Is AI driving away your best customers? 3 fixes for bridging gaps with growth audiences
It’s the last week of Black History Month (BHM) and it’s clear Americans are over performative values. Trite BHM-inspired merchandise sits on retailer shelves untouched while media is abuzz covering the artistry, activism, and symbolism of Bad Bunny’s Super Bowl halftime show. The signal is clear: consumers are looking to brands for real solutions to real problems, not products that commodify culture.
Most companies build everything from advertising to AI for the “average user,” but in doing so, they react to rather than lead markets. Strategic leaders look to growth audiences—underserved groups who are the fastest-growing demographics—as lead users. They are the “canaries in the coal mine” because they navigate the highest levels of systemic friction, making them the first to experience “average” design failures.
What does championing these lead users look like at a communications, product, or systems level? It looks like Elijah McCoy automating engine lubrication—an innovation bred from the friction between his engineering degree and the menial labor he was forced to perform, thus creating the “real McCoy” quality standard. It looks like Jerry Lawson changing the economics of the gaming industry by inventing the video game cartridge that divorced its hardware from its software. And it looks like emergency medicine becoming a global standard after being piloted by the Pittsburgh Freedom House Ambulance Service who, in the face of medical bias and systemic unemployment, also redefined emergency care as a public right.
Drawing from their lived experiences in underserved groups, these pioneers didn’t just solve problems; they mastered environmental friction. Today, that friction also manifests in algorithms. Championing growth audiences as lead users means ensuring they are critical AI system “stress testers.” When we fail to design for them, we allow AI data, development, and deployment to default to obtuse “averages” that can frustrate or drive away valuable customers. Three recent examples highlight issues and opportunities.
Relying on ‘Data Infallibility’ versus Lived Realities
In this Infallibility Loop bias, a brand’s AI trusts a data source—like a flawed GPS coordinate or outdated government map—as an absolute truth, even when customers provide contrary evidence. This is a digital echo of historical redlining: a systemic refusal to see humans over faulty data.
The Experience: A Black homeowner in an affluent area is penalized by an AI that confuses her address with a property in a different town, automatically forcing unnecessary flood insurance onto her mortgage and increasing the payments. Despite providing human-verified deeