The exemplary performer

You cannot distribute a personality. You can distribute what made one possible.

The experiment that is already running

In almost every unit, some people get the result and others do not: under the same rules, with the same systems, the same customers and the same manager. That is as close to a controlled comparison as an organization ever produces, it has already been run, and nobody has read the result.

Gilbert built his method on it, and gave the comparison a number. Then he did the thing that matters more: he subtracted. Take what the exemplary performer does and has, take away what the typical performer does and has, and read what is left.

What is left is almost never a personality. It is a list of conditions the organization gave to one person and withheld from the others, usually without deciding to. That list is the finding, and unlike a personality it can be handed out. The number comes first, though, because it decides whether the subtraction is worth doing at all.

One caution before anything else. This is a comparison, not an experiment. Nobody randomized the territories, the customers, the tenure or the manager. Establishing that the two groups are genuinely comparable is the first task of the analysis, not a formality, and section 08 says what happens when they are not.

Gilbert’s number, and what it is for

The performance improvement potential is the oldest instrument in this discipline and one of the most misused. It is a ratio, it takes two minutes to compute, and its purpose is the opposite of what its name suggests to most readers.

Gilbert defined it as the ratio between exemplary and typical performance on the same measure (Gilbert, 1978). Not between the best and the worst: between the best and the ordinary, because the ordinary is where the volume is. A PIP of 2 says the exemplary performer produces twice what the typical one produces. A PIP of 1 says there is nothing to find here and you should go and look somewhere else.

Two mechanics matter and both are routinely got wrong. Direction. Where a low number is good (days, errors, cost, complaints) the ratio is typical over exemplary, not the other way round. Reported the wrong way a PIP of 2.12 arrives as 0.47 and reads as a problem rather than an opportunity. The comparison group. Gilbert’s denominator is the typical performer, not the mean of everybody including the exemplar, and not the worst. Using the worst inflates the number and makes the whole finding easy to dismiss.

What the ratio cannot tell you

A number this easy to compute invites more weight than it can carry. Four cautions, and then the way to use it that survives all four.

Using the ratio so that it survives the room

None of those four cautions is a reason to skip the calculation. All of them are reasons to report it with its conditions attached, and to be careful about the sentence it arrives in.

So use it as a sorting instrument rather than a conclusion. Compute it across the jobs that touch the result, take the largest, and spend the analysis there. Report it as a range rather than a point (“between 1.8 and 2.1, depending on whether the two newest advisers are included”) because a range invites the argument about comparability that the number needs to survive anyway.

And state it in the form that carries the argument. Not “our top person is twice as good”, which is a sentence about a person and provokes a defense of everybody else. But: “the distance between what this unit produces and what one of its own members produces is a factor of two, under the same conditions.” That sentence is about the organization, which is where the finding actually lives.

Expert or exemplary performer

These are routinely treated as the same person and they are not. The difference decides who you interview, and interviewing the wrong one is the commonest way this question is wasted.

Ask a manager for “the expert” and you will be given the most visible person: the one who runs the induction, answers the questions, sits on the working group. Ask for “whoever produces the outcome best, on the measure” and you will sometimes be given the same name and often not. The second question is the one this analysis needs.

And a word about deliberate practice

The popular move at this point is to turn exemplary performers into experts through deliberate practice. How much practice explains is an open debate in the research, and it is not close to being settled, which does not leave a practitioner without guidance.

The debate in one sentence: Hambrick, Macnamara and Oswald put the variance it explains at about 14 percent overall and roughly 1 percent in professions (Hambrick et al., 2020); Debatin and colleagues reply that the meta-analyses understate it by pooling studies in which the practice was not individualized (Debatin et al., 2021). Both sides are arguing partly about measurement, so what would move the argument is not another pooled estimate but studies that define the practice tightly and observe it in professional work rather than in chess and music. That research is worth doing and has not been done at any scale.

Finding one without a popularity contest

The failure mode here is nomination. Ask who the good ones are and you will be handed the well-liked, the long-serving and the loud. The method that avoids it is not complicated, and its first step is the one that gets skipped.

