How to Estimate a Job Type You Have Never Done

Why this matters

Every job type you run profitably today was once a job type you had never done. Somebody bid the first one. How they bid it decided whether the shop learned a new line of work or learned to be afraid of it.

Both failure modes are expensive. Guess low and you fund the customer's project with your own hours, then write off that job type on a sample size of one. Guess high and you lose the bid and learn nothing. The method below separates what you know from what you do not, prices the unknown with structure instead of nerve, and makes the first job buy the data that prices the second one.

Before anything: this is a competence gate before it is a costing gate

If the new job type crosses into a hazard class or a scope your shop is not licensed and trained for, the estimate is irrelevant and the answer is no until the gate is cleared. Name the requirement, and never decide it on site:

• Fuel gas work: hold the license or certification your jurisdiction requires for gas piping and appliance connection, and pull the permit. Pressure-test and leak-check per the applicable fuel gas code before the appliance goes into service.
• Confined space: entry requires a written permit program, atmospheric testing before and during entry, an attendant outside, and a retrieval plan. Not something to work out on the first job.
• Work at height: fall protection suited to the surface and height, in place before anyone goes up, never improvised from what is on the truck.
• Anything energized you cannot isolate: if it cannot be done de-energized with lockout and a verified-dead test, it is qualified-person work under a different set of rules.

Those are gates, not line items. A shop that prices around them is not estimating a new job type, it is estimating a violation.

Step 1: Decide whether this is a first one worth taking

Bid it when three things are true: the downside is recoverable if you are wrong by a factor of two, you can get a credible outside bracket on the part you do not know, and it is a job type you want more of. A first instance is a training purchase, and it only pays back if there is a second one.

Decline, or price it to lose, when the job carries a hard deadline with a penalty attached, when the customer is buying on price alone (so you get all the learning risk and none of the relationship), or when the unknown part is the majority of the job and no outside bracket exists. A first instance under schedule pressure with no data is a gamble with a customer's project, not a learning job.

Step 2: Decompose into operations you have already timed

Do not estimate the job. Estimate its operations, and sort each into one of three bins:

• Known: an operation you already run elsewhere and have actuals for. Access, setup, isolation, haul-out, cleanup, testing and paperwork are nearly always known even on a brand-new job type, and together they are frequently half the hours.
• Analogous: not run in exactly this form, but it maps onto one you have with a size or count adjustment you can state out loud.
• Unknown: an operation with no counterpart anywhere in your history.

The method depends on this sort being honest, and the pressure runs one direction: it feels competent to call things analogous. An operation is analogous only if you can name the known operation and state the adjustment factor. If you cannot say "same as X but roughly 1.5 times the size," it is unknown.

Step 3: Bracket the unknowns from outside your own shop

For every operation in the unknown bin, get a range from a source that has done it: manufacturer literature with a stated labor figure, a supplier's technical rep, a peer shop in a non-competing market, a trade association labor manual, an industry production-rate table.

Take the top of the bracket into the bid, not the middle, unless Step 5 lands you on a structure with a ceiling of its own: a not-to-exceed already caps your exposure, so there the expected case can go in the bid with the bracket top as the ceiling. Without a ceiling, bid the top. Record both ends on the job file, because the bracket's width tells you which structure Step 5 should pick.

No bracket from anyone means no basis for a fixed price on that operation. That is not a reason to guess harder, it is the answer to Step 5.

Step 4: Apply a first-time factor to the unknown operations only

Add a learning multiplier of about 1.25x to the unknown operations. Not to the analogous ones, and definitely not to the whole job.

Applying it across the whole estimate inflates the half you already know how to do, which is how you lose bids you should win while staying exposed on the part that is actually risky.

The 1.25x is a default for a competent crew on an operation adjacent to their skill set. Widen it toward 1.5x when the operation needs a tool or technique nobody on the crew has used, and drop it toward 1.1x when a tech has done this operation at a previous employer even though your shop has not.

Step 5: Let the uncertainty pick the pricing structure

Compute what share of the estimated hours sits in the unknown bin after the first-time factor.

• Under about 15% unknown: bid fixed price. Exposure is inside ordinary variance.
• About 15% to 30% unknown: fixed price with a written conditions clause naming the unknown operation and what triggers a change order.
• Over about 30% unknown: do not bid fixed. Use time and materials with a not-to-exceed, or a paid two-stage approach where the customer buys an assessment, then gets a firm number.

Tune those bands once you have a few first-instances behind you. What matters is that the structure comes from the arithmetic rather than from how confident you feel, because confidence on an unfamiliar job type is not correlated with accuracy.

Step 6: Instrument the job so the first one buys the data

Before the crew goes out, tell them the job is being timed per operation, not in total, and give them the operation list from Step 2 to log against. Without that, you get one number back and it teaches you almost nothing, because you will not know which operation ate the hours.

