Picture the spreadsheet. Sixty franchise locations, one column of revenue growth, sorted high to low. Someone circles the top five and says, "Interview those." It feels efficient, but nobody has asked what the column measures, how long the top five have been there, or what they were handed to start with.
So here is the answer up front. Choose your top performers by writing a selection rule before you look at names: one outcome, a stated calculation, a threshold, a minimum period over which it must hold, the context you will record, and a guardrail that must not get worse. Check that rule against the records. Then add a small group of ordinary performers who face the same conditions, so you can see what the best actually do differently. This guide gives you a worksheet to copy.
This is the selection step only. For the full method, read our end-to-end guide to top performer analysis.
Start with the decision, not the leaderboard
A leaderboard answers "who is highest?" Your study needs "what could we copy, and from whom?" The second question tells you how to pick.
Write the decision first, in one sentence. "We need to choose which handover routine to test across all branches." "We need to decide what to put in the onboarding plan for new reps." Each implies a different outcome and a different person to talk to. If you cannot finish the sentence, you do not have a study yet.
Then pick the unit: locations, branches, reps, agents, franchisees, channel partners. Choose the one that controls the practice. If the branch manager decides the routine, your unit is the branch, not the sales assistant.
Pick an outcome that resists gaming
Choose an outcome that is hard to fake and hard to luck into.
Be suspicious of any measure that people know is being watched. A Lancet analysis of hospital performance data argued that comparative measures of mortality and morbidity are often overinterpreted, and that the stigma of a poor ranking can pull attention towards superficial improvement or even gaming.1 Your sales and service figures are not hospital data, but the lesson carries over: a ranked number invites people to work on the number. A rep with the highest bookings may be pulling deals forward from next quarter. A branch with the fewest complaints may simply make complaining hard.
Three habits help:
- Prefer an outcome nobody can move by redefining terms. Choose repeat purchases, renewals or returned-job rates over "tickets closed" or "calls made".
- Choose a rate, not a total, when units differ in size. Revenue rewards the big territory. Conversion, retention, or margin per job rewards the practice.
- Pair it with a guardrail. Pick a second measure that would go the wrong way if someone were gaming the first, such as cancellations within thirty days next to bookings.
If two outcomes both matter, run two selections. A blended "composite score" hides what you wanted to learn.
Separate sustained results from luck and position
One great quarter is not a track record. The trap has a name, regression to the mean: a statistical effect that can make natural variation in repeated data look like real change. It becomes more noticeable when you follow up only a group chosen for a high baseline reading.2 That is exactly what "interview the top five" does.
So make "sustained" a written rule. Decide how many consecutive periods the result must hold, and include at least one stretch that tests ordinary operating conditions, not only the season everyone does well in. Fit the number to your business cycle: enough periods to cover at least one quiet stretch. If you report monthly, start at four and adjust to your cycle. Treat one standout month as a lead, not a qualification.
Then look at where the result came from. Two locations can post the same number for very different reasons:
- Territory. A site on a busy road, in a growing town, or beside a large employer has an advantage that comes from the place, not the practice. Record it.
- Customer mix. One branch serves mostly repeat commercial accounts, another serves one-off walk-ins. Their conversion rates are not comparable.
- Size. Treat the best period of a small location with more suspicion than the best period of a large one; a handful of cases can move its figure a long way.
Comparisons between units should show how precisely each figure is estimated, not only the figure itself. A funnel plot does this by charting each unit's figure against how precise it is, which helps avoid spurious ranking. Extra variation between units can also come from risk factors nobody measured.3 You do not need a statistician for this. Write down, for every candidate, the conditions they work under, and ask whether a fair observer would call the comparison like for like.
A location with a built-in advantage is a fine business, but a poor teacher. Look for people who succeed under everyone else's constraints.
The selection kit
1. The selection worksheet
Copy this table. Fill in every row before anyone is invited. An empty row is the next thing to fix.
| Field | Your answer |
|---|---|
| Decision this study will inform | |
| Unit (location, branch, rep, franchisee, partner) | |
| Outcome, as a rate | |
| Calculation (numerator, denominator, source system) | |
| Threshold that counts as "top", and why you set it there | |
| Periods in a row it must hold (check the rate on the pooled period first, then month by month) | |
| Periods that must be included (for example, one busy and one quiet) | |
| Smallest number of cases behind a figure that you will accept | |
| Context to record for every candidate (territory, customer mix, size, years open, inherited pipeline) | |
| Guardrail metric and the line it must not cross | |
| Exclusions and the reason for each | |
| Candidates who meet the rule | |
| Candidates close to the rule (kept as reserve, labelled as near-miss) | |
| Comparison units and the conditions you matched them on |
Write the threshold as a statement you can defend, such as "at or above the network median plus a margin we chose for this reason" or "above the natural gap in the sorted rates", rather than copying a percentile from a blog post.
2. Questions to ask the records before you invite anyone
Run each candidate through these. Write down every "no" and "don't know".
- Is the outcome calculated the same way at every location, with the same denominator and the same reporting period?
- Has any location changed how it records the outcome during the period?
- Does the figure hold for the full sustained period, or did one strong stretch carry it?
- Did anything unusual land in the period: a one-off contract, a closure nearby, a promotion, an inherited book of business?
- Does the guardrail metric hold up for the same period?
- How many cases sit behind the figure? Would a few deals or tickets moving change the picture?
- Who handles the work: the named person, or a team, a shared account, a manager covering?
- Did the candidate's conditions (territory, customer mix, size) differ sharply from the typical unit's?
- Would you be comfortable showing the candidate's own team the record that qualified them?
