Two estimators take off the same set and return different quantities

When two competent estimators measure one set and return different quantities, they have usually measured different things. The drawings were the same; the rules were not. What counts as inside a condition, where one trade's scope stops, whether an item is tallied per symbol or per assembly, and how much waste is carried — none of that is printed on a drawing, and none of it survives in anybody's head between jobs.

The instinct is to compare bid totals and relax when they land close. Resist it. A total is the sum of every condition, so an over-measure on partitions can cancel an under-measure on ceilings and produce a number that looks like agreement while both underlying quantities are unbuyable. The comparison has to happen one level down.

Compare condition by condition, never total to total

Put the two takeoffs side by side at condition level, with a percentage difference and a blank column for the explanation. What you are looking for is shape: variance concentrated in two or three rows is a rules problem you can fix this afternoon, variance spread evenly across forty rows is a scale or method problem and a different investigation entirely.

Five of those six rows are explained without anyone re-measuring anything. That is the normal outcome: the disagreement is about rules and inputs, and the measurement itself is rarely the culprit. The exception is the last row, which is not a variance at all.

First, prove both people measured the same documents

A whole-condition difference — one estimator has a quantity and the other has zero — usually means the two were not looking at the same issue. Addenda land mid-takeoff, someone downloads from the portal on Tuesday and someone else on Thursday, and neither notices. Before anything else, confirm the issue and revision of every sheet each takeoff was based on.

• Record the issue date and revision per sheet on the takeoff itself, not in an email. This is the only way the question is answerable a week later.
• Where an addendum landed mid-exercise, reprice the affected conditions rather than re-measuring everything, using the change-scoped method in repricing an estimate after an addendum.
• Where the two estimators genuinely used different issues, produce the sheet-level change list first — comparing two issues of a sheet set properly — because the number of affected conditions decides whether reconciling is worth more than re-measuring.

The boundary problem, which is most of what remains

The largest and most expensive variances sit where two packages meet, because that is where the drawings stop saying who does what. Both estimators read the same sheet and reached opposite conclusions about whose scope an item falls in — and one of them is going to be right only by accident.

• Interfaces between trades. Fire-stopping at service penetrations, backing steel behind sanitaryware, builder's work in connection with services, and the last stretch of a run between a riser and a terminal unit. Every one is drawn by one consultant and installed by another.
• Vertical versus horizontal splits. Where a package boundary is drawn at a level, at a grid line or at a fire compartment, the item sitting on the line has to be assigned explicitly or it is either counted twice or not at all.
• Provisional and excluded scope. An item deliberately excluded by one estimator and quietly included by the other produces a difference that looks like a measuring error and is actually a commercial position.
• Temporary works, access and hoisting, which appear on no plan and get carried by whoever remembers them.

The remedy is dull and it works: the inclusion rule lives on the condition record, not in the estimator's memory. Scope boundary, sheet scope, deduction rule and unit are exactly the fields set out in what a takeoff condition actually defines, and a condition carrying them produces the same quantity in two different pairs of hands.

Double counting between plan, detail and schedule

The second family of variance is the same item measured twice from two representations of one reality. It always inflates, never deflates, so a consistently higher takeoff is worth checking here first.

• Plan and enlarged plan. A toilet core drawn at 1:50 on its own sheet and again at 1:100 within the floor plan is one toilet core, and only a sheet-scope rule stops it becoming two.
• Plan and schedule. Doors counted off the plan and again from the door schedule, or equipment counted from the layout and again from the equipment schedule.
• Plan and riser or single-line diagram, which is the classic route to an inflated services count.
• Overlapping sheets at a match line, where the same 3 m strip appears on both sheets and gets measured on both. Automated counting makes this failure faster rather than rarer — it is one of the patterns in an automated count that finds devices on some sheets and none on others.

Counted per symbol, or counted per assembly

This one produces the largest arithmetic differences and the least argument once it is spotted, because both estimators are right about what they counted. Take a run of 214 socket outlets drawn as individual symbols. Counted per symbol, the quantity is 214. Counted as assemblies of three-gang faceplates, it is 71 — and if the unit rate is per outlet in one estimate and per faceplate in the other, the two estimates can price the same scope with a threefold gap in quantity and a similar total.

The same trap runs through light fittings on a common driver, sprinkler heads per zone, and structural connections per node. The rule is that the unit on the condition has to match the unit on the cost item, and both estimators must be reading the same one. Where that link is missing, the quantity has to be re-derived by hand on export — which is where mapping takeoff quantities onto cost codes either preserves the definition or loses it.

Waste, laps and the gap between measured and bought

A measured quantity and a purchased quantity are different numbers, and the adjustment between them is an assumption rather than a measurement. On 3,120 m2 of board, 5 percent waste gives 3,276 and 10 percent gives 3,432 — a 156 m2 difference produced entirely by a factor neither estimator wrote down. Reinforcement laps do the same thing on tonnage, and cable pull-in allowances do it on runs.

Turning two numbers into one

1. Agree the document baseline: one issue, one revision per sheet, recorded on the estimate.
2. Rank the variance table by impact and draw a line where the remaining rows cannot move the bid by more than your tolerance. Everything below the line is documented, not resolved.
3. For each row above the line, write the inclusion rule in one sentence and have both estimators agree it before either re-measures. Half the rows close here without a second measurement.
4. Re-measure only the conditions where the rule change alters what is in scope, and record the new figure against the rule rather than against the person.
5. Convert every unresolved difference into a written qualification on the bid form. An unexplained variance is a scope gap, and a scope gap at award becomes a change request — the same reasoning behind treating a suspiciously low bid as a missing-scope signal rather than a bargain.
6. Put the agreed rules back into the condition library so the next job starts with them, and note the estimating hours the reconciliation consumed. Those hours are real, and they belong in the estimating budget set at the bid or no-bid gate.

Two estimators agreeing to within one percent on every condition is a controlled process. Two estimators agreeing on the total while disagreeing on nine conditions is a coincidence that will not repeat.

There is a cost-control consequence as well. Whichever quantity wins becomes the basis of the budget the project manager inherits, and if the winning number was never reconciled to a rule, the reasons for it evaporate at handover. That is one of the mechanisms behind a job budget that stops agreeing with the estimate that won it.

Most of this is automatable in the boring sense: a comparison that lines up two takeoff exports by condition, computes impact-ranked variance and produces the explanation sheet is a small internal build, not a platform, and it is the sort of thing we scope under AI agents and automation. If you are weighing up who builds it, the questions worth asking are in choosing an AI development partner. The rest of this silo sits under preconstruction, takeoff and estimating, within our construction and contracting work.

• takeoff
• estimating
• quality control
• scope boundaries