How to Size a Contingency You Can Defend

Why this matters

A contingency you cannot break into named parts cannot be defended, cannot be spent down, and cannot be improved. It sits on the quote as a percentage, the customer asks what it is for, you say "unexpected conditions," and the honest answer they hear is "padding." Worse, at the end of the job nobody can say whether it was right, because there is nothing to compare the outcome against.

This piece builds a contingency from the bottom up: a short register of named unknowns, each carrying an impact in hours and a likelihood, summed into a number you can explain line by line and delete line by line as the unknowns resolve. A sibling article covers the top-down method, sizing from the spread of your own historical outcomes at a chosen percentile. See related: The Contingency Line and How to Size It. The two are complements. Use the register when the job is unusual enough that your history does not describe it; use the distribution when you have done forty of them.

The bound to understand before you start

A register-built contingency is a portfolio number, not a worst-case cushion. It is the sum of expected values, so on any single job it is almost certainly wrong in one direction or the other. It converges across many jobs, not within one.

That has a direct operating consequence: if any single register line's full impact exceeds the whole contingency, that unknown does not belong in the contingency at all. It belongs in an exclusion with a stated consequence, or in a separate allowance, or the job gets priced with that condition investigated first. A contingency is for the accumulation of moderate unknowns. A single unknown that can swallow the job is a scope decision wearing a percentage.

Step 1: name the unknowns, and only the ones you can name

Write down the specific things about this job you do not know. Not "things could go wrong." Concrete, checkable statements: the routing path through the second floor is not visible; the existing isolation device has not been operated in years and may not seat; the tenant's operating hours may restrict access.

Cap the list at about six. A register longer than that is a sign you are listing worries rather than unknowns, and the long tail will be full of items with negligible expected value that make the total look rigorous without changing it.

What breaks if you skip this: you go straight to a percentage, which is the whole problem. The percentage is an output. Starting from it means you have no line to delete when an unknown resolves and nothing to compare against at close.

Step 2: give each unknown an impact in hours

Not a feeling, not a category. If this thing turns out to be true, how many additional hours of work does it create? Use hours consistently, including for material-driven items where you convert the delay or the rework into the hours it costs you.

Estimate the impact for the typical occurrence, not the worst imaginable one. Worst-case impacts inflate every line and produce a contingency the market will not pay.

What breaks if you skip this: likelihood without impact is a mood. Two unknowns can both be even-odds and differ tenfold in what they cost.

Step 3: give each unknown a coarse likelihood

Use bands, not invented precision. About 1 in 10, about 1 in 5, about 1 in 4, about 1 in 3, about even. If you have real history for a line, use the observed frequency and say so. If you do not, a band is honest and a two-decimal probability is not.

What breaks if you skip this: you end up summing full impacts, which prices the job as though everything goes wrong at once. That number loses every bid, and losing bids on a fantasy is expensive in a way that never shows up in a job costing report.

Step 4: multiply, sum, and express as a percentage of sold hours

Expected hours per line is impact times likelihood. Sum them. Then state the total both ways: as hours, and as a percentage of the job's sold hours, because the percentage is the form the customer and your own margin math will use.

Step 5: resolve what you can before the number goes out

This is the step that separates a register from a formula. Look down the list and ask which lines could be closed by an hour of work before the quote is issued: a second site visit, a phone call to the tenant, opening one access panel, pulling the permit history.

Every line you close comes off the number, which makes your quote more competitive and your risk smaller at the same time. That is the only move in estimating that does both.

If closing a line means opening a concealed space, treat it as work, not as looking. Before cutting any exploratory opening, establish what is in the cavity from the panel schedule and drawings, and treat any concealed conductor as energized until it is proven dead with a meter rated for the circuit, checked on a known live source before and after the test, which is the live-dead-live sequence in NFPA 70E-2021, 120.5, an NFPA consensus standard that binds you through your employer's electrical safety program or where your jurisdiction or your contract has adopted that edition. Cutting or abrading finishes in a building built before 1978 can release lead dust, an inhalation and ingestion hazard: for renovation for compensation in pre-1978 housing and child-occupied facilities, EPA's Renovation, Repair and Painting rule at 40 CFR 745 Subpart E requires a certified firm and lead-safe work practices, so either test the finish first or presume it and work under those practices.

