How to Price Risk You Can Name but Cannot Size

Why this matters

There is a risk you can describe perfectly and cannot put a number on. You know exactly what might be behind that wall, in that chase, under that slab. You know it changes the job if it is there. You have no idea whether it is there, and no honest basis for a percentage. Most shops respond by adding a contingency anyway, which feels like risk management and is not: a percentage is a price for spread you have measured, and here you have not measured anything. What follows is the alternative. When you cannot size a risk, you stop trying to price it and you change the instrument that carries it.

A sibling owns contingency sizing itself. See related: The Contingency Line and How to Size It. This article picks up exactly where that one stops working.

Step 1: Write the risk as a trigger and a consequence

One sentence, two clauses: what would have to be true, and what it does to the job. "If the existing connection point is not where the drawing shows it, the run changes and the wall has to be opened in a second location." If you cannot write both clauses, you have a worry rather than a risk, and worries do not get priced, they get looked at.

Skipping this step is why estimates carry a vague uplift on gut feel. An unwritten risk cannot be assigned to anyone, and an unassigned risk belongs to you.

Step 2: Test whether you actually have a distribution

This is the fork the whole method turns on. You have a sizeable risk when you can pull past instances of the same trigger and read what they cost you. Set the bar at a median of at least five prior instances with recorded outcomes, and be strict about what counts as recorded: a job where the file is silent is not a job where the condition was absent.

Run the test honestly and most shops discover they have fewer instances than they thought. That is not a failure, it is the finding, and it selects the method.

Step 3: If you have the distribution, price it and stop reading

Five or more recorded instances, and you size a contingency from your own spread at a percentile you choose deliberately, attached to the named operation rather than to the whole job. That is the sibling article's subject and it does it properly. Do not carry on down this page out of habit; a measured risk priced as an unmeasured one is money left on the table.

Step 4: If you do not, pick the instrument that moves the unknown

Five instruments, each of which puts the unknown somewhere specific. The choice is not about which is safest for you. It is about which one matches who can actually control or absorb the thing.

Most jobs with a named-but-unsized risk want two of these together: a unit rate for the ordinary case and a stop-and-call gate for the case that stops being ordinary.

Step 5: Buy the information when it is cheap against the spread

The instrument nobody offers is the one that usually wins. If two hours of exploratory work removes a spread of many hours, sell those two hours as their own small job with its own scope, and quote the main work afterwards from what you found. This sells better than shops expect on commercial and managed property, where somebody has a budget line for it. It sells badly to a homeowner who has already had two companies out, where it reads as a charge to look. Where it will not sell, price the exclusion and the unit rate instead and say so plainly; the investigation is the better instrument, not the only one.

Where the investigation means opening a building cavity, the opening itself carries hazards your entry block probably did not cover. Scan or probe for concealed conductors and piping before cutting, and treat any conductor found as live until de-energized, locked, tagged and proven dead under 29 CFR 1910.333(b)(2). Cutting or grinding masonry, concrete or mortar releases respirable crystalline silica, which is an inhalation hazard a glove does nothing for: use water suppression or on-tool extraction with respiratory protection under a written program, per 29 CFR 1926.1153 on construction work and 29 CFR 1910.1053 in general industry. In buildings constructed no later than 1980, thermal system insulation and surfacing material are presumed asbestos-containing under 29 CFR 1926.1101 unless shown otherwise, so an exploratory opening that would disturb either stops there and goes to a licensed abatement contractor rather than being cut through to save a trip.

Step 6: Write the trigger so a tech recognizes it without calling you

An instrument is only as good as the field recognition of its trigger. "Unforeseen conditions" is unrecognizable. "More than two lengths of the existing run found corroded through the wall" is something a tech can identify standing there at eight in the morning.

State the trigger in observable terms, in the quote, in the language the person on site will use. If the trigger requires your judgment to identify, then in practice the trigger is a phone call to you, and you should write it that way instead of pretending otherwise.

Step 7: Set the stop-and-call threshold as a number

Do not write "notify the customer of significant additional work." Commit to a value and tune it later. A workable default for a residential service job: work stops and the customer is called when unplanned work reaches 2.0 hours, or before starting any single unforeseen item you expect to exceed 4.0 hours. On a job whose whole planned scope is under 4 hours, scale the first number down: a threshold larger than half the job is not a gate. For a multi-day project, scale it to the day rather than the job: 25% of a day's planned hours.

Two parameters, both of which have to be in the quote: the threshold, and what happens when it is hit. "Work pauses and resumes on written approval" and "work continues at the stated unit rate up to a stated ceiling" are both legitimate, and they are very different promises. Pick one and say which.

