Why $57 Billion in Location Advertising Is Targeting the Wrong Customers

The standard playbook for local advertising says: draw a circle around your store, target everyone inside it. It is intuitive, easy to execute, and according to a new Harvard Business Review study, systematically wrong.

A six-year analysis of millions of retail visits across grocery, pharmacy, home improvement, and mass merchandise categories reveals what researchers are calling a "donut effect": customers within 4 miles of your location are the least responsive to your ads. The customers most likely to change their behavior based on what you show them live in a ring from 4 to 14 miles out. Past 14 miles, responsiveness falls off again.

If you have been running geotargeted campaigns using standard radius-based tools, you have been paying to reach the people who need you least.

Why the Inner Ring Does Not Respond

The intuition behind tight geotargeting is sound - proximity implies relevance. But HBR's analysis of retail visit data suggests the relationship breaks down at close range for a simple reason: customers who live near your store already know you exist. They have made their decision. Advertising to them is not influencing behavior - it is overhead.

The researchers found that within 4 miles, promotional ads (discounts, limited-time offers) still move the needle somewhat, because people already predisposed toward you are willing to act on a deal. But brand-building messages - the ones that shape perception and pull customers away from alternatives - have their highest impact at moderate distances, in the 4-to-14-mile band.

The Competitive Proximity Variable Nobody's Measuring

The deeper finding is more structurally interesting: relative distance to competitors is a stronger predictor of ad responsiveness than absolute distance from your own store.

A customer who lives 8 miles from your grocery store but 10 miles from every competitor is more responsive than a customer who lives 3 miles from your store but also 2 miles from two direct rivals. The first customer's decision is still open. The second has already been settled by geography.

Current geotargeting platforms - Google's local campaigns, Meta's radius targeting, most DSPs - cannot express this constraint. They draw circles around your store. They cannot filter for "closer to us than to any rival." This is not a minor omission. It means that across $57 billion in location-targeted U.S. advertising spend in 2025, a key predictive variable is invisible to the bidding infrastructure.

This connects to a pattern we have written about before: why agencies confusing brand visibility with brand relevance keep getting geographic targeting backwards. Visibility is about showing up. Relevance is about showing up for people whose decisions are still open to influence.

This is the kind of structural gap STI's research tracks systematically - where the tool and the truth diverge, and organizations keep paying the difference.

What Donut-Aware Targeting Would Actually Look Like

The researchers suggest three operational changes that marketers can act on now, even before platform capabilities catch up.

First, layer competitive store locations into your audience modeling. If you have access to census-block or zip-level competitive density data, use it to identify customers who are geographically equidistant from you and a key competitor, or who are closer to your location. These are genuinely persuadable audiences regardless of absolute distance.

Second, differentiate creative by distance band. Promotional and discount messaging converts best in the 1-to-4-mile range, where you are reaching your existing loyalists and semi-loyalists who respond to activation triggers. Brand and awareness creative belongs in the 4-to-14-mile ring, where perception is still being formed. Mixing these signals across a single radius campaign is why many location campaigns produce mediocre average results - the two audience types need different messages.

Third, suppress the innermost ring for brand spend. This is the most counterintuitive recommendation, and also the most defensible based on the data. Customers within a mile or two of your store have already formed a view of you. Brand dollars directed at them are unlikely to shift anything meaningful. Redirect that budget outward.

If you are evaluating targeting strategies against these criteria, our analysis tools can help surface what the pitch decks won't - including whether your current geotargeting setup treats competitive proximity as a variable at all.

The Housing Market Is Running the Same Calculation Error

The donut effect is not a quirk of advertising mechanics. It is an example of a broader problem: decision systems that optimize for the observable proxy while missing the predictive variable.

The U.S. housing market has been running a version of this error since 2022, and new analysis from Of Dollars and Data makes the pattern legible. Seventy-five percent of the largest U.S. metro areas saw inflation-adjusted home price declines over the past year. New home sales dropped 17.6% in January 2026, reaching their lowest point since October 2022. Purchase cancellations hit record highs. The top 15 cities by rental volume have all seen double-digit rent declines since 2022 - Austin leading but not alone.

