The symptom
Consider a network of twenty locations whose combined performance reporting shows stable, consistent results period after period. On the surface, this reads as a healthy, well-run network. Underneath the aggregate, though, four sites are declining while the rest are steady or improving enough to offset them in the combined figure. Whether those four sites share anything in common has not been examined — two happen to be in the same region, two are not, and no one has yet established whether that is a meaningful pattern or coincidence.
This is not a case where the obvious explanation is either clearly right or clearly wrong. The reporting problem is straightforward: the aggregate figure conceals real divergence, and that much is not in question. What is genuinely unresolved is the operational question underneath it — whether the four declining sites are four separate, unrelated stories, or evidence of something the network has not yet identified.
What aggregation does to information
V. F. Ridgway’s Administrative Science Quarterly research on performance measurement, published through Cornell’s Johnson Graduate School of Management, is frequently cited for its warning about single measures distorting behaviour — but its broader finding applies just as directly here: composite measures, not only single ones, can produce misleading pictures of overall performance. A network-wide average is a composite measure by construction. It is designed to answer “how is the network doing,” and it answers that question reasonably well. It was never designed to answer “how is any particular part of the network doing,” and using it for that second question, which is what happens whenever an average substitutes for site-level review, will systematically hide exactly the kind of divergence described above.
Steven Kerr’s research on reward systems, published in Academy of Management Journal, offers a related and important warning about what organisations pay attention to: attention tends to follow whatever is measured and rewarded, and what is measured and rewarded is often the aggregate figure that is easiest to report at the board or investor level, not the site-level detail that would surface early divergence. This is not a criticism of any individual’s diligence. It is a structural consequence of what gets built into standard reporting. A network can be genuinely well-run at the level its dashboards track and still be blind at the level where the underlying performance actually varies.
Jeffrey Bradach’s field study of five large US restaurant chains found that chains combining company-owned and franchised units achieve consistent overall performance partly by managing variation actively across the network, not by eliminating it — a finding consistent with the idea that divergence between sites is a normal, expected feature of a healthy multi-site network, provided it is visible enough to be understood, rather than a warning sign that something has gone wrong.
Why wide variation is the normal state
It is worth being explicit about something the symptom above easily obscures: performance varying “widely” across a franchise network is not, by itself, evidence of a problem. Different locations operate in different local markets, with different demographics, competitive conditions, staff tenure, and physical formats. Some real variation is the expected consequence of a model being applied across genuinely different conditions, not a sign that the model or its execution has failed anywhere.
The actual operational question a network needs answered is not “why does performance vary” — it always will — but “does the variation follow a pattern worth understanding, or is it noise.” These are different questions and they call for different responses. Noise-level variation, where sites move up and down somewhat unpredictably around a stable average, generally does not need central intervention; it reflects the ordinary texture of local business conditions. Patterned variation — where a specific group of sites, sharing some identifiable characteristic, consistently moves in one direction — is a different matter entirely, because it suggests something structural: a regional factor, a format difference, a supply issue, a shift in the demographic assumptions the model was built around.
The difficulty, as in the scenario above, is that from inside the aggregate figure the two look identical. Distinguishing them requires deliberately looking underneath the average at a level of granularity most standard network reporting is not built to provide, and asking a comparative question — what, if anything, do the moving sites have in common — rather than a purely descriptive one about whether performance overall is acceptable.
There is also a timing problem that compounds this. By the time enough declining sites accumulate to move the network average visibly, the pattern behind their decline — whatever it turns out to be — has typically been operating for some time already. A reporting structure built around the aggregate will always be a lagging indicator of exactly the kind of divergence that matters most, because averages are mathematically designed to smooth out the early stages of any trend before they become large enough to show through. This is not a flaw that can be fixed by reporting the average more frequently; a faster-updating average is still an average, and still hides the same underlying movement it is built to summarise. What changes the picture is looking at a different level of the data, not looking at the same level more often.
Where this shows up
This perspective is written with franchise networks in mind. It sits within the Multi-Site and Franchise grouping, and the questions it raises recur in other sectors too.
Questions worth considering
- Does current reporting show site-level variation clearly enough to distinguish a pattern from noise, or only the network average?
- When a subset of sites moves in the same direction, has anyone examined what they share, or is each treated as an individual case?
- Is stability in the headline figure being read as evidence of health, when it may simply reflect offsetting movements underneath?
Your team is facing similar challenges?
Subscribe to The Riddle Ledger
Perspectives are sent as they are published, with no promotional email, and you can unsubscribe at any time.