Step 8: Record the outcome so this risk is sizeable next time

The last step is the one that ends the problem permanently. Every time a named risk resolves, write down whether the trigger occurred and what it actually cost in hours, in a field you can query, on the job type rather than in a narrative note. Five instances later you have a distribution and this article stops applying to that risk.

Silence is the enemy here. Record the negative result too. A job where the condition was absent is data, and it is the data that makes a rate computable at all.

Worked example: the risk that averaging destroys

A shop is quoting a job type it has run eleven times. There is one named risk: a concealed condition that, when present, adds substantial rework. The planned scope is 20.0 hours.

The distribution test fails, and it fails informatively. Of the eleven prior jobs, the file explicitly records the condition on three. The other eight are silent, and silence is not absence: nobody was asked to record it. So the shop cannot compute a rate at all. What it does have is three jobs where the condition was present, and on those it added roughly 12.0, 14.0 and 16.0 hours.

What a contingency would do here. Suppose the shop guesses the condition shows up about one job in three and prices the mean: one third of a 14.0-hour midpoint is about 4.7 hours, which is a 23% uplift on the 20.0-hour base. Now look at what that buys. If the condition is absent, the customer paid 4.7 hours for nothing and the shop's price lost to a competitor who did not. If it is present, the shop is 7.3 to 11.3 hours short and eating the difference on a fixed price.

That is the structural point, and it is worth naming: this outcome is bimodal, so the mean occurs never. The job is either about 20.0 hours or about 34.0 hours. There is no version of it that lands at 24.7. A contingency prices a spread around a central case, and where there is no central case it produces a number that is wrong on both branches. Pricing the mean of a two-branch outcome is the single most common way a well-run shop loses money on a job it estimated carefully.

What the shop does instead. Three instruments together:

• A paid investigation of 1.5 hours to open and inspect the two accessible points where the condition would be visible, quoted separately and scheduled before the main quote is issued. That is 7.5% of the 20.0-hour base, and it collapses the whole question for most jobs.
• If the customer declines the investigation, the main quote goes out at 20.0 hours with an exclusion naming the condition explicitly, plus a trigger-priced unit rate for the rework so nobody is negotiating under pressure at the moment of discovery.
• A stop-and-call gate at 2.0 hours of unplanned work, with the quote stating that work pauses and resumes on written approval.

Carry it through. The customer takes the investigation. It runs 1.5 hours and finds the condition present at one of the two points. The shop now quotes the real job: 20.0 hours plus 13.0 hours of rework, as one fixed price on a known scope. The 13.0 is not the midpoint of the three prior instances: the condition turned out to be present at one of the two points rather than both, and the estimator sized it from what he could now see. That is the difference between a number from a job and a number from a distribution. The customer paid 1.5 hours to find out, which is the cheapest line on the whole job, and the shop is not carrying a 13.0-hour surprise on a fixed price.

The other branch matters just as much. Had the investigation found the condition absent, the shop quotes 20.0 hours flat with no uplift and wins against a competitor still carrying a guessed 23%. The investigation pays for itself on that branch too, in a way that never appears in the job record because nothing went wrong.

Failure mode. The version of this that goes badly is subtle: the shop writes the exclusion, wins the job, hits the condition on day two, and the customer says the exclusion is not what they understood the exclusion to mean. That is not an estimating failure, it is a surfacing failure, and it is a different subject. See related: The Exclusion That Was in the Quote and Still Cost the Job.

Verifying the instrument before the quote goes out

• Can you state the trigger in one sentence that a tech on site could evaluate without calling you?
• Does the quote say which party carries the unknown, in a sentence a customer would recognize as being about them?
• If you used a contingency, can you name the five or more recorded instances it came from? If not, you guessed, and one of the five instruments belongs there instead.
• Does the stop-and-call gate carry an actual number and a stated consequence?
• Is there a field where the outcome of this risk will be recorded when the job closes, whether or not it occurred?
• If the trigger occurs on the busiest day of the year, does the paperwork tell the tech what to do without a conversation?

References

• 29 CFR 1926.1153 (construction) and 29 CFR 1910.1053 (general industry), respirable crystalline silica, for exploratory cutting or grinding of masonry and concrete
• 29 CFR 1926.1101, asbestos in construction, including the presumption applied to thermal system insulation and surfacing material in buildings constructed no later than 1980
• 29 CFR 1910.333(b)(2), safety-related work practices for work on or near energized electrical parts, general industry
• See related: The Contingency Line and How to Size It; The Allowance and When It Is Honest; When a Fixed Price Is the Wrong Instrument; The Conditions Clause: Protect the Quote