Narrow Framing: Deciding One Job at a Time
Why this matters
An owner who argues every disputed callback on its merits will usually win most of the arguments and still lose. Each fight looks worth having on its own: the customer is wrong, the hour of free work is real. Added up over a year, the fights cost more office time, more free work and more customers than a simple goodwill rule would have. The same pattern shows up in bidding, discounting and taking risks on new work. Judging each case as if it were the only one is called narrow framing, and the fix is to ask a different question: not "what should I do about this one?" but "what rule would I want if this came up forty times a year?"
The idea
Daniel Kahneman and Dan Lovallo described narrow framing in a 1993 paper on why managers make "timid choices": decision makers tend to treat each risky choice as a one-off and evaluate it in isolation, rather than seeing it as one of many similar choices they will face. Because losses feel heavier than equal gains, a single risk viewed alone often looks unattractive even when a steady series of the same risk is clearly worth taking.
The classic illustration comes from the economist Paul Samuelson, who offered a colleague a coin-flip bet: win twice as much as you lose. The colleague turned down one flip, because the possible loss of one flip stung more than the possible win pleased, but said he would take a hundred of them. A hundred flips at those odds almost never lose money overall. One flip loses half the time. Same bet, different frame.
A shop makes the same choices over and over: bids, warranty calls, small concessions, which jobs to take. Each one feels like a single coin flip, and it is actually one of the hundred.
Where it shows up in a shop
Warranty and callback disputes. Each ambiguous claim gets argued as a matter of principle, when the shop faces dozens a year and the useful question is what a standing rule would cost and keep.
Fixed-price bids. An owner refuses a fixed-price job because it could run 25% over. Across thirty bids a year, some run over and some under, and the question is whether the bid method is right on average, not whether this one could go wrong.
Small concessions. A tech who gives away a filter here and a trip charge there is deciding one ticket at a time. The office sees the total only at year end.
Customer firing. Each difficult customer looks worth keeping in isolation ("they do pay"), when the pattern across the bottom twenty accounts may be a third of the office's time for a twentieth of revenue.
In every case, the cure is the same move: name the class of decision, count how often it happens, and decide the class once.
Worked example: the callback disputes
A pest and lawn shop with ten people logs every disputed callback for a year: a customer wants a free return visit and the claim is ambiguous (was it the treatment, the weather, a neighbor's yard, or a missed instruction?). There were 40 of them. The owner argued each one.
What the year looked like, from the shop's own records:
- Each dispute took about 0.8 hours of the owner's or office manager's time in calls and emails. 40 times 0.8 is 32 office hours.
- The shop won 24 of the 40 (60%) and the customer accepted a paid visit or nothing. It lost 16, and did the return visit free anyway, at about 1.5 tech hours each: 16 times 1.5 is 24 tech hours.
- Of the 40 customers, 9 did not renew the following season, against about 2 in a comparable group of 40 customers with no dispute.
Total shop time for the year's disputes: 32 plus 24, 56 hours, and 9 lost customers.
Now the policy the owner wrote afterwards: any first ambiguous callback from a customer, within 30 days of the treatment, gets a free return visit with no argument. A second within 12 months goes to the owner. Run on the same 40 claims:
- All 40 get a free visit: 40 times 1.5 is 60 tech hours.
- Office time drops to scheduling, about 0.1 hours each: 40 times 0.1 is 4 office hours.
- Assume non-renewal falls to the no-dispute rate of about 2.
Total: 60 plus 4, 64 hours, and about 2 lost customers. The policy costs 8 more hours of shop time a year (64 minus 56) and keeps about 7 more customers (9 minus 2).
Those two are different currencies, so convert. This shop's typical residential customer is worth about 6 billed hours a year. 7 customers times 6 is 42 billed hours a year, and at the shop's gross margin of about 45%, 42 times 0.45 keeps roughly 19 billed hours' worth of gross profit a year, recurring for as long as those customers stay. Against it, the 8 extra hours are paid labor hours, which cost the shop less than a billed hour brings in. So the policy wins by more than 19 against 8 suggests, before counting the owner getting most of 28 office hours back (32 minus 4), and before counting reviews.
Two cautions about that number. The 7 retained customers is an estimate drawn from one year and two groups of 40, so it could be off by several in either direction (see related: Regression to the Mean and the Tech Who Had a Great Month for why one year of a small count can mislead). And some disputes really were customers gaming the shop; the second-claim clause routes those to the owner rather than pretending they do not exist.
