Two stores serving the same trade area can report very different results because a site model evaluates the location, not the manager assigned to it. In research on two large retail chains, economists attributed 25% to 35% of the variation in store-level productivity to individual managers. That factor is real and sizable, but it remains separate from the site score.
This is a familiar store. The demographics, traffic counts, and co-tenants all looked right, and its score was good enough that no one challenged it in committee. Two years later, it ranks in the fleet's bottom quartile. One person says the model missed; another says the model was fine but the store was poorly run. Both are guessing because no one recorded which bet they were making.
Why two stores with the same trade area do not perform the same
A site model scores the market using population, incomes, daytime traffic, competitors, and access. Store results reflect those conditions along with the operation built on top of them. Rent, build-out, format, opening timing, staffing levels, and the person running the floor all affect the P&L, while a trade-area score captures almost none of those factors.
The gap doesn't indicate a flawed score. It shows the score is doing its job: a trade-area model tells you what the ground can support before you've hired anybody. The issue starts when a store misses and the team treats that score as though it had promised something bigger.
The disagreement rarely gets easier because the two sides have different incentives. Real estate is trying to defend the site, while Operations is trying to defend the operator. Neither side has evidence, so tenure and volume end up carrying the argument. Our piece on when to trust your gut in a site decision explains why that instinct is never corrected: this decision comes up a few dozen times in a career, and the answer appears years later, with a manager change buried inside it.
What the research says a store manager is worth
Economists Robert Metcalfe, Alexandre Sollaci and Chad Syverson examined store-level data from two multibillion-dollar retail companies, where managers transfer among stores while firm policy determines management practices and keeps them fixed. This setup separates the individual manager from the company's playbook and quantifies an issue most operators otherwise debate.
Published as NBER Working Paper 31192 in April 2023: the finding is that managers account for 25% to 35% of the variation in store-level productivity. The authors estimate this at roughly 50% to 70% of the explanatory power of store fixed effects. That comparison needs careful reading: a store fixed effect captures everything permanent about the store, including the trade area, building and format together. The person running it therefore accounts for somewhere around half to two thirds as much variance as all of those factors combined.
The paper finds that, across the four largest store groups connected through manager moves, shifting a manager from the 10th percentile to the 90th raises productivity by between 22% and 82%. Results vary widely across those store groups, so the authors report a range rather than one clean headline number. Even the bottom of that range exceeds most of the site differences a committee spends its afternoon arguing over.
For a real estate team, two additional points in that same paper matter more than its headline number.
Good managers are not where you would guess. The study finds negative assortative matching between managers and stores and presents several possible explanations. One possibility is that companies intentionally place their strongest managers in weaker locations to support them. The authors also identify a selection bias that can affect this kind of research, rather than dismissing it. If the first explanation is correct, manager quality is not mere noise in store results. It creates a misleading signal because your best operators are disproportionately assigned to your worst buildings.
Manager quality barely shows up in the data you already hold. Looking at tenure, gender, wages, and distance to the closest competing store, the analysis found no statistically significant relationship with manager quality. The only observable that showed one was the ratio of full-time to part-time workers, and its relationship ran in opposite directions for the two companies. Even personnel researchers with payroll-level detail couldn't identify a reliable leading indicator of who would turn out to be a good manager, so a site model built from census blocks and mobile panels isn't going to do better.
The finding extends beyond retail. In The Value of Bosses, Lazear, Shaw and Stanton examined technology-based services workers, not store staff. They found that replacing a bottom-decile boss with a top-decile boss increased a team's output more than adding a tenth worker to a nine-member team would have. The industry differed, but the conclusion was the same.
Why putting the operator inside a site score breaks both decisions
Once operator quality is built into a site score, the result no longer describes the ground alone. When something goes wrong, you can't tell whether the score or the hire was at fault, so the feedback loop needed to improve next year's scores goes silent. Both decisions worsen together, in ways no one can see.
A simpler concern is that, when you sign, you usually don't know who will run the store. The lease binds you before that hire exists, so any operator quality included in the score is an estimate about someone not yet selected, presented as though it were a property characteristic.
The sequence makes the argument. Staffing can be changed through the schedule. A manager involves hiring and ramp-up, and within a year you'll know whether it worked. A lease isn't something you can revisit, while rent is due for the space either way. Building a ten-year commitment around a variable that can be replaced in a quarter puts the math in the wrong order.
The model's back-test identifies this same failure from the opposite direction. When a vendor demo scores your worst store an 87, the more candid interpretation is often that the real estate is fine, while the problem sits elsewhere in the business. That is useful information only when someone states it plainly rather than quietly expanding what the score is meant to cover.
Three comparisons that separate a site problem from an operator problem
A store that misses its plan raises three questions, not one, and sequence matters. Keep one factor constant in each comparison: begin with the market, then assess the format and the economics you signed, and finish with everything physical about the store. Once you run all three, the variable still moving is your answer.
