When a Fixed Price Is the Wrong Instrument

Why this matters

Shops that lose money on fixed-price work almost never lose it because the estimate was bad. They lose it because the job type has a wider spread of outcomes than the margin can absorb, and no amount of estimating skill fixes that. A fixed price is not a prediction. It is a bet that your variance is smaller than your margin, and the whole question of whether to offer one has a number attached that most shops never compute. Once you can compute it, a set of jobs you have been quoting fixed for years turns out to be jobs you should never have quoted fixed, and a set you have been running on time and materials turns out to be jobs where a fixed price is free money.

A sibling compares the three contract structures side by side. See related: Time and Materials vs Fixed Bid vs Not-to-Exceed Decision Matrix. This card is narrower: it is the arithmetic that decides whether the fixed option is on the table at all.

What a fixed price actually is

When you fix a price you are not promising the customer that the job will take the hours you estimated. You are selling them certainty and keeping the difference between the estimate and the outcome, in both directions. On the jobs that run short you win, on the jobs that run long you lose, and the deal is sound as long as the long ones do not run further than your margin reaches.

That reframes the estimating question completely. The relevant number is not "am I close on the average." It is "how far wrong can I be on the bad end, and how far wrong can I afford to be."

The gate: your tolerance is your margin over one minus your margin

Here is the arithmetic, and it is short enough to do standing up.

On a labor-dominant job with a fixed sold price, where the overrun is technician hours and the materials do not move, every extra hour worked adds cost but no revenue. If your gross margin on that job type is m expressed as a fraction, then the cost of one sold hour is (1 - m), and the margin you have to spend is m times the sold hours. Divide one by the other and the breakeven overrun, as a percentage of the estimate, is:

tolerance = m / (1 - m)

Read it at a few points, because the shape is the lesson:

State the conditions that formula was derived under, because they matter: it assumes a labor-dominant job, a fixed sold price, materials unchanged by the overrun, and no additional mobilization. A parts-heavy job carrying thin parts margin has less labor margin available to absorb hours, so its real tolerance is lower than the table suggests. An overrun that forces a return trip carries travel and setup on top and lowers it further.

Now the gate itself. Pull the upper end of your plausible hour range for this job type - the outcome you hit roughly one job in ten, not the worst thing you can imagine - express it as a percentage over your estimate, and compare it to the tolerance. If your upper-decile overrun is bigger than your tolerance, a fixed price is the wrong instrument, however good your estimate is.

Case one: passes the gate comfortably

A repeat service job type. The shop has twenty-two closed jobs of this type with recorded hours, estimated at 6.0 hours, actuals running from 5.0 to 7.5, with the upper decile around 7.0. Gross margin on this job type is 38%, and it is labor-dominant.

• Upper-decile overrun: 1.0 hour on 6.0, which is 17%.
• Tolerance at 38%: 0.38 / 0.62, which is 61%.

Seventeen against sixty-one is not close. Even at the bad end this job still returns real margin: the extra hour consumes 0.62 of a billing-hour out of the 2.28 billing-hours of margin in the job, leaving 1.66, which is about 28% of the price. The shop should quote this fixed every time, and should notice something else - a job type with this much headroom is a candidate for a slightly sharper price, because the certainty it is selling costs very little to provide.

Case two: same shop, same trade, fails the gate

A project version of similar work. Estimate 20.0 hours, plausible outcomes from 18.0 to 34.0 hours, upper decile around 32.0. Gross margin 30%.

• Upper-decile overrun: 12.0 hours on 20.0, which is 60%.
• Tolerance at 30%: 0.30 / 0.70, which is 43%.

Sixty against forty-three, and the gate fails. Carry it to the end so the size is visible: at the upper decile the extra 12.0 hours cost 8.4 billing-hours against 6.0 billing-hours of margin, so the job finishes 2.4 billing-hours underwater, which is 12% of the sold price. Not a thin job. A loss, on a job the shop estimated as well as anyone could.

Notice what is not wrong here. The 20.0-hour estimate is the best number available, and on the typical job it is right. The instrument is what fails. Fixing the estimate cannot help, because the estimate is not the problem: the distribution is wide and the margin is not.