The point of the example is not the precise answer. It is that the owner who argued all 40 never saw the 56 hours or the 9 non-renewals, because no single dispute contained them. They existed only in the total.
A second case: one rule, two bids, opposite answers
The same shop's sister business, a six-person remodeling crew, ran a rule test the following spring. The owner had a habit of declining fixed-price bathroom jobs whenever the house was old enough to hide surprises behind the tile, on the grounds that "this one could go negative."
The crew's records for the previous year: 30 fixed-price bathroom jobs. Average gross margin across the 30 was about 36%. Three of the 30 lost money, the worst by about 8% of that job's price. The other 27 ranged from about 20% to about 55%. Old-house jobs were in the mix and averaged slightly lower, around 31%, but none was the worst loss.
Viewed alone, any old-house bathroom "could go negative," and three in thirty did. Viewed as a class, the owner was turning down work that averaged about 31% margin to avoid a roughly one-in-ten chance of a single-digit loss on that job. The policy he wrote: take any residential bath at the standard price with a 10% contingency line for concealed conditions; decline only where the worst plausible loss on that one job would exceed a month of the crew's gross profit.
Then the test case arrived: a commercial tenant fit-out, four bathrooms, roughly a fifth of a normal year's revenue in one contract, fixed price, with a builder the crew had never worked with. Run through the same rule, the answer flips. A 25% overrun on a job a fifth of the year's size wipes out about 5% of a year's revenue (0.25 times 0.20), against a month of gross profit that, at a 36% margin, is about 3% of a year's revenue (0.36 divided by 12). That is not one flip in a hundred. It is one flip with the shop on the table, and the rule sent it to a separate decision, where the owner priced it with a larger contingency and a schedule of progress payments.
The rule did not tell the owner to be bold or cautious. It told him which frame each bid belonged in.
Turning a frame into a policy
A good policy has four parts, and each one is there to stop the case-by-case argument creeping back:
- The class. Precisely which situations it covers ("first ambiguous callback within 30 days"), so nobody has to decide whether a case qualifies.
- The default action. What happens without discussion.
- The exit. The condition that sends a case to a person, so the rule does not get abused or applied to cases it was never meant for.
- The review date. When you will look at the class totals again, because a policy built on one year's numbers is a hypothesis.
Writing these as rules the office can apply without you is its own skill; see related: Writing Decision Rules So the Office Stops Asking You.
Where the portfolio view does NOT apply
This is the half of the idea people skip, and it is where narrow framing is actually the right instinct.
A risk that can sink you. Samuelson's hundred flips only work if you survive the bad runs. A fixed-price commercial bid that could lose a quarter of the shop's annual gross profit is not one coin flip among many; it is one bet that can end the game. Portfolio thinking assumes many small, independent bets. A single large one should be judged alone, and judged harshly.
Bets that are not independent. Ten fixed-price bids for the same builder, on the same house plan, in the same market, are closer to one bet than ten. If the plan is underpriced, all ten lose together. A portfolio of correlated risks gives none of the averaging you were counting on.
Too few repetitions. A shop that does two commercial bids a year does not have a portfolio. The averaging needs volume to show up within any useful time.
Customers who talk to each other. In a small community or a tight neighborhood, a dispute with one customer is not independent of the next one's decision to call you. That usually strengthens the case for a generous policy rather than weakening it, but it means the per-case numbers understate the effect.
Choices that set precedent with staff. A concession a tech makes on one ticket can become the expected behavior on every ticket. That is a portfolio effect too, but one that runs against generosity: the class is bigger than it looks.
How to tell you are framing narrowly
You are arguing about principle on a case you will see again next week. You know how this dispute is going but not how many disputes there were last year. You remember the customers who took advantage of you better than the customers you lost. And you have never written down what the shop does in this situation, so every instance gets decided fresh by whoever is on the phone. When you notice one of those, pull the class totals before you decide the case in front of you. The expected-value arithmetic for a single bet is covered in its own card; see related: The Expected-Value Habit.
References
- Kahneman, D. and Lovallo, D., "Timid Choices and Bold Forecasts: A Cognitive Perspective on Risk Taking," Management Science (1993)
- Samuelson, P. A., "Risk and Uncertainty: A Fallacy of Large Numbers," Scientia (1963)
- Benartzi, S. and Thaler, R. H., "Myopic Loss Aversion and the Equity Premium Puzzle," Quarterly Journal of Economics (1995)
- See related: The Expected-Value Habit
- See related: Writing Decision Rules So the Office Stops Asking You