Construction AI — Paths & Open Questions

A working note for us to align on before we commit. This is not a plan or a recommendation — it lays out the three paths I see, the concerns attached to each, and the questions we need to answer ourselves to get a clearer picture of risk vs. reward. Read it as a checklist of unknowns, not a pitch.

1. Where we are

• We have a working, deployed demo of an AI-first field-intelligence system for construction.
• We have warm access to one or two UK-based construction firms — already running real, sizeable operations.
• We have a multi-week window to decide how we go to market, not whether.

The decision in front of us is a business-model choice, and underneath it a risk/reward choice. The goal of this note is to make the unknowns explicit so we can size each path honestly.

2. The decision lens

For every path, I want us to be able to fill in four cells:

Right now most of these are blanks. The point is to systematically replace each ? with a number or a range, so we can compare risk-adjusted payouts — not just headline payouts.

A high reward with high risk isn’t automatically worse than a low reward with low risk, or vice versa. We need the actual numbers before we can judge. Sometimes a high-reward / high-risk option beats a low-reward / low-risk one; sometimes it doesn’t. We can’t tell yet.

3. The three paths

Path A — Consulting (in and out)

What it is: They hand us one or two concrete problems. We solve them — vibe-code / build fast — and move on. Possibly repeat with the next problem.

Why it’s attractive:

• Quick. Bounded scope.
• Low ownership, low long-term commitment.
• Likely the lowest-hanging fruit — fastest path to cash.

What concerns me:

• For them to keep seeing value, we’d have to keep solving new problems. Unclear how sustainable that is.
• No recurring revenue, no compounding asset. Every new client is a fresh sell.

Open question for this path: is each thing we build reusable across firms, or thrown away after one engagement?

Path B — Agency / retainer (bespoke, replicated)

What it is: Not a SaaS. We take on ad-hoc projects, ideally on a retainer. Each year we agree: this much new dev time, this much maintenance; you can always upgrade; we keep all your systems running. For firms with no tech team, we handle everything — including hosting, deployment, credits, the cloud-code instance, the lot. They don’t have to manage any of it.

Then we take what we’ve learned and pitch firm #2, #3, #4 — possibly extending beyond the AI agent into broader back-office automation (e.g. even answering phone calls, ops automation).

Why it’s attractive:

• Bespoke and tailored to each firm’s workflow — high perceived value, less competition.
• Retainer gives steadier cash flow.
• We can run a small team in maintenance mode while we keep building.
• We accumulate enough domain understanding to sell to the next client credibly.

What concerns me:

• It’s service-heavy — bound by our own hours.
• Risk of becoming a body shop; churn if a client builds its own internal team.
• Margins lower than product.

Open question for this path: how much of each build deposits into a reusable core vs. being one-off labour?

Path C — SaaS (AI-first product)

What it is: Build an actual product. AI-first, differentiated from everything on the market today (task management / tracking software with real automation on top — e.g. “this work was planned today, but X issue happened, so we couldn’t do Y”).

Why it’s attractive:

• Biggest payout if it works.

What concerns me — and this is the heaviest list:

• From what I can tell, what they actually want is a full-fledged task-management + tracking system before the automation layer even makes sense. That’s a 1–2 year build, maybe longer, to understand the industry and go deep.
• The glamorous, well-funded AI problems get taken by people in California. A niche vertical solution may see little money for a long time.
• The 2-year clone risk: if, after two years, a bigger player enters, we may have made nothing. This is the fear that has to be de-risked before we commit.
• It demands the most from the founders we’d partner with — time, attention, access.

Open question for this path: can we de-risk a multi-year build enough that a late-entering competitor doesn’t zero us out?

4. Cross-cutting concerns (apply to all paths)

These are the realities that color every option:

• Founder time & motivation. Our prospective partners are UK-based and already running large construction firms. How much time will they realistically sit with us? How motivated are they?
• Access / contacts. Will they open their network to us? That network is what makes paths B and C scale.
• Logistics & cost. Do we need to keep flying to the UK? Who foots that bill? What’s the realistic budget to solve the problems they’ll put in front of us?
• Time horizon vs. cash. Consulting pays now; SaaS pays late, if at all. Agency sits in between.
• Capital strategy. Bootstrap vs. raise — and each implies a very different operating reality.

5. The questions we need to answer ourselves

Organized by what each answer unlocks. I want us to come out of the next couple of calls (with each other and with the firms) able to fill these in.

A. Opportunity sizing

• What is the cost of the problem to them today, in £/month or £/year? (He mentioned a legal issue that would cost a few hundred thousand to handle — that kind of number is our value anchor.)
• If we solve their problems to ~90% with AI, what’s the realistic ceiling on what one firm would pay annually?
• How many firms of this size exist that have the same problem? (i.e. can we replicate to 10–15 others?)

B. Pricing / willingness to pay

For each problem, ask the firm directly:

• “If we solved this to 90% with AI, what would you sign a cheque for today?” — knowing us.
• “If this came to you via a close referral, but from someone who isn’t us — what would you pay?” — strips the friend-discount and gives us a truer market price.
• Is this spend from an existing software/ops budget line, or net-new money they’d have to justify?

C. Replicability (does consulting/agency secretly become product?)

• What fraction of what we build for firm #1 is reusable for firms #2–#4?
• Is the workflow we’re automating standard across firms of this size, or unique to each one?
• Which problems recur across every firm vs. which are one-offs?

D. Client access & motivation

• How many hours/week can the founder actually give us in the first month?
• Will they introduce us to peers — and let us name them as a reference client?
• Do they want to just use what we build, or do they want a stake in us selling it onward?

E. Logistics & cost to serve

• Can the work be done remote, or does it require on-site presence?
• If on-site: who pays for travel, and is it baked into the engagement?
• What does it cost us (time + infra) to keep one firm’s systems running in maintenance mode?

F. Capital strategy

• If this is bootstrap: what does the cash-flow runway look like from consulting/agency revenue alone?
• If we’d need to raise: at what milestone, how much, and on what kind of business (services P&L vs. product traction)?
• Which game are we actually playing — a profitable agency/lifestyle business, or a venture-scale SaaS? (These imply different capital, different risk appetite, different exits — and we should pick on purpose, not drift into one.)

6. What a good outcome of this exercise looks like

By the end of the next couple of weeks, we should have:

1. The risk/reward table in §2 filled in with real ranges, not blanks.
2. A populated set of the §5 questions — at least from firm #1, ideally one cold-ish data point via a referral.
3. A clear read on the three cross-cutting unknowns that gate everything: founder time, network access, and cost-to-serve.

I want us to be tactical, not generic. Not “go run the mom test.” The concrete question is: what is the lowest-risk path to maximizing what we can earn here — and we can only answer it once these blanks are numbers.

Curious where your read differs from mine on any of the paths, and which of these questions you’d add, cut, or reorder.