The nominal price story was relatively calm for most of this period. The real price story - adjusted for inflation, income growth, and carrying cost - was deteriorating faster. The income required to afford a typical home nearly doubled from pre-COVID levels. That affordability ratio is the predictive variable. Nominal list prices were the observable proxy.

This is not a new dynamic. We looked at how financial brands should position into affordability cycles after McKinsey's affordability data last quarter. The underlying conclusion was the same: the brands that navigate these transitions well are the ones tracking real variables, not headline numbers.

The Equity Cushion Argument, and Why It Does Not Close the Gap

The standard counterargument to housing pessimism is the equity cushion thesis. Homeowners built record equity during the 2020-to-2022 price run. They can cut asking prices and still profit. This creates a structural buffer against the forced-sale cascades that accelerated the 2008 collapse.

That argument is correct, and it does not resolve the demand problem.

The ability to cut prices does not create buyers who can afford to buy at the cut prices. Income-to-price ratios are the binding constraint, and they have been moving adversarially for two years. The equity cushion is the inner ring: it is real, it is comfortable, and it makes sellers feel safe while the actual demand destruction is happening further out - among the population that would buy, but cannot qualify at current rates and prices.

Why Organizations Consistently Optimize for the Wrong Variable

There is a consistent pattern across the advertising data and the housing data that deserves a direct answer: why do organizations keep optimizing for observable proxies when better predictive variables exist?

The answer is usually one of three things. The predictive variable is harder to compute. The tooling only supports the proxy. Or the proxy worked well enough in the past that questioning it never became urgent.

Radius-based targeting exists because drawing circles around coordinates is trivially easy to compute. Affordability indices exist but are not how real estate platforms price listings or how mortgage originators think about pipeline. Follower counts are how most brands evaluate creator partnerships, even though cultural velocity - the rate at which a creator's audience actually adopts behaviors - is the variable that predicts whether the partnership moves product.

Cam Newton made exactly this argument at the 2026 IAB NewFronts. His production entity (Iconic Saga Productions) has built genuine audience ownership across podcasting and live events targeting HBCUs. His pitch to brands was precise: "It doesn't have to cost you or change who you really are." The undervalued variable is not raw reach. It is the authenticity premium that determines whether reach converts. Most brand partnership deals are still priced on follower count because follower count is the number that fits in a spreadsheet.

This is the same logic that produces the intent/impression confusion in performance marketing more broadly. Impressions are countable. Intent to act is harder to surface. The tendency to optimize for the countable over the predictive is one of the most consistent structural biases in how organizations use data - and one of the most expensive.

The Pattern Worth Stress-Testing Before You Spend

The HBR location advertising study is useful precisely because it is operational. It is not a theoretical argument about decision quality. It is a documented, six-year, multi-vertical dataset showing that the standard practice costs real money and produces measurably worse outcomes than an alternative that requires slightly more sophisticated targeting logic.

That kind of documented counterfactual is rare. The housing market's affordability gap will not produce a clean study showing the buyers who could not qualify and would have bought. The athlete creator authenticity premium will not appear in a single campaign's attribution dashboard. These gaps persist because the cost of the wrong variable is diffuse and the benefit of fixing it is hard to attribute to a single decision.

What the pattern suggests, practically: the variables that are hardest to put into a standard dashboard are often the ones worth building toward. Competitive proximity is not in your geotargeting platform. Real affordability is not on listing sites. Cultural velocity is not in influencer analytics. These are the outer rings - where the decisions are still open, and where the returns on reaching the right audience are highest.

The question worth asking before the next campaign, market entry, or partnership decision: which variable are you actually optimizing for, and is it the one that predicts behavior?

That's the kind of analysis our research tools are built for. Start there before committing the spend.