How we size and de-risk this — working note

Working note, partner-to-partner. This is deliberately not an answer or a recommendation. It’s how I think we should go about figuring out whether there’s a real business here, how big it could be, and how to slot in with the least risk for the most value. Everything below is a way to turn our guesses into evidence. Click any link to go deeper.

Jump to: The frame · What we have and what we lack · The three ways we could slot in · How we assess it · The questions we need to answer · What we do not do yet · Next moves

The frame

Bro — I think there’s a real opportunity here. But right now it’s unsized. We don’t know what the problems are worth, what anyone would actually pay, how big the market is, or how much of it a two-person team can realistically take. So the job for the next few weeks is not “pick the business model.” The job is:

1. Size it — put real numbers on the reward for each way we could play this.
2. De-risk it — figure out where the risk actually lives and buy information to shrink it before we commit real time.
3. Then slot in — choose the entry point that gives us the most value per unit of risk.

De-risking, concretely, means spending cheap information to avoid expensive mistakes. The cheapest, highest-signal information available to us right now is a small number of well-run conversations with people inside the industry and one investor who sees the whole market. This note is mostly about how to make those conversations count.

The whole thing reduces to one table we can’t fill in yet:

Every ? is a research task, not a debate. When the table is full, the decision mostly makes itself. A high-reward / high-risk row can still beat a safe / low row — we just need the numbers to judge, not vibes.

What we have and what we lack

Have: a working, deployed AI field-intelligence demo; strong AI + engineering ability; warm access to one or two senior people inside real UK construction firms, and possibly an investor.

Lack: any reliable read on what these firms’ problems are worth, realistic deal sizes, what a product could charge, what budget exists, or what they’d truly adopt versus politely nod at. That gap is the whole game.

The three ways we could slot in

These are the same three paths from the three-paths note — restating them here as slots, each with the shape of its risk and reward and, crucially, the unknowns that decide it. I’m not ranking them.

Slot 1 — Consulting (in and out)

• Reward shape: one-off fees, fast cash, capped. No compounding asset.
• Risk shape: low — worst case we eat some dev time. But no recurring revenue and every client is a fresh sell.
• What decides it (must learn): how much a firm would actually pay to solve one bounded problem, and whether the thing we build for firm #1 is reusable or thrown away. → sizing + reusability questions below.

Slot 2 — Agency / retainer (bespoke, replicated)

• Reward shape: steadier retainer cash, replicated across 10–15 firms, a reusable core forming underneath.
• Risk shape: medium — service-heavy, bound by our hours, risk of becoming a body shop, churn when a client builds its own team.
• What decides it (must learn): whether the same workflow recurs across firms (productizable) or every firm is bespoke (service trap); whether firms will pay a monthly retainer and open their network. → replicability + access questions.

Slot 3 — SaaS (AI-first product)

• Reward shape: biggest payout if it works.
• Risk shape: highest — long build before revenue, we don’t yet know the domain, and a funded competitor could zero us out in ~2 years.
• What decides it (must learn): whether there’s a workflow enough firms want the same way to justify a product, and whether we can defend it long enough to matter. → appetite + ceiling + capital questions.

How we assess it

Three steps, in order:

1. Agree the table. Lock the columns above (reward, probability, downside, time-to-cash) so every conversation is hunting for the same cells.
2. Buy the information cheaply. Run a handful of disciplined interviews with our warm insiders + the investor. This is the de-risking engine. We already have the full interview kit built:
   • The call guide — the runnable script: an ordered set of firm-insider questions and a separate investor track, with an opener and a close for each.
   • The theme-by-theme depth banks to go deeper where a call gets interesting: current tool stack, purchase history & churn, adoption & realized value, spreadsheet/WhatsApp workarounds, budget, authority & procurement, inbound vendors & funding, AI (behavioral), external-help buying, the forward-leaning signal, the natural ceiling, and the investor set what founders pitch, portfolio buy/keep/renew, budget moving vs talk.
   • Interviewing notes — how to pull real numbers out of “it varies,” what to trust vs. discount, and what not to say so we don’t contaminate the answer.
3. The discipline that makes it work: every question is about past and present real behavior — money already spent, tools already bought and dropped, things already tried — never “would you buy our thing.” If we pitch, they’ll be nice, say yes, and we learn nothing. We ask what they do, not what they’d like.

The questions we need to answer ourselves

Grouped by what each group unlocks in the table. The interview kit above is how we get most of these; a few we answer internally.

• Sizing the reward — what the problems cost these firms today, and what they already pay for software. This sets both the ceiling (a conservative firm is anchored to what it already spends) and the value we can price against. → current tool stack, workarounds.
• Pricing / willingness to pay — what they’ve actually bought, kept, churned, and why; what a thing is worth to them behaviorally, not aspirationally. → purchase history & churn, adoption & realized value.
• Access & channel — will they open their network, give intros, act as a reference. Decides whether slot 2/3 can scale. → inbound vendors & funding.
• Appetite for AI / external help — what they’ve already tried with AI and outside dev, and whether this specific person acts on new tech or just talks. This is where the “founder gets it but has no time/expertise, so will pay for help” hypothesis gets tested. → AI (behavioral), external-help buying, forward-leaning signal.
• The ceiling — what makes them not adopt even something good (IT, data, contracts, conservatism). Bounds every reward estimate. → the natural ceiling.
• Budget & authority — who owns the money, who signs, how a new tool actually gets in. → budget, authority & procurement.
• The market view — from the investor: what founders pitch, what got funded vs. passed, what portfolio firms actually buy and keep, where budget really moves. → what founders pitch, portfolio buy/keep/renew, budget moving vs talk.
• Capital strategy (internal) — bootstrap vs. raise, and what each does to the risk/reward of each slot. We decide this from the evidence, not before it. See the three-paths note, §5.F.

The fuller original question set lives in the three-paths note, §5.

What we do not do yet

• Don’t pick a model. The table isn’t full.
• Don’t build anything new. We already have enough to have the conversations.
• Don’t treat the external AI research as the answer. We ran two long GPT-5.5 research passes — they’re useful input to pressure-test, not conclusions. Heads-up: when we fed them our full research corpus, they both converged on the same narrow thesis, which tells me the convergence is our corpus talking, not the market. Treat them as one opinionated voice: the tactical pass and the partner pack. Read them after we form our own view, not before.

Next moves

1. Sign off on the table columns above.
2. Run the interviews with the call guide — one or two firm insiders + the investor.
3. Fill in as many ? cells as the conversations allow; log the numbers.
4. Reconvene, look at the filled table, and then decide which slot we enter and how.

The point isn’t to be right today. It’s to run the cheapest experiment that forces the answer to reveal itself — while we’re getting paid or at least getting smarter, not guessing.