Add a mid-job checkpoint even on a job small enough not to normally warrant one. On a first instance the checkpoint is worth more than usual, because the burn ratio is the only early warning you have.

Step 7: Debrief within 48 hours, per operation

Sit down with whoever ran it and fill in the actual against the estimate for each operation. Two days, not two weeks - the per-operation detail decays fast and it is the whole point of the exercise.

The output is a template with three fields per operation: your figure, the confidence (known, analogous, unknown), and the instance count. The first-time factor comes off only after three instances agree, not after one good day.

A worked example, carried through

A shop is asked to bid a job type it has never run. Decomposition into five operations:

Known and analogous total: 1.0 plus 0.75 plus 2.0 plus 1.5 is 5.25 hours.

Bracketing the unknown. Manufacturer installation literature and a peer shop in another market bracket the install operation at 3.0 to 5.0 hours. Bracket width is 2.0 hours.

Two figures come out of this, and both go in the quote. At the bracket midpoint of 4.0 hours with the 1.25x first-time factor, the install is 5.0 hours, so the expected total is 5.25 plus 5.0, or 10.25 hours. At the top of the bracket, 5.0 hours with the same 1.25x factor is 6.25 hours, so the ceiling is 5.25 plus 6.25, or 11.5 hours.

Pricing structure. The unknown block at the expected case is 5.0 of 10.25 hours, or about 49% of the estimate - well past the 30% band. So this does not get a fixed price. It gets quoted as expected around 10.25 hours, not to exceed 11.5 hours, with the headroom of 1.25 hours (about 12% of the 10.25-hour expected figure) named in writing.

The actuals, logged per operation:

The job landed at 10.8 hours, 0.55 hours above the 10.25-hour expected figure (about 5% over the expected figure) and 0.7 hours under the 11.5-hour ceiling. On the not-to-exceed structure the shop gets paid for 10.8 hours and nothing is in dispute. On the 9.0-hour fixed guess it would have eaten 1.8 hours, about 20% of the 9.0 hours it quoted.

The lesson is in which operation missed. The unknown install, the one everybody worried about, came in 0.4 hours under its first-time-adjusted 5.0-hour figure: the bracket plus the 1.25x factor slightly over-covered. The largest miss was the analogous removal at 2.7 hours against 2.0, 35% over its own estimate and more than the whole job's net overrun.

That is the reliable pattern on a first instance: the unknown gets bracketed, factored and watched, so it behaves. The analogous operation got a number from memory with a size adjustment nobody checked, and that is where the hours go.

The template correction. Set removal at 2.7 hours for this job type and re-check it next instance. Set install at 4.6 hours but keep the confidence flag at unknown and the first-time factor in place until three instances agree: one data point on an unfamiliar operation is an anecdote, not a rate.

What changes the answer

A tech on the crew has done this elsewhere. Move the operation from unknown to analogous, use their figure as the bracket midpoint rather than the top, and drop the first-time factor toward 1.1x. Then verify against the actual, because a figure carried from a previous employer usually reflects that employer's tooling and crew size.

The customer will not accept a not-to-exceed. Then either the job gets a fixed price with a conditions clause naming the unknown operation specifically, or you decline. A fixed price on a job that is half unknown with no clause is not an estimate, it is a wager.

It is a first instance of many, already sold. If a customer has committed to a series, price the first closer to expected than to ceiling, because the learning has somewhere to pay back. Do it explicitly, and write down which future instances justify it.

The bracket sources disagree wildly. A 2-to-1 spread between credible sources usually means the operation is not one operation: it varies with a site condition neither source stated. Find that condition before bidding rather than averaging two numbers that describe different jobs.

How to verify you got this right

After the first instance, check whether your bins were honest: did the operations you called known come in within about 10% of their figures? If a known operation missed badly, your existing templates have a problem unrelated to the new job type, and you just found it.

Check whether the unknown behaved inside its bracket. Landing inside the bracket means the sourcing worked and the method is reusable. Landing above the top of the bracket means your sources were describing a different job or a better-equipped crew, and the next first-instance needs a wider bracket or a different source.

After three instances, compare the three actuals for each operation. Where they agree within about 15%, retire the first-time factor on that operation and move it to known. Where they still disagree, the operation depends on something you are not asking at intake, which is a scoping question rather than an estimating one.

References

• See related: Estimating the Unknown: The Conditions Clause for the contract language Step 5 needs.
• See related: Time & Materials vs Fixed Bid vs Not-to-Exceed Decision Matrix for the structure choice.
• See related: How to Read Your Own Job Costing Data for the sample-size gates behind the three-instance rule.
• OSHA general industry standards on permit-required confined spaces, fall protection, and control of hazardous energy, for the competence gates named above.
• Manufacturer installation literature and published trade production-rate tables as bracketing sources.