The shop has three legitimate moves and no fourth. Raise the margin on this job type until the tolerance covers the spread, which at a 60% spread means a margin near 38%. Narrow the spread by buying information before quoting, which is a separate method. Or change the instrument to a not-to-exceed or a time-and-materials arrangement with a cap. See related: How to Price Risk You Can Name but Cannot Size, and What Time and Materials Actually Protects.

The three ways shops fake passing the gate

Padding the estimate. Quote 26.0 hours instead of 20.0 and the arithmetic looks better. It does not survive the market: on the large majority of these jobs, the ones that land near the 20.0-hour typical case, you are 30% high and you lose them. Padding does not widen your margin, it narrows your win rate, and the jobs you keep winning at the padded price are disproportionately the bad ones, which is the worst possible selection.

Counting on change orders. The plan is a fixed price plus recovery through variations when things go sideways. This works when a genuine scope change occurs and fails when the overrun is your own scope taking longer than expected, which is the case the gate is about. A change order needs a change. Your own hours running long is not one, and on a residential fixed-price agreement that is generally the end of it. Commercial and public forms sometimes provide routes to relief for conditions materially different from what was represented, but those are terms in that contract rather than a background right, and they do not cover an estimate that was simply optimistic.

Betting the relationship. "They are a good customer, they will make it right." Sometimes true, and it is not an instrument, because it is unenforceable and it draws down goodwill you were saving for something else.

What the customer is buying, and why they will pay for it

Do not read the gate as an argument against fixed pricing. Certainty is genuinely valuable to a customer, and on a job type that passes the gate you should sell it hard. A homeowner comparing a fixed number against an hourly rate is comparing a decision they can make with a decision they cannot, and the fixed number wins even when it is higher.

The point of the gate is to know which product you are selling. On a passing job type you sell certainty and price it. On a failing one, selling certainty means selling something you cannot deliver, and the customer finds that out in the middle of the job, which is the worst place for both of you.

The pressure a fixed price puts on the tech, and the rule that answers it

There is one hazard this instrument creates and it is worth naming plainly. A tech who knows the job is fixed and running long is under pressure to compress, and the steps that compress most easily are verification steps: the combustion or leak check, the proving sequence before contact, the operational test at the end. That is a hazard the pricing decision created, upstream of anybody touching equipment.

So the rule that comes with a fixed price is a stop rule, and it belongs in the shop's policy rather than in the quote: verification steps are never inside the variable part of a job. A tech does not de-scope proving dead before contact under 29 CFR 1910.333(b)(2), does not skip isolation and release of stored energy under 29 CFR 1910.147 before opening a machine, and does not omit a combustion or leak verification because the hours are gone. If the hours are gone, that is the shop's loss and the estimate's lesson, and the tech is told so out loud rather than left to infer it.

The contract side, which varies more than the arithmetic does

The formula is the same everywhere. What surrounds it is not. Commercial and public construction forms commonly include differing-site-conditions and change-order provisions that give a contractor a route to relief when conditions turn out materially different from what was represented, and those provisions are contract terms, not a background right. A residential fixed-price agreement generally has whatever it says and nothing more.

Several states also regulate the content of residential home improvement contracts, including what must be in writing, how changes must be documented, and what a contractor may collect and when, and the requirements differ substantially between states and between consumer and commercial customers. Have your own attorney review the form you use in each state you work in rather than assuming a clause you copied does what its heading says.

Verifying you applied the gate correctly

• Did you compute the upper decile from recorded actuals on this job type, or from memory? Memory reliably reports the typical case and forgets the tail.
• Did you use the margin on this job type, not the shop blend? A job type below the blend has less tolerance than the blend implies.
• Is the job labor-dominant? If a large share of the price is parts at thinner margin, recompute rather than reading the table.
• Would an overrun trigger a second mobilization? If so, subtract that from your tolerance before comparing.
• If the gate failed, did you change the instrument, or did you change the estimate and tell yourself that was the same thing?

References

• See related: Time and Materials vs Fixed Bid vs Not-to-Exceed Decision Matrix; What Time and Materials Actually Protects; How to Price Risk You Can Name but Cannot Size; The Contingency Line and How to Size It
• 29 CFR 1910.333(b)(2) and 29 CFR 1910.147, cited here for the verification steps that must stay outside any commercial compression
• Confirm residential home improvement contract requirements with your own attorney for each state you operate in; requirements and consumer/commercial scope vary by state