Start with the market. Assess the store alongside your other locations in the same metro. Regional demand, weather, local economic conditions, and competitive entry affect that group together. If every location misses at once, the issue is with the market assessment, not an individual store. The discussion should then involve whoever defined the trade area, rather than whoever manages the floor.
Then hold the economics still. Assess the store alongside your own locations that share its format, rent band, and access pattern. If it still stands apart from those comparables, the ground really is contributing something they do not, giving you a site finding worth recording.
Only then look at the operator. Compare a store's performance before and after its manager changes. The trade area, box, and lease stay fixed, so the real estate remains constant. When results change with the people, you've isolated the operation about as cleanly as a fleet of stores will allow.
Most teams head straight for the third rung. Usually, someone in the room has already formed an opinion about the manager. That shortcut makes the trade area take the blame for the person, or the person for the trade area. Our guide to fixing underperforming stores with location data looks at that same diagnosis from the portfolio side.
Write the operator assumption down, then check it after you open
Your site package already rests on an unstated operating assumption: this location will be run roughly as well as your average store. Recording that assumption in the deal costs nothing, and eighteen months later it gives somebody a claim to check against what happened instead of a vague excuse.
Get specific. Identify the person assigned to operate the location and whether they've opened a store before. Then assess the local hiring bench: can you replace them with one phone call, or could recruitment take six months? Check whether the pro forma is based on a labor model you've actually operated, not the one you wish you had. None of this affects the score, but it does shape what you should expect from the store. You can know all of it before signing.
Then use it in the post-opening review. The question isn't "did the store hit plan"; it's "which of our assumptions missed." A defensible forecast already records the numbers, so record the operator assumption the same way. If a store misses with the manager you planned for, that's a site finding. But when two managers have turned over in a year, the miss isn't evidence about the ground at all.
That distinction carries more weight in restaurants than most site packages acknowledge. Black Box Intelligence, in its survey of 158 restaurant brands, reported full-service management turnover of 38% in the third quarter of 2024, compared with 31% in 2019. Limited-service management turnover was 55% versus 45%. These figures came from brands that responded, not an industry census. But at anything close to those rates, the operator isn't a permanent store attribute you assess once. The operator keeps changing while the lease remains in place.
How GrowthFactor keeps the site question and the operator question apart
GrowthFactor scores each site through configurable lenses and displays every input that changed the score, keeping the model's claims about what it knows visible on the page. The score, trade area, and deal all belong to one pipeline, so user-entered assumptions are recorded beside the number rather than buried in someone's inbox.
That is the mechanism supporting the distinction this entire piece relies on. Open a score and inspect its variables, and you can identify precisely which parts of a store's result the score was ever intended to explain. You can also record the operator assumption in the deal record with the lease terms and site score, keeping all three together for the review two years later.
We don't receive a percentage of any lease our customers sign, so there's no reason to broaden what a score covers just to make a deal look better. The score addresses the ground question. The operator question belongs to your team. Keeping them separate is what makes either answer worth having.
Frequently Asked Questions about operator quality and site selection
Why do two stores with the same demographics perform differently?
Because a site model grades the trade area and your P&L grades the whole business. Rent, build-out, format, opening timing, staffing levels, and the person managing the floor all move store results, and almost none of them appear in a trade-area score. Two stores can sit on statistically identical ground and still separate by a wide margin on everything built on top of it.
How much does a store manager affect store performance?
More than most site packages assume. Studying two multibillion-dollar retail chains where managers move between stores while company policy stays fixed, Metcalfe, Sollaci and Syverson found that managers explain 25% to 35% of the variance in store-level productivity, roughly 50% to 70% of what everything permanent about the store explains. The effect is real and it is large enough to swamp a modest difference in trade area quality.
Should a site selection model include operator quality as an input?
No, for a practical reason rather than a philosophical one. At the moment you sign a lease you usually do not know who will run the store, and a score that folds in a guess about staffing stops being a statement about the ground. You also lose the ability to tell whether a miss was the site or the hire, which is the one comparison that would make next year's score better.
How do you tell whether a store is underperforming because of the site or the manager?
Run three comparisons in order. Compare the store against your other stores in the same market, which holds regional demand constant. Compare it against your own stores with the same format, rent band, and access, which holds the economics constant. Then compare the store against itself before and after a manager change, which holds the real estate perfectly constant. Only the third one isolates the operator.
How does GrowthFactor compare to Buxton on what a site score can and cannot see?
Buxton, now part of Audiense, delivers its Customer Value Site Score through a consultative, analyst-mediated engagement built on the brand's CRM data and psychographic segmentation, so the model sits with their team rather than opened inside the product. GrowthFactor scores sites across configurable lenses and shows which inputs moved each score in the app, so your team can see exactly which parts of a store's performance the score is claiming to explain and which parts it never touched. Neither one can score your operator. The difference is whether you can tell that from the number.