3. Nominations versus records
Ask area managers and regional leads who they think is best. Do it, and use the answer as a list of leads. Then check every nominee against the worksheet. Use nominations to find people the records miss, such as a new site with a brilliant manager and only a few months of data. Use records to test the popular, long-serving names.
Studying exceptional performers is particularly suited to situations where organisations can be ranked reliably on valid performance measures.4 If your records fall short of that, drop the label "top performers", write "nominated by their managers" in the study notes, and read the interviews as exploration.
Who else belongs in the sample
Interviews with the best people alone only tell you what they say they do. Do not take "we greet customers promptly and follow up quickly" as a finding until you have heard whether everyone else says it too. To find what is different, you need someone to compare against.
A systematic review of positive deviance studies in healthcare found that comparison groups were rarely included.5 A small group of typical performers who work under the same conditions turns "they all do X" into "they do X and the middle of the pack does not".
Purposeful sampling means choosing cases because they are rich in information about your question, not because they are easy to reach.6 For this study, treat typical cases, and sometimes a case that tried something and dropped it, as information-rich in the same way as the stars.
A short decision guide for the comparison group
| If your question is... | Add typical performers? | Who to pick |
|---|---|---|
| "What do our best do that we should teach everyone?" | Yes | Locations or reps with similar territory, size and customer mix, but middling results |
| "Why did one region pull ahead?" | Yes | Neighbouring units in other regions with the same product and customer mix |
| "What does a strong first year look like?" | Yes | Newer units with average first-year results, so you see where they diverge |
| "Is there something unusual and rare going on here?" | Optional | One or two ordinary units, to check the practice is not universal |
| "We only need ideas for a first draft of a playbook" | Not required | Say clearly that no comparison was made |
How many people
There is no law for this. The right number depends on how narrow your question is, how specific your sample is, how rich the conversations are, and how you will analyse them: the more relevant information the sample holds, the fewer participants you need.7
Plan a first wave and read the notes. Then check coverage: for each kind of context you recorded (territory type, size band), is there at least one qualifier and one typical unit? If not, add the missing unit before you add more of the same. List any open questions in the write-up.
A worked example
Here is an invented example to show the method.
A home-services franchisor runs sixty territories and wants to know how its best territories handle the first call from a new enquiry. The decision: which first-call routine to test across all territories.
The owner's first list is the top five by monthly bookings. Then they fill in the worksheet. The outcome becomes "enquiries that turn into a booked job within seven days", a rate, so large territories have no head start. The guardrail is "jobs rebooked within thirty days for the same fault". The smallest figure the owner accepts is twenty enquiries in a month. Then the owner sorts the conversion rates with the territory names hidden and sets the threshold at a natural gap: nine territories sit above it, and the next one down is well below. All nine clear the twenty-enquiry minimum. A tighter line would have left only two.
The rule requires four consecutive months, including the quiet one after the summer rush. The rate threshold is checked first on the pooled four months, then month by month for the four-month rule.
Of the original five, two do not clear the rate threshold at all: they are big territories that win on volume. The other three are among the nine.
Running the record questions then cuts the nine territories down to four:
- Two fail the four-month rule. Each had one month inflated by a one-off batch of enquiries from a housing contract and was ordinary in every other month. Both were in the original five.
- One fails the guardrail. It books fast, but many of its jobs come back.
- One changed how it logs a booking part-way through the period, so its figures are not comparable.
- One serves almost only commercial maintenance accounts, so its customer mix does not match the rest.
The exclusions row of the worksheet records the territory that changed its logging and the one with almost only commercial accounts, each with its reason. That leaves four qualifying territories, and only one of them was in the original five.
A neighbouring territory with a new manager has only three months of data. It is not one of the nine, because the pooled sort needs the full four months, and it cannot meet the four-month rule, though its three-month rate would clear the line. It goes on the worksheet as a near-miss, and the study notes label it as a reserve, not a qualifier.
The final sample is eight territories: the four qualifiers, three typical territories matched on size and customer mix, and the labelled reserve. The reserve is interviewed too, but its answers are read separately and not counted with the qualifiers. In the interviews, all four qualifying territories mention a call-back within the hour. So do two of the three typical territories. What only the qualifiers do is ask the enquirer to describe the fault in their own words before offering a time. That is the practice worth testing, and the call-back would have looked like the answer if the typical group had been left out.
Inviting people without making others feel ranked
Inviting a handful of locations and not the rest of the network sends a message. Decide what it says, and treat the invitation as part of the study.
Tell every invitee what the research is for, how you will use the results, and that taking part is voluntary and can stop at any time. These are among the things participants must understand for consent to be informed.8 Also tell them who is running the research, whether you will record, and how you will keep what they say. Follow your organisation's own consent and privacy rules for anything beyond that.
Four wording choices help:
- Describe the study by its question, not by a status. "We are learning how first calls are handled across the network" is true and neutral. Avoid "You are one of our top performers"; it turns an invitation into a ranking.
- Invite the comparison group in the same words. Nobody should be told they are the control.
- Say what happens next for everyone. Tell the wider network the study is happening, and share what you learn with everyone afterwards, so the interviews are not a secret contest.
- Keep identities out of the write-up. Report practices, not rankings.
What this will not tell you
The label "top performer" is only as good as your records, and a worksheet cannot make poor data good. Interviews describe what people say they do; they do not prove that the practice causes the result. A practice found only in your best locations may be a symptom of their advantages, which is why the comparison group and a later test matter. Choosing a threshold, a period and a sample size is judgement, and sensible people will draw the lines in different places. Write yours down, and let someone else challenge them.
Your next step
Open a blank copy of the worksheet and fill in the first four rows for your own network today. Then run the record questions on your first list. Related: top performer analysis.