Establish the metric before you ask for names. Which number distinguishes key from standard performance in this job, and what value does each of them sit at? Only then do you ask a manager for names, and you ask them against the metric rather than against an impression. Four criteria have held up across our own analyses: the person produces the outcomes in question markedly better than a standard performer, has met the metric consistently for more than two quarters, is recognized by their own team as well as by the organization, and can say what the organization is trying to achieve. The third criterion filters out the performer whose numbers come at someone else’s expense; the fourth filters out the one who cannot explain a single choice they make.

The exemplary performer cannot tell you what they do

This is the point on which the whole question turns, and it is the point most approaches get wrong. You cannot copy what somebody achieves by asking them how they achieve it. Not because they are unwilling, because they do not have access to the answer.

The reason is how skill is stored. Work that is repeated thousands of times stops being a sequence of decisions and becomes a single automated routine. The steps are still executed; they are no longer represented in a form the person can retrieve and put into words. What returns when you ask is not the procedure but a reconstruction: a plausible account assembled at the moment of asking, out of what the person believes must have been going on.

The size of the omission is now documented. A systematic review of cognitive task analysis in clinical and health services research states it plainly: experts may leave out up to 70 per cent of the components and decision steps needed to perform a task successfully, with the consequence that protocols and training built on their accounts are incomplete or ineffective (Swaby et al., 2022). Seventy percent is the upper end of a range rather than a constant, and the review says so, but even the lower end of that range destroys the assumption that an interview will do.

This is not a new discovery so much as a newly measured one. Nisbett and Wilson established the general case in 1977: when people report on their own mental processes, those reports are frequently not based on introspective access at all but on a theory of what would plausibly have influenced them (Nisbett & Wilson, 1977). What has changed since is that the effect has been measured in real work rather than argued from the laboratory, and that a method exists for getting round it.

How much goes missing

The field that has measured this most carefully is surgery, because there the cost of an incomplete description is immediate and countable.

Clark and colleagues put expert surgeons in front of a procedure they had performed hundreds of times and asked them to describe it for a trainee. Then the same procedure was reconstructed using cognitive task analysis (structured probing, observation and comparison across several experts) and the two accounts were set side by side. The reconstructed version contained steps and decision points that the surgeons’ own descriptions had left out entirely (Clark et al., 2012). Not details of style: actions they perform every time and had stopped noticing.

The method used to recover them has since been reviewed for organizational research specifically, which matters here because most of the older evidence comes from medicine and the military. Brown, Power and Gore set out what cognitive task analysis is for and where it applies in organizations: it is a family of techniques for eliciting the cognition that sits behind skilled work in its own context, precisely because that cognition is not available to interview (Brown et al., 2024).

And the consequence is measurable rather than theoretical. Training built from a reconstructed description rather than a self-report produces better performance, in general instructional settings (Tofel-Grehl & Feldon, 2013) and in surgical training specifically (Edwards et al., 2021). Something real is being recovered, and it is exactly the part the expert could not say.

What to do instead of asking

The remedy is not a better interview. It is to stop relying on recall and start relying on the record, and there are four ways to get one.

The fourth deserves its own sentence, because it inverts the usual instinct. The exemplary performer has automated the obstacle away and can no longer see it. The standard performer is standing in front of it every day and can describe it precisely. Ask one what they do and the other what stops them, and the two answers assemble the finding between them.

Reading the difference

Now do the arithmetic. Put what the exemplary performer does and has next to what the standard performer does and has, and subtract. What remains is the whole value of this question, and the subtraction is more literal than it sounds.

Work down four columns rather than writing a description. What they produce: the outcome and its number, for each. What they do: the observed sequence, from the record you built in section 04, step by step. What they have: every tool, screen, permission, list, contact and piece of information either of them touched. What they know: the rules and cues you saw them apply, which is the shortest column and the last one to fill, because most of what looks like knowledge turns out on inspection to be one of the other three.