Step 6: keep the register live and spend it down in front of the customer

The register is not a pricing artifact you file. As each unknown resolves during the job, mark it closed and note what the resolution cost against what the line carried. A customer who watches lines close and hears "that one resolved clean, it comes off" reads the contingency as a risk instrument rather than a fee.

Step 7: reconcile at close and feed the numbers back

At job close, compare collected contingency against drawn contingency, line by line, not just in total. The total is the least informative number in the reconciliation, for the reason the worked example below shows.

Worked example: a 60.0 hour job

Four named unknowns on a job selling 60.0 labor hours.

That is 4.92 hours against 60.0 sold, so 8.2 percent.

Check it against the bound stated at the top: the largest single impact is 5.0 hours, and the whole contingency is 4.92 hours. One line can exceed the total. That is a warning, not a disqualification, because 5.0 hours on a 60.0 hour job is 8.3 percent of the job rather than a job-swallowing event. If that line's impact had been 30.0 hours, it would come out of the register and into an exclusion.

Step 5 in practice. Before issuing, the estimator calls the tenant and confirms access is unrestricted, in writing. The access line closes at zero. The register drops to 2.92 hours, which is 4.9 percent of 60.0 sold. One phone call removed 41 percent of the contingency, and the quote goes out lower than a competitor who carried the same worry as an unnamed percentage.

Step 7 in practice, across a year. The shop ran 22 jobs of this type. Register-built contingency averaged 5.1 hours per job, so 112.2 hours collected. Actual draws totalled 98.0 hours. Collected exceeded drawn by 14.2 hours, about 13 percent more collected than drawn, which reads as a well-sized contingency running slightly rich.

Read line by line and that reading falls apart. The routing-obstruction line was carried on all 22 quotes at 1.67 expected hours, so 36.7 hours collected on that line. It actually fired on 9 of the 22 jobs, a frequency of 41 percent against the 33 percent the band assumed, and the average draw when it fired was 6.8 hours against the 5.0 assumed. So that line drew 9 x 6.8 = 61.2 hours against 36.7 collected. It under-collected by 24.5 hours.

The other three lines together collected 112.2 - 36.7 = 75.5 hours and drew 98.0 - 61.2 = 36.8 hours, so they over-collected by 38.7 hours. The comfortable-looking net of 14.2 hours is two large errors in opposite directions cancelling, and a shop reading only the total would conclude nothing needed changing while its most frequent unknown was funded at about 60 percent of what it costs.

The correction, with its step size stated. Unit of analysis is one job of this type. With at least 20 observations, move both parameters straight to the observed values; with fewer than 20, move halfway, because a small sample that swings a parameter fully makes the register oscillate for a year without settling. Here there are 22 observations, so the routing line moves to a likelihood of 2 in 5 and an impact of 6.8 hours, giving 2.72 expected hours. The register becomes 2.72 + 0.75 + 2.00 + 0.50 = 5.97 hours, about 10 percent of a 60.0 hour job, up from 8.2 percent.

The three over-collecting lines get the same treatment, downward, from their own observed frequencies. Do not cut them by feel to keep the total flat. The total was never the thing that was right.

How to verify a register before the quote goes out

Three checks, each under a minute.

• Every line names a thing someone could go and look at. If a line cannot be resolved by any imaginable investigation, it is not an unknown, it is anxiety, and it does not belong on the register.
• No single line's full impact exceeds the register total by a margin that would hurt. If one does, decide deliberately whether it stays or becomes an exclusion, and write down which you chose.
• Every line you could have closed before issuing, you closed. Any line surviving into the quote that a phone call would have killed is a line you are charging the customer to carry on your behalf, and that is the version of contingency that deserves the name padding.

References

• NFPA 70E-2021, 120.5, establishing an electrically safe work condition including the live-dead-live instrument check; a consensus standard that binds through an employer's electrical safety program or through adoption by a jurisdiction or contract
• U.S. Environmental Protection Agency, Renovation, Repair and Painting rule, 40 CFR 745 Subpart E, certified firm and lead-safe work practice requirements for renovation for compensation in pre-1978 target housing and child-occupied facilities
• See related: The Contingency Line and How to Size It; Why a Contingency You Do Not Explain Reads as Padding; What an Assumption in a Quote Is Doing