Transparent pricing: $0.05 per square foot of plans you run. See what that costs on a real house

Product · Accuracy

Producing a number is the easy half

Any tool can return a quantity. The question worth asking a vendor is what happens when it's wrong — whether anything in the system is trying to catch it, and whether you can tell. Here is exactly what we do.

Layer one

An audit that argues with the takeoff

After the takeoff is produced, a separate pass re-derives it from the plans without trusting a single number in it — the way a senior estimator checks a junior's work before the bid goes out.

What it re-counts

Studs from wall lengths and spacing. Joists and rafters from spans. Headers opening by opening against the window and door schedules. Sheathing area against sheet counts. Hardware against the details that require it.

What it hunts for

Omissions — any framing member, hardware family or consumable the plans require that the takeoff lacks entirely — and duplicates, where the same material got counted in two packs.

Error-level findings block the quote from being sent until someone accepts or dismisses each one. Catching an error and letting it ship is worse than not catching it.

Layer two

Guards on identity and on price

A pricing engine matches text. Text similarity is not the same question as “is this the same product”, and the two come apart in expensive ways.

Identity

Fire-rated gypsum shaftliner and a concrete forming tube can describe alike and cost within a percent of each other — so a total can look right while the material is completely wrong. Sheet goods can't bind to a furring strip, panel sizes must agree, lumber can't bind to hardware, and both sides must share a material term.

Price

Every unit price is checked against what that class of material actually costs in your own catalog. A connector priced like a truss package — a real failure we found and fixed — is flagged for review, not quoted.

Layer three

It learns from your estimators

Every correction an estimator makes is recorded — which lines they edited, by how much, what they added by hand and what they deleted. Those corrections are distilled into directives that steer later takeoffs for your yard specifically.

Measured, not asserted

Edit rate, median quantity correction and dollar variance are recorded on every sent quote. Whether the tool is improving is a number you can look at, not a claim we make.

Your yard's conventions

Waste factors, preferred lengths, what you carry and what you never stock. The tenth bid should need less fixing than the first because it learned from the nine before it.

Nothing shared between yards

Directives learned from your corrections apply to your account only. Your pricing and your practices are not training material for a competitor.

Layer four

New models are tested before they touch a bid

When a better AI model ships, it doesn't just get switched on. It runs invisibly alongside the current one and is judged on evidence that can't be gamed: did it produce the lines your estimator had to add by hand, and did it avoid the ones they deleted? A model only takes over when a human approves the evidence.

An estimator still signs the bid

None of this replaces judgement, and we don't pretend it does. BidLoggers produces a defensible draft and shows its reasoning so a professional can check it quickly — which is the difference between a tool you can use and one you have to re-do.