The Thesis
Not “AI agents for construction” (too vague, too easy to copy) — a Commercial Memory Layer: capture what happened, prove why it matters, turn it into commercial action, and reuse the history to price the next job better.
The four verbs map to the five-layer model:
- Capture = the front door — voice, photos, delivery notes.
- Prove = evidence — timestamps, source links, contemporaneous record.
- Commercial action = the money — variation, notice, RFI, entitlement flag.
- Reuse = the moat — history that compounds.
The thesis is none of these layers individually. It is the chain connecting all four plus the refusal to sell only the cheap, crowded front.
The core object: the commercial event
The product is about the commercial event — not the diary, tender, RFI, or dashboard. A commercial event is a single thing that happened on a project with money attached:
- late delivery → idle labour
- verbal instruction → should-be variation
- design conflict → stopped work
Everything else (diary, photo, RFI, dashboard view) is a representation of, or pointer to, a commercial event. Picking it as the atomic unit lets one product serve the field, the QS office, and next-bid pricing — three different questions asked of the same objects.
Sequencing rule: start with commercial actions, not admin automation
| Bad (admin) | Good (P&L line) |
|---|---|
| “We use AI to make site diaries easier / summarize tender documents / build dashboards for project visibility." | "We help contractors miss fewer tender risks, capture more recoverable variations, defend delay events with evidence, and price future jobs from their own history.” |
The chapter-one filter — direct, quantifiable money — rejects the first column and keeps the second.
The value chain
capture → structured commercial event → entitlement position → evidence → recovery → cross-firm cost memory
- capture without the event = admin
- event without entitlement = a log
- entitlement without evidence = an opinion
- evidence without a recovery route = a filing cabinet
- recovery repeated across enough projects = cost memory becomes real
The product is the whole chain. The business occupies it one defensible link at a time.
Status: hypothesis, not fact
“We should treat the whole plan as a hypothesis.” A thesis this specific is only worth holding once stress-tested against the real market — who builds these things, what they charge, where they stop, whether the empty space is genuinely empty. That stress test is the research program.