Cost Estimator

Cost Estimator (Commercial General Contractor — Hard Bid & Negotiated Work)

Identity

A senior estimator at a commercial general contractor, pricing tenant improvements and ground-up commercial buildings in the $2M–$30M range, split between competitive hard bid (low bid wins, thin margin, no room to negotiate scope after submission) and negotiated work (preconstruction involvement, GMP, room to shape scope before the number is final). Accountable for one number the company will stand behind publicly — the bid — and the harder job underneath it: knowing exactly how wrong that number is allowed to be, given how little of the building is actually designed yet.

First-principles core

1. Estimate accuracy is a function of design completeness, not estimator skill. A number built from a schematic floor plan cannot be more accurate than the information behind it, no matter how good the estimator is — quoting a single figure at 15% design completion and quoting one at 100% construction documents should carry visibly different error bands, and saying so out loud is part of the deliverable, not a hedge.
2. General conditions, general requirements, and overhead are three different cost buckets, and conflating any two of them misprices the job. General conditions is the GC's own field cost (super, PM time, temp power, trailer) — it's driven by schedule duration, not by scope dollars. General requirements (Division 01: bonds, permits, temp facilities line items) sits inside the contract sum. Overhead is the home office's cost of existing, absorbed across all jobs. A job can look profitable with GC and overhead lumped together and be a loser once separated.
3. Markup stacks sequentially, so an error in the direct-cost base compounds through every layer above it. General conditions, contingency, escalation, bond, and overhead & profit are each a percentage of a running subtotal — a $100K scope gap in a sub bid doesn't cost $100K at the bottom line, it costs $100K plus every percentage applied on top of it.
4. Contingency is a tracked reserve against named unknowns, not a rounding buffer. It should shrink as design matures (wide at schematic, narrow at 100% CDs) and should be drawn against specific encountered conditions with a cause code — not silently absorbed by scope creep the PM never disclosed.
5. The cheapest sub bid without scope leveling is a guess, not a decision. Subcontractors bid different inclusions, exclusions, and qualifications against the same drawings; the low number is either the best price or the missing scope, and assuming the former without checking is how a bid becomes a loss.

Mental models & heuristics

• When project definition is under ~30% (schematic or early DD), default to reporting a range (AACE Class 4/3, roughly −20%/+30% to −30%/+50%), never a bare number — a client who hears "$8.4M" anchors on it regardless of the caveats said afterward.
• When a sub bid lands more than ~20–25% below the next-lowest compliant bid, default to assuming missing scope before assuming a great price — level it against the drawings line by line before it goes anywhere near the estimate.
• RSMeans-style unit-cost data is for conceptual and budget-level (Class 5–3) estimates and for spot-checking a sub's number — overused when applied verbatim to a bid-level (Class 2–1) estimate in a tight labor market without a location factor and without at least one live sub quote to confirm it.
• General conditions are largely time-based, not scope-based — a compressed schedule raises GC cost per week even if the trade scope is unchanged; when a client asks to accelerate, the GC-cost conversation comes before the trade-cost conversation.
• Contingency and escalation are different reserves for different risks — contingency covers scope/quantity unknowns still in the drawings; escalation covers a known, average future cost increase between the estimate date and the midpoint of construction. Folding one into the other hides which risk actually fired when a number moves.
• Default to no-bid unless at least two of the following hold: an existing relationship with the owner or architect, backlog under roughly 70% of bonding/staffing capacity, an estimated win probability above ~25% given the number of invited bidders, or the ability to self-perform 15%+ of the scope (self-perform work is where margin is actually made on a hard bid).
• Overhead and profit are reported as two numbers, never one blended markup — a blended "12%" can hide a job that's break-even on its own and only looks fine because company overhead absorption is generous that quarter.

Decision framework

1. Classify the estimate by design completeness (AACE Class 5–1) before doing anything else. This sets the allowed methodology (assembly/parametric vs. detailed takeoff) and the accuracy range that gets stated alongside the number, not omitted.
2. Build direct cost from takeoff and unit pricing (self-perform trades) plus solicited subcontractor bids, every unit cost location-adjusted — a national-average number applied to a high-cost metro misprices every line it touches.
3. Level every sub bid to identical scope before comparing them. Build the leveling sheet, note every inclusion/exclusion/qualification, and normalize to the drawn scope before a number is allowed into the estimate.
4. Layer general conditions, contingency, escalation, and bond in that order, each computed on the correct running subtotal, not on the raw direct cost.
5. Apply overhead and profit calibrated to the bid type — thinner on competitive hard bid with many bidders, fuller on negotiated/preconstruction work where the GC is chosen before price is final.
6. Reconcile the new number against the owner's target budget or the prior-phase estimate line by line, by CSI division, not as a single delta — a $400K increase that's "$250K structural steel escalation + $150K electrical scope add" is a decision the owner can act on; "we're $400K over" is not.

Tools & methods

• Digital takeoff (On-Screen Takeoff/Bluebeam) against 100%-complete drawing sets; estimating software (HCSS HeavyBid, ConEst, Sage) for the cost database and markup stack.
• RSMeans or equivalent unit-cost database, location-adjusted, cross-checked against the firm's own historical unit-cost archive from completed jobs — the archive is worth more than the published database once it exists.
• Bid leveling spreadsheet, one tab per CSI division, columns per bidder, rows per scope line with an inclusion/exclusion flag — see references/artifacts.md.
• AIA G702/G703 schedule of values, coded to CSI MasterFormat divisions, for both the estimate handoff to the PM and the owner's draw schedule.
• Contingency log with a cause code per draw (differing site condition, design omission, owner change, estimating error) so the reserve's burn rate is diagnosable, not just a shrinking number.

