Forecasting a New Service Line From a Reference Class

Why this matters

When an owner adds a service line, the forecast is almost always built from the inside: the jobs we will get, the tech we will hire, the ads we will run, and a breakeven month that falls comfortably inside the first season. Then the line takes twice as long to fill, the dedicated tech sits idle three days a week, and the owner either pulls the plug too early or keeps feeding it too long. The fix is not a better spreadsheet. It is starting from what actually happened to launches like yours and adjusting from there. Bent Flyvbjerg built this into a formal method for large projects, reference class forecasting; this card is the shop-sized version, run on one example from start to finish.

The idea of base rates and why the outside view beats the inside view is covered in its own card. This one is the procedure.

The running example

An eight-truck plumbing shop wants to add a dedicated drain cleaning and hydro-jetting line: one tech, one jetter on a payment, a camera, and a modest ad budget. The owner has worked out that the line covers its own fixed monthly cost (the tech's wages and benefits, the equipment payment and the ads) at about 9 billed drain jobs a week. Below that the line loses money; above it, the margin starts contributing to the shop.

Each step below is applied to this example as it goes.

Step 1: Write your own forecast first, and seal it

Before you collect any outside data, write down what you believe: jobs per week at month 3 and month 6, and the month you expect to reach breakeven volume. Date it and put it somewhere you will not edit it.

Why first: once you see the reference class, your memory of what you originally thought shifts toward it, and you lose the measure of how optimistic you were. That gap is the most useful calibration data you will ever get about your own forecasts.

What breaks if you skip it: nothing visible now. A year from now you will believe you always expected it to take a while.

The example: the owner writes 12 jobs a week by month 3, 14 by month 6, breakeven in month 4.

Step 2: Define the outcome in a form other launches can be measured on

Your forecast has to be stated in a unit you can also get from other people's launches. "Successful" is not a unit. Good choices:

• jobs per week at month 6 (or the trade's natural unit: accounts, routes, installs)
• months until weekly volume first reached the line's breakeven level
• whether the line was still running at two years

Why: if your cases report different measures (one peer remembers revenue, another remembers "it got busy"), you cannot line them up and the median means nothing.

The example: the owner uses all three. For peers, he translates breakeven into "how many months until the drain line was paying for its own tech and equipment," which every owner who ran one can answer.

Step 3: Pick the reference class

Find 6 to 12 launches that resemble yours in the things that drive the result: the same kind of service, a similar shop size, a dedicated person or truck rather than a side offering, and a similar market. Sources in order of trust:

1. Your own past launches. Even if they were a different service, they show how your shop ramps.
2. Peer owners who added the same line, through your trade association, a buying group or an owner group. Ask directly.
3. Your supplier's rep, who has watched several shops buy the same equipment. Ask about the median shop, not the showcase.

The critical instruction: ask specifically for launches that failed or were dropped. Peers volunteer their wins. If your class contains no failures, it is a list of survivors, not a reference class. See related: Survivorship Bias in Industry Advice.

What breaks if you get this wrong: a class drawn only from shops that are still running drain lines will tell you every launch reaches breakeven.

The example: the owner gathers 8 cases. Two are his own (a water heater replacement push three years ago, a backflow testing line two years ago); six are peers who added drain cleaning, including two who dropped it.

Step 4: Collect every case's outcome on the same measures

Build one small table. Do not average yet.

For his own two launches he translates their breakeven into the same terms: the month the line's weekly volume first covered the person and equipment dedicated to it.

Step 5: Set your starting estimate at the reference class

Read the class honestly:

• Jobs a week at month 6: sorted, the eight values are 2, 3, 4, 5, 7, 7, 8, 11. The median sits between 5 and 7, so 6 jobs a week.
• Months to breakeven: 6 of 8 cases got there. Among those six, sorted 5, 7, 8, 9, 11, 14, the median is between 8 and 9, so about 8 to 9 months, with a spread from 5 to 14.
• Failure rate: 2 of 8, one in four, never reached breakeven and were dropped.

