How a Two-Person Construction Company Runs a Nine-Role AI Review Panel Instead of One Assistant
Case study · Nexen Construction, exterior building envelope subcontractor · PA/NJ/DE/MD/VA/WV/OH/NY
people run the entire company
overall bid win rate across active trades
win rate in the best-fit pricing zone (James Hardie, $250K–$1M bids)
of a disputed debt forgiven in a single negotiation ($8,894.88 of $18,894.88)
differentiated review roles used on every high-stakes decision — bids and legal alike
The problem
Nexen Construction bids EIFS, stone veneer, Nichiha fiber cement panels, Longboard aluminum panels, and James Hardie siding packages, most in the $250K–$1M range, competing against subcontractors with full estimating departments. With only two people managing the entire business — bidding, legal, collections, and delivery — the team could not afford a single point of failure in a six-figure takeoff.
The team’s early attempts to use one general AI assistant for estimating produced exactly the failure mode you’d expect from a single generalist: on the Millbourne bid, the assistant assumed a metal panel spec was a thin panel when it was actually a thick one — a mistake that would have cost real margin if it had gone out uncaught.
What they built instead
Rather than trust one assistant’s single pass, Nexen built a nine-role review panel — each role with a specific, opposing job, all reviewing the same bid before it goes out:
- One role calls out every product zone and does the math (width × height, face by face)
- A second role hunts specifically for what the first role missed — corner returns, parapet backs, canopy soffits, recesses
- A third role's only job is to challenge the other two — flag double-counts, ambiguous scope, and unverified assumptions
- A fourth role tallies everything into one final number
- Separate roles independently check pricing, field feasibility, and legal exposure before anything reaches the business owner for approval
No single role sees its own output as final. The bid doesn’t go out until all nine have had a pass at it — the same structural idea as AIRE’s four dimensions and nine types: different lenses catch different failure modes that one blended view cannot.
What changed, in real numbers
Across active trades, Nexen’s overall win rate is 30.7%. The panel’s pricing checks found the pattern by trade: Hardie, Nichiha, and vinyl work are price-sensitive — pricing to the market wins; EIFS, stone, and metal panel work are not price-sensitive — the losses on those trades were never about price, so the team holds margin there instead of discounting into a loss. On bids above $500K, win rate drops to roughly 3%, so those get sharpened 10–15% or qualified out entirely rather than chased.
The Millbourne panel-thickness error was caught by the challenge-role specifically because that role’s only job is to distrust the first pass — not by a second read from the same kind of assistant that made the error.
A disputed $18,894.88 balance owed to a supplier was negotiated down to a $10,000 settled-in-full agreement with tradeline deletion — $8,894.88 forgiven, a 47% reduction — closed in six days of negotiation starting from a $5,000 opening position.
Why this matters for AIRE Assessment
Nexen didn’t need a name for what they built — they needed the outcome. AIRE Assessment exists to give every team the vocabulary and the structure for what Nexen already proved works: put your Pilot, Analyst, Verifier, and the other six types on the same decision, and you catch what one person (or one AI) alone would have missed. The Nine Types aren’t a personality quiz result — they’re a description of the actual jobs that need to be covered before something expensive goes out the door.