Doing the subtraction, line by line

The figure opposite is the finished result. Getting there is a mechanical procedure, and it is worth setting out because the mistakes people make in it are all the same three.

Step three is the one that decides whether anything happens afterwards. The same observation can be written two ways, and only one of them can be acted on. “She anticipates what the credit desk will ask” is an observation about a person and it terminates there. “The two fields the credit desk most often queries are not flagged anywhere in the application screen” is the same observation written as a property of the work, and it has a cost, an owner and a completion date attached to it before anyone has finished reading the sentence.

Four qualities that look like character and are conditions of the work

The same four differences turn up so often that they are worth naming. Each of them is normally written down as a personal quality, and each of them is in fact a statement about the environment.

The reframing in the third column is not a rhetorical preference. It is what turns the finding into an assignment. “She takes ownership” has no owner, no cost and no completion date; “the query screen sends one question at a time” has all three, and it can be assigned on Friday. An explanation in terms of disposition closes the inquiry, because there is nothing further to establish and nothing to do except admire one person and hope for the others.

Some of the difference does belong to the person: nine years of pattern recognition, an aptitude, a relationship built over a decade. That part is real and it is not distributable. But it is the residue left after the subtraction, not the explanation offered before it, and in practice it is smaller than anybody expects.

From one person to the average of a unit

Here is where most attempts at this fail, and they fail in the same way: the organization tries to copy the behavior. Behavior is the one thing that cannot be handed over, because it only happens when the conditions permit it.

Three standard moves and why each disappoints. Making the exemplary performer a trainer converts a producer into a teacher, leaves the conditions untouched, and, as section 04 established, asks them to teach the part of the job they can least describe. Writing up a best-practice document produces a description of behavior, which the reader cannot enact for the same reason. Pairing people up transfers whatever the exemplar can articulate, one person at a time, at the cost of the exemplar’s own output. All three leave the advantage where it was.

Gilbert’s own purpose is the corrective. He did not compute the ratio to celebrate the exemplar; he computed it to size what the organization was losing across everybody else, and the whole of his method points at the middle of the distribution. The question is therefore never “how do we make more people like her?” It is: what does she have that the other thirty-three do not, and which of those conditions is the organization free to hand out?

Four ways to hand it over, in order of how well they hold

Not every condition can be distributed the same way, and the four routes differ sharply in how much they depend on anybody remembering to do something. Prefer the ones that do not.

The ordering is not a preference; it is what the evidence about behavior at work keeps producing. Number 03 sets it out in full: what most reliably changes what people do is a change in what they can reach at the moment of the decision, and instruction is the weakest of the available levers. This question is where that argument stops being general and acquires a name, a screen and a phone number.

Two more conditions have to change

Distributing the conditions is necessary and it is not sufficient. Two further moves turn this from an anecdote into a result, and both of them are about what you measure afterwards.

Reset the desired level to what the exemplary performer actually achieves. Not to perfection: to the number somebody in the building is already producing. This is the most useful thing this question hands to the rest of the analysis, and it is worth being explicit about why.

A desired level set by ambition invites one argument and always the same one: that it cannot be done here, with these customers, on this system. A desired level set by observation cannot be argued that way, because it has already been done here, with these customers, on this system, by a colleague whose name is known. The target stops being a demand and becomes a fact about the organization’s own capability. Number 06 takes how desired performance is established in general; this is the version of it that survives contact with a skeptical room.

Two refinements keep it honest. Set it at what the exemplar reliably achieves rather than at their best month, because a best month is not a level. And where the exemplar’s advantage turns out to be partly undistributable, set the target at the distributable part only: in the case below, the six days that the consolidated request explains rather than the full nine.

Measure the average of the unit afterwards, not the exemplar. This sounds obvious and is almost never done, because the natural thing to report is the success story rather than the mean of thirty-four people.

The reason it matters is that the two numbers can move independently, and which of them moves tells you what actually happened. If the mean moves and the exemplar does not, a condition was distributed and everybody else caught up, which is the result you were after. If the exemplar improves and the mean does not, you have given attention to one person. If both move, something changed for everybody including her, which is usually a systems change and worth understanding separately.

