmetic on the two. The notes below state which.

Xero, invoiced totals for the 12 months to July 2026. 21 completed contracts, average value £148,000, blended gross margin about 9.5%. Invoicing trails site completion, hence 21 against the 22 you finish. The 21 are priced tenders plus repeat and negotiated work; none came from the unpriced half.
Sprint conversation, 12 August 2026. "If a week came back I would put it into pricing more of the sixty tenders a year we are sent. We only price about half of them." And later: "We win about one in five of what we price, so pricing more is the whole thing."
Walkthrough notes, 12 August 2026. One live tender timed from first read to submitted price: 5.5 hours, of which 4 on reading the pack and listing what it contained.
Sprint conversation, 12 August 2026. "The O&M manual takes the office manager the best part of two days per project, and we finish about 22 projects a year."
Sprint conversation, 12 August 2026. "I price most of the tenders myself and it takes two evenings a week."
The enquiries inbox, opened during the walkthrough. 3 of the 10 most recent enquiries were past their stated deadline, unopened.
Site WhatsApp group, one live job. 41 messages logged on Friday 7 August 2026; none carried to the job file.
Shared programme sheet, file history. Last saved 3 August 2026, 9 days before the walkthrough.
Dropbox, job folders. The 4 live projects opened side by side: 4 different folder structures, the same warranty filed 3 ways.
Sprint conversation, 12 August 2026. "We pay for Microsoft 365 Copilot on three seats and barely touch it."
Purchase ledger, sampled during the walkthrough. 5 subcontractor invoices drawn; 2 had no purchase order to match.
Sprint conversation, 12 August 2026. "The office manager could give half a day a week. Nobody else has anything spare. If something worth building comes out of this, I could go to about £15,000 for the first one."
Sprint conversation, 12 August 2026. "The bookkeeper spends a day a week matching subcontractor invoices to purchase orders, and we tend to find the mistakes when we are already arguing about the final account."

Quotes are verbatim. Nothing is paraphrased.

Illustrative. Not client data.