The Landscape
Everyone reads documents; recovery-at-scale and cross-firm cost are the two empty columns — but the honest correction is that recovery is contested, not empty.
60 competitors scored on one 21-area grid. Two columns near-empty across the field. Cross-competitor heatmap.
| Empty column | What it is |
|---|---|
| Col 15 | Change / variations / claims / entitlement recovery — substantiated event → recovered money |
| Col 21 | Cross-firm historical cost — pricing the next job against what the whole market actually paid |
Everyone reads — nobody recovers (TRUE half, stays true)
| Tier | Companies | Captures | Stops at |
|---|---|---|---|
| AI-native frontier | Document Crunch, Trunk Tools, Civils.ai, Pype | document reading / extraction / coordination | ”here is the risk, here is the log, here is the notice” |
| Field-capture | Raken, PlanRadar, BuildPass | beautiful field capture | the diary |
Converting captured evidence into recovered money — the adversarial, P&L-bearing step — is where they stop by their own product boundaries.
”No one recovers money” is FALSE
Found by hunting the claims-and-delay field head-on, not inferring from general tools. A real, sub-scale recovery cohort exists:
| Vendor | Motion | Price |
|---|---|---|
| Magra | Almost the exact notice-to-recovery motion: deadline radar, evidence ingestion, claim drafting, Eichleay + measured-mile quantum. Owner and contractor. | ~$36k/yr |
| ClaimDD / ClaimEOT, Masin AI, Delay Claim Builder | Genuinely assembling cost, time, causation into quantified recoverable claims | — |
Absolute claim retired: “Area 15 is more contested than the synthesis language still sometimes implies… The opportunity is not ‘no one quantifies claims.’”
The precise claim (both reviews converged)
“no scaled, trusted, jurisdiction-specific recovery platform with cross-firm outcome data.”
| Word | Why it holds |
|---|---|
| Scaled | Cohort is all 2024–25 micro-entrants, unfunded/seed, some one person |
| Trusted + jurisdiction-specific | Recovery judgement depends on contract form + payment law; credibility earned, not shipped |
| Cross-firm outcome data | The genuinely empty column — empty for a structural reason |
Col 21 is empty BY DESIGN
AI benchmarking entrants — Gauge, Rate QS, BenchIt — all explicitly single-firm. Rate QS markets “your private data, not cross-firm” as a selling point. Vendors are building cost benchmarking; none pool across firms. Cross-firm pooling is a trust + competition-law problem, not a modelling one. “Empty by design — and that is the tell.” Demand proven, gap unfilled → durable moat for whoever earns the trust to pool it.
Two structural patterns
- The seam is real market structure. QS commercial suites (Causeway, Eque2, RedSky) run the money workflow to variation valuation / final account, then stop. Delay & disruption entitlement is sold by a separate forensic-claims vendor category, never bundled. The wedge sits on that join.
- Absorption is the default exit, proven 4x. Pype→Autodesk, Payapps→Autodesk, Document Crunch→Trimble, Disperse→OpenSpace. Trunk Tools’ Procore API revoked while Procore shipped a rival agent. An independent AI layer on a platform’s data + buyer is rented land. See landscape synthesis; supply side (Sablono et al.) mapped there.
Net read
Build alongside open-API field tools (not on a platform). Consume the evidence tools (don’t rebuild). Flank the document neighbours into the money. Avoid distribution-owners head-on. Which jurisdiction, in what vehicle → the decision.