Behind this sits a quieter point about how performance is distributed at all. O’Boyle and Aguinis examined a large body of performance data and found the familiar bell curve to be a poorer description than usually assumed, with a heavier tail of very high producers than a normal distribution allows (O’Boyle & Aguinis, 2012). For this question the implication is practical rather than statistical: if a small number of people account for a disproportionate share of output, then the mean is dragged by the middle and the middle is exactly where a distributed condition does its work. Reporting the exemplar tells you about the tail. Reporting the mean tells you whether the organization changed.

The lending officer who sent one email

A regional bank, 34 small-business lending officers, 4,200 applications a year. The figures are illustrative and reconcile from the rates stated with them.

The business need was decision time. The bank promised a credit decision within ten working days and averaged 17.2, and applications that ran past twelve days were dropping out at 22 percent against 9 percent for the rest. The proposal on the table was a two-day course on credit-file quality for all 34 officers.

Before the course was approved, somebody asked the third question. On the bank’s own dashboard one officer was running at 8.1 days, had been for six quarters, and nobody had mentioned it: the dashboard reported the branch average, and her branch had two officers.

What was distributed, and what it returned

Nothing was taught. Three conditions were changed, and none of them was a description of her behavior.

The screen acquired a second button, hold and consolidate, which parks a query until the officer has finished reading and then sends the accumulated list as one message. The credit desk published a checklist of the two fields that account for most returns. And the desk’s direct number, which she had and the others did not, went on the intranet page next to the shared mailbox.

Three changes, all of them shared by construction. Within two quarters the unit average moved from 17.2 days to 10.4. She herself did not improve at all, which is exactly what should happen: she was already doing it.

What this does not claim

This question is the one most often done badly, and the failures are predictable. Four of them.

• The comparison is not an experiment. Nobody randomized anything. The exemplary performer may have an easier territory, a longer tenure, a better manager or a quieter branch. Establish that the groups are comparable before you attribute anything to the difference, or the whole finding is confounded and will be dismantled by the first skeptic in the room.
• Selecting on a metric selects on what the metric captures. If the measure is volume you will find whoever is best at volume, including at the expense of everything the measure does not see. Check the exemplar against a second measure before treating them as one.
• Formalizing a practice can destroy it. Some exemplary performance depends on discretion: judging when the rule should not apply. Writing that into a procedure removes the judgement and often the benefit with it. Distribute the tool and the information freely; be careful with the rule.
• Part of the difference really is the person, and when it is, the finding is about selection and staffing rather than about learning. Saying so plainly is more useful than pretending everything is environmental, and it is the only honest way to keep the claim that most of it is not.

The comparison, in six lines

Twenty minutes, on the problem in front of you. Answer these six and you have question three. Leave one blank and you have found the first thing to establish.

Two problems, one method, sixteen papers

Two models carry the method. Gilbert’s six conditions ask what kind of cause; Rummler and Brache’s three levels ask where it sits. Both are sound and this series uses both, and it leads with Gilbert, because his unit is the accomplishment, which is what an organization already counts, and because his six conditions are the categories an organization can actually change.

Where the method comes from

Two further pieces close the set and carry no number, because each points at all the numbered papers. Before the guide sits Arets L&D Business Models, the destination piece for the people who receive the requests: four models a learning function can run, and the route from HR-driven to business-driven. Last sits Solutions that cannot work, the field guide, ordered by the sentence that lands on your desk: seventeen forms a performance problem arrives in, each with an address in one of the numbered papers. Three orders in all: the destination you choose, the question you are on, and the form that landed on your desk.

Sixteen papers, six parts

Number 01 is the map and number 02 is the foundation. From number 05 on, every number is depth on exactly one step of the method. The filled block is the paper you are holding; the two unnumbered pieces close the set.

Where every claim in this paper comes from

Where every claim comes from · continued

The three books named earlier in this paper are listed in full in number 02, so that they do not recur in every reference list.