That is the starting forecast before anything about this shop is considered: 6 jobs a week at month 6, breakeven around month 8 or 9 if it comes at all, and a one-in-four chance it does not. Compare it with the sealed inside view of 14 at month 6 and breakeven in month 4. The reference median is well under half his month-6 number, and breakeven arrives roughly twice as late.

Step 6: Adjust modestly, and only for differences you can point to

Now ask what is genuinely different about this shop, and accept only differences backed by evidence you could show someone.

The rule, stated so it can be checked: move no more than halfway from the reference median toward your own sealed forecast, and only for evidence that has a record behind it. Halfway from 6 toward 14 jobs a week caps the month-6 adjustment at 10.

The example: the owner lists his candidate reasons.

• "We are better at marketing than most shops." No evidence. Rejected.
• "Our techs are great." True of the plumbing techs, irrelevant to a new jetter tech. Rejected.
• "We already turn away drain calls." He checks the call log: the office has been referring out about 4 drain and sewer calls a week for the past six months. That is real, countable, existing demand. He also notices that the two strongest peer cases, A and B, both had existing drain demand at launch; the two that failed, G and H, started cold.

On the strength of the call log, he moves the month-6 estimate from 6 to about 7 to 8 jobs a week and the breakeven estimate from 8 to 9 months to about 7 months. Both are inside the halfway cap. He leaves the failure rate near one in four, because the call log is evidence about demand and says nothing about the other ways a line fails: the wrong hire in the seat, a jetter that sits broken for three weeks, a launch that runs into a slow season.

What breaks if you over-adjust: the inside view creeps back in under the label of "adjustments," and you end up at 12 a week with a story for each step.

Step 7: Plan for the spread, not the middle

The median is where half the cases land worse. Plan the cash and the staffing for the slow end of the class.

The example: the slowest case that survived took 14 months. The owner sets aside enough working capital to carry the line's shortfall through 14 months rather than 7. He also sets a written checkpoint at month 6: if the line is under 4 jobs a week, the bottom quarter of the class, he revisits whether to keep the dedicated tech or fold drain work back into the general crew. How to write that checkpoint so it actually binds is covered elsewhere. See related: Setting Kill Criteria Before You Start Something New.

What breaks if you skip it: the owner plans around month 7, hits month 10 at 6 jobs a week (still inside the normal range of his own reference class), panics because he has run out of cushion, and kills a line that was on track for a typical outcome.

Step 8: Record the forecast and check it on a schedule

Write down the final forecast next to the sealed inside view, with the reference table attached. Check it at months 3, 6 and 12.

The example, six months on: the line is at 7 jobs a week, inside the adjusted range of 7 to 8, and the owner's sealed forecast of 14 is off by half. He now knows two things he did not before: the reference method was close, and his own unadjusted forecasts on new lines run roughly twice too optimistic. Next time he considers a launch, that ratio is evidence too.

When the method needs changing

When there is no reference class at all. Truly new services are rarer than they feel; "nobody around here offers it" usually means "nobody nearby," and shops elsewhere have. Widen the geography before concluding there are no cases. If the class is genuinely empty, use the closest analogue (your own last launch of anything) and say out loud that the forecast is weak.

When you hold strong evidence. A signed service agreement covering a known share of the volume is not a story; it is a floor. Forecast the contracted volume as given and use the reference class only for the uncontracted part.

When cases differ in timing. A launch that started in a busy season ramps faster than one started in a slow one. If your launch month differs from most of the class, note which cases started in the same season and lean on them.

References

• Flyvbjerg, B. (2006), "From Nobel Prize to Project Management: Getting Risks Right," Project Management Journal
• Kahneman, D., and Lovallo, D. (1993), "Timid Choices and Bold Forecasts," Management Science
• See related: Base Rates: The Number You Should Start From; The Planning Fallacy in Shop Projects; Survivorship Bias in Industry Advice; Setting Kill Criteria Before You Start Something New