operations

Measuring a Rep Without Demoralizing Them

Research on rank feedback says a bare leaderboard can lower output. Here is how to measure canvassers so the numbers help.

Canvassing is one of the most measurable jobs in sales. Every door is an event. Every event has a timestamp, a location, and an outcome. A manager can know more about a canvasser's day than about almost any other employee's.

That is exactly the problem. Measurement this granular makes it very easy to build a management system that produces accurate numbers and worse performance.

There is real research on this, and it is less flattering to the standard sales floor than most managers expect.

Rank feedback, by itself, can lower output

Iwan Barankay at Wharton ran a three-year randomized experiment with 1,754 full-time furniture salespeople, published as the working paper Rank Incentives: Evidence from a Randomized Workplace Experiment. The design varied whether reps were told their performance rank and whether they were given benchmarks showing what performance was needed to reach the top 10, 25, or 50 percent. Pay was commission on absolute performance, so rank carried no money.

The headline result: "removing rank feedback actually increases sales performance by 11%, or 1/10th of a standard deviation."

The mechanism Barankay describes is the one every field manager has watched happen: "telling people that their rank is worse than expected leads to a demoralization effect, a drop in performance." And the fix he identifies is specific. Adding benchmarks to the rank significantly raised performance, because knowing what is needed to climb "dampens this demoralization effect."

So the finding is not that measurement is bad. It is that a bare ordinal rank, with no information about the gap, is a discouragement device.

Anik Ashraf's field experiment with 366 piece-rate garment workers, published as CESifo Working Paper 9591 and later in Management Science, found something adjacent and worth knowing. Private ranks did not affect productivity. Public ranks did, and not always upward: "publicly-ranked workers reduce productivity to conform to their social groups in the workplace," producing a 2.7 percent decrease among those competing with friends. That is a factory, not a sales floor, so do not oversell the transfer. But a canvassing team that rides together, eats together, and sees each other's numbers is socially closer than most workplaces, and "conforming down to your friends" is a recognizable failure mode.

Measure inputs the rep controls, separately from outcomes they do not

The most common measurement error in canvassing is comparing two reps' set rates without accounting for where they knocked and when.

A rep's contact rate is mostly a property of the neighborhood and the hour. The Bureau of Labor Statistics' American Time Use Survey for 2025 found 81 percent of employed people worked on an average weekday against 30 percent on an average weekend day. A rep assigned weekday daytime hours in a commuter subdivision will show a worse contact rate than a rep working Saturday mornings in a retiree neighborhood, and neither number says anything about either rep.

Split the metrics by who controls them.

The rep controls: doors attempted, hours in the field, logging discipline, how long they stay at a door, whether callbacks are honored on time, and the quality of the first fifteen seconds.

The rep does not control: contact rate, the housing stock, the weather, whether a storm hit, and whether the offer is any good.

Jointly determined: contact-to-pitch and pitch-to-set, which is why those are the ratios worth coaching on.

Rank reps on the first list. Diagnose the territory with the second. Mixing them produces a leaderboard that is partly a map of who got the good turf, which every rep in the truck can see even when the manager cannot.

Specific and difficult beats vague, and "do your best" beats nothing by very little

Locke and Latham's 35-year review in American Psychologist is the standard citation here and it is unusually blunt. "We found that specific, difficult goals consistently led to higher performance than urging people to do their best," with meta-analytic effect sizes ranging from .42 to .80. Their explanation: "do-your-best goals have no external referent and thus are defined idiosyncratically. This allows for a wide range of acceptable performance levels."

They also name the limit. Performance "leveled off or decreased only when the limits of ability were reached or when commitment to a highly difficult goal lapsed."

Two practical consequences for canvassing.

First, "knock hard today" is close to worthless as a goal. "Sixty doors logged before 7pm in your assigned territory" is a goal. The specificity is what makes it work, and the difficulty is the active ingredient, not the specificity by itself. Locke and Latham are explicit that "goal specificity in itself does not necessarily lead to high performance because specific goals vary in difficulty."

Second, there is a ceiling. A goal set past the point where a rep believes it is reachable stops producing effort and starts producing the commitment lapse Locke and Latham describe. This is the same cliff Barankay found with bare ranks, arriving by a different road.

Quotas have a specific failure mode worth knowing about

Harikesh Nair and Sanjog Misra studied a Fortune 500 sales force that moved to a plan with no quota requirement. Stanford GSB's summary, Eliminating Sales Quotas May Stimulate Profits, reports that overall revenues rose 9 percent, roughly a million dollars a month.

The mechanism matters more than the number. Quotas create what the researchers call a "perverse incentive to postpone their effort to the next cycle." A rep who has already hit quota stalls the next sale into next month. A rep who cannot reach it stops trying and banks the effort. Both behaviors are rational and both are invisible in the quota number itself.

Two caveats before anyone rips up their comp plan: the 9 percent came from an entire redesigned plan, not from quota removal in isolation, and the Stanford page is the school's summary rather than the paper. But the gaming pattern is worth watching for in your own data. If your sets cluster suspiciously at the end of a pay period and go quiet right after, you are watching it happen.

What to actually put on the wall

Pulling the research together, a measurement system that helps rather than discourages looks like this.

Show the gap, not just the rank. Barankay's benchmark condition is the single most actionable finding here. "You are 7th" is a discouragement. "You are 7th, and 3 more sets this week puts you 4th" is a goal. If your leaderboard cannot show the gap, it is doing the thing the research says lowers output.

Normalize for territory and hour band. Publish doors per hour and set rate within comparable conditions. If a rep worked a daytime pass, compare them to daytime passes.

Make the daily number an input number. Doors logged, hours worked, callbacks honored. These are achievable every single day regardless of luck, which means a rep can have a bad outcome day and still have a good work day. That distinction is what keeps a new rep in the truck during week three.

Keep one private view and one public view. Ashraf's result suggests the public version carries social risk the private one does not. A private detail view with full diagnostics, plus a public view limited to input metrics and progress toward a stated benchmark, gets most of the motivation with less of the conformity drag.

Never present a ratio built on fewer than a few hundred doors. A rep with 70 logged knocks has a set rate with an error bar wide enough to make any ranking meaningless. Publishing it anyway teaches reps that the numbers are noise, which is the fastest way to lose logging discipline.

The one-sentence version

Measure the work, benchmark the gap, normalize for the turf, and keep the ranking honest about sample size. The evidence says a bare rank with no path to climb is not neutral; it costs you output.

Canvass Pro logs doors with timestamps and territory assignment, so input metrics and outcome metrics can be read separately rather than blended into one number.