Editor’s note: Today’s blog comes from Laurent Madeleni, Principal, Spatiotemporal Science at AKQA. Laurent shares a summary of his analysis using Places Insights and Population Dynamics Insights, which helps customers train better predictive models with embeddings derived from Google search, maps, and weather trends.
Some brands thrive in unexpected places. Now, we can measure exactly why — analyzing 85,000 locations across eleven major brands to map out brand market fit.
The same rating but opposite performance ratings
Two coffee shops from the same chain both hold a 4.3-star rating. In a town where the local rating average is 4.5, a 4.3 rating falls short of the local expectation. In another town, the norm is 4.0 — the same rating makes it one of the best shops around.
|
Town A - Local coffee norm 4.5 |
Town B - Local coffee norm 4.0 |
|---|---|
|
4.3 ★ — below the bar Customers here expect more, and this store isn't delivering it. |
4.3 ★ — above the bar Same rating. Here it beats what the town usually gets. |
Every place quietly sets its own bar, and sets it differently for coffee, fast food, pizza, convenience stores. So the useful question is "where do we beat what the locals expect”? Rather than "where are our ratings highest?" We call that difference brand market fit.
What AKQA researched to predict brand market fit
The study was developed using two Google Maps Platform insights products – Places Insights and Population Dynamics Insights.
Places Insights helps customers derive custom statistical insights by securely joining their business data with Google's comprehensive Places dataset in BigQuery, providing aggregated venue counts, ratings for millions of American businesses. From it we measured the average ratings for fifteen everyday Places Types in every US postal code, then checked roughly 85,000 locations of eleven familiar brands — against the average ratings of the local area.
Population Dynamics Insights provides the second ingredient: it synthesizes search trends, mobility, and environmental data into rich, privacy-preserving embeddings that can be joined with your own outcome data to predict location performance and fill data gaps. It lets us ask a simple question at national scale: do markets that behave alike perceive Places alike?
The answer: They do. And we tested it the hard way: to predict a brand's fit in any state, the model was only allowed to learn from other states. The model worked — in the same direction for all eleven brands.
Location is a small share of a rating but can make (or break) a store rating
When the same brand runs two stores in the same neighbourhood, their ratings are often wildly different. Measured across 47,570 same-brand locations sharing a postal code and comparing such pairs demonstrates exactly how much rating weight is based on the store (79.6%) versus the place (20.4%) - or 1 star out of 5.
What defines a brand place rating?
No place-based method could ever predict most of a rating — the signal simply isn't there. But what matters is that the share belonging to a place is real, measurable, and different for every brand. That share is what we can now predict and our method captures about two thirds of everything it contains.
Two brands in the same category succeed in different places
The clearest example is two chicken chains. Same food category but in near-opposite geographies:
Chicken chain A
Rates higher than its local average ratings in spread-out, suburban and small-town America — places with drive-thrus and gas stations rather than cafés, brunch spots and dog-friendly patios.
Chicken chain B
Rates higher in denser, takeout-oriented neighbourhoods with more delis and corner stores — places where parking matters less and foot traffic matters more.
Place attributes impact brand location rating.
And the data offers a humbling reality check for brand value propositions and priorities. A well known coffee chain talks about being an urban "third place" — sofas, laptops, lingering. But the places where the coffee chain most outperforms expectations look different: suburban, car-friendly, group-and-kids, order-ahead. The data doesn't know what a brand wants to be. That is exactly what makes it useful.
A brand's proposition has a geography. Now it's one you can measure — before you commit resources to a market.
The test and limitations
Over two and a half years we watched these brands open 1,445 new locations — stores that didn't exist when our model was built. Where we had predicted a warm reception, new stores were received better; where we had predicted a cold one, worse. The model held for all eleven brands studied.
Equally important are the boundaries confirmed by the analysis:
-
It can't predict where chains open stores — that's a real-estate decision, and our test confirmed it. We predict how a store is received, not where one gets built.
-
It says nothing about any individual person's tastes — only about places.
-
It doesn't prove cause, doesn't measure sales, and can't rescue a badly run store — remember the 80%.
How this model can work for other brands
For any brand with enough locations, this method answers three practical questions:
-
Where does our proposition already over deliver?
-
What do those places have in common in plain words (not model weights)?
-
And what other locations in the country look just like them?
This is a new input for choosing new places, picking test cities, and writing customer-friendly value propositions that fit the local place. The traditional question was where are the people we want? This adds a different layer: where does our brand belong?
Developed with Google Maps Platform Places Insights (the counts) and Population Dynamics Insights (the population embeddings). The study included eleven QSR brands, ~85,000 outlets, 33,791 US postal codes. Every claim included is measured and the boundaries tested.
