Work / Margin protection

The Review Desk That Surfaced $221,915.70 in Missed Margin

An estimator assembled every bid from five documents that disagreed with each other, and nothing caught the gap before it shipped. A reconciliation pipeline cross-checks those documents automatically and holds every approval with the estimator: $221,915.70 surfaced across $3,442,234.02 reviewed, a 6.4% verified catch rate, zero escapes.

Anonymized. Client, employer, project, manufacturer, part, and pricing identifiers are removed; the mechanism and the numbers are real.

The user and their week

The user is an estimator with five documents open and a bid due: internal quotes, manufacturer quotes, drawing takeoffs, service and labor rules, and the contractor's own bid breakouts. Each input is prepared by a different person, under deadline, in a different format. When those inputs disagree, a lot line gets under-scoped, an accessory drops off, a service is priced from the wrong job, or a breakout total quietly diverges from the takeoff behind it. Catching that means holding all five documents side by side and reconciling them by hand for every bid, so the gaps ship, and they eat margin the company already earned.

Why it mattered

Margin on a commercial job is won or lost before a shovel hits dirt, and the reconciliation work that would catch a mis-scoped line was too slow to do by hand on every bid.

The bet, and what was rejected

The bet was that the estimator's judgment was fine and their attention was the scarce resource, so the product automates the reconciliation and leaves every decision with the estimator. The machine holds the thousands of relationships between documents; the estimator rules on the disagreements it surfaces, each one arriving with its evidence attached.

Two alternatives lost. A spreadsheet lost because the hard part was never arithmetic, it was maintaining consistent relationships across independently maintained documents that keep changing through the estimating process. Full automation lost because a system that prices bids without an approval gate has to be right every time, and a system that only has to be trustworthy about what it flags can start delivering value on day one.

Acceptance criteria, set before build

What the record does support is the shape those criteria took in the finished product: reconcile internal quotes against manufacturer quotes so pass-through pricing lines up instead of drifting, reconcile bid breakouts against the takeoff behind them so a group total cannot silently diverge from the devices it should contain, reconcile line items against a catalog price source so an under-scoped or stale line surfaces instead of shipping, and reconcile service and labor against a rules model so labor is scoped consistently on every bid. Every flagged disagreement had to carry the evidence that triggered it, and the pipeline had to hold every commercial decision with the estimator; it resolves none of them on its own.

What shipped

A bid-reconciliation pipeline that ingests every input to a bid, quotes, manufacturer quotes, drawing takeoffs, service and labor rules, and bid breakouts, and cross-checks them automatically against the four rules above. Each flagged disagreement carries its evidence: which input triggered it and what it was checked against.

The human gate sits at every flag. The estimator reviews each surfaced disagreement and rules on it before the bid goes out; findings link to their source evidence and approval is retained before customer delivery. Nothing reaches a sent quote without that sign-off, which is also why the ledger records zero escapes.

Every flag carries its evidence. Nothing reaches a sent quote until an estimator signs off on it.

What failed, and what changed

The measurement discipline lagged the reconciliation logic itself: the pipeline was already checking bids before the ledger existed to grade it, and the ledger was re-based twice before the numbers on this page were trustworthy enough to publish. The early catches had to be re-traced to their source documents after the fact, work a ledger built alongside the pipeline from day one would have made free.

The catch rate is falling over time, and that reads as the product working: once the desk is checking every bid, the process upstream cleans up, and there is less left to catch.

Result

Across $3,442,234.02 of bid work reviewed, the pipeline has surfaced $221,915.70 in missed revenue and estimating errors, a verified catch rate of 6.4%. The ledger itemizes 34 jobs, traces every line item to a source document, and records zero errors escaping into a sent quote.

Adoption, and what carries forward

What the record does support: the pipeline has been in production since 6/28/2026, reconciling every job that passes through the desk, 34 of them itemized in the current ledger.

What carries to the next engagement is the pattern, which survives any change of tools: ingest the scattered inputs that make up an estimate wherever they live, reconcile them against each other on rules that encode how a correct bid has to hang together, surface only the disagreements with evidence attached, and keep a human at the approval gate. The same layer sits on top of a GC's existing estimating, project-management, and ERP stack without requiring authority over the estimating team using it.