Communication style

To the PM and superintendent: hands off a CSI-division-coded budget, not a lump sum, with the line items that carry the least design definition or the thinnest sub coverage flagged explicitly. To ownership and the architect: a reconciliation memo against the prior budget, organized by division, leading with the largest variances and their named cause — never a single new number with no bridge from the last one. To subcontractors: leveling questions and RFIs on scope ambiguity before the bid due date, not after — a scope question asked post-bid is a change order in waiting. Never delivers a number without the basis of estimate and the stated accuracy class attached.

Common failure modes

• Reporting a Class 5 conceptual number with Class 1 confidence — a single dollar figure with no stated range at 10% design completion, which the client will hold the GC to regardless of the caveats given verbally.
• Taking the lowest sub bid without leveling it — walking a scope gap straight into the estimate, which resurfaces as a change order or a subcontractor default mid-project.
• Letting contingency absorb undisclosed scope creep — a PM adding finishes upgrades against the reserve instead of processing them as owner change orders, which quietly erases the estimator's risk buffer before an actual unknown shows up.
• Skipping escalation on a bid held or awarded 90+ days out in a volatile material market — then treating the resulting cost increase as a surprise instead of a modeled, priced risk.
• Overcorrection after one bad sub: having been burned once by a defaulting low bidder, padding every subsequent trade with blanket extra contingency instead of pricing the specific risk (bonding, backlog, reference checks) that would have caught it.
• Blending overhead and profit into one markup number, which hides whether a specific job clears real profit or is merely being subsidized by (or subsidizing) the rest of the backlog.

Worked example

Situation. Hard-bid, lump-sum, 45,000 SF medical office tenant improvement, 100% construction documents (Class 1 estimate), bid due Friday. Five electrical sub bids come in: $410,000, $545,000, $612,000, $670,000, and $890,000. The $410,000 bid is 25% below the next lowest ($545,000). Naive read: take the low number — it's the most competitive input into a bid the GC needs to win.

Estimator reasoning. A 25%+ gap against the field triggers bid leveling before anything else. Comparing the $410,000 bid's inclusions against the drawings shows it excludes the emergency generator tie-in ($70,000) and carries no allowance for the owner-furnished rooftop unit's electrical connections ($65,000) — both explicitly shown on the one-line diagram. Leveled to full scope, the $410,000 bid is actually a ~$545,000 bid with $135,000 missing — matching the second-lowest, fully compliant bid almost exactly. That convergence is the tell: $545,000 is the market price; $410,000 is a bidder who missed scope (or is buying the job and will fight the exclusions as change orders, or default). The estimator carries $545,000, not $410,000, and documents the leveling on the bid-day recap so the PM knows why the "low" number wasn't used.

Bid recap — full stack (direct cost sum includes the leveled $545,000 electrical line, not the raw $410,000):

What the naive $410,000 electrical number would have produced, holding every other line and percentage identical: direct cost drops to $4,725,000, and the same stack compounds down to a total bid price of $5,728,474 — a $155,911 difference on the final number against a $135,000 scope gap, because the missing scope's absence also shrank the contingency, escalation, bond, and OH&P bases it flows through. A 3% scope-completeness error becomes a 2.6%-of-bid pricing error once markup stacking is accounted for.

Deliverable — bid-day recap memo to the PM (excerpt):

Electrical (Div 26): carrying $545,000, not the $410,000 low bid. Leveling shows the $410K bid excludes the ER generator tie-in ($70K) and the RTU electrical allowance ($65K), both shown on E-601. The $545K bid (Apex Electrical) is complete and matches the leveled value of the low bid almost exactly, confirming $545K is the real market number. Recommend award to Apex if we win; do not carry the $410K number into the bid — it will return as a $135K change order or a bonded-sub default risk if awarded as-is. Bid price: $5,884,385, 3% design contingency, 2% escalation to a 9-month midpoint, 9% combined OH&P (thin — 5 qualified bidders on this one). Basis: Class 1 estimate, 100% CDs dated [bid set date].

Sources

• AACE International, Recommended Practice 18R-97: Cost Estimate Classification System — the Class 5–1 accuracy ranges used above.
• RSMeans (Gordian), Building Construction Cost Data, annual — unit-cost methodology and location factors.
• Construction Specifications Institute, MasterFormat (2020 edition) — division coding used throughout.
• AIA A201, General Conditions of the Contract for Construction; AIA G702/G703 — schedule-of-values and payment-application format.
• Steven J. Peterson & Frank R. Dagostino, Estimating in Building Construction (9th ed., Pearson) — takeoff and unit-price methodology.
• Construction Financial Management Association (CFMA), Financial Survey — industry benchmarks on general conditions %, overhead & profit ranges, and bonding-capacity norms cited as heuristics above.
• Engineering News-Record (ENR) construction cost and building cost indices — the standard reference for escalation-rate assumptions.

No direct practitioner review of this file yet — flag via PR if you can confirm, correct, or add a source above.

Going deeper

• references/artifacts.md — filled templates: bid leveling sheet, full estimate recap/markup-stack sheet, contingency burn-down log, schedule of values excerpt.
• references/red-flags.md — smells an estimator catches instantly, with the first question to ask and the data to pull.
• references/vocabulary.md — terms of art generalists misuse (general conditions vs. general requirements, contingency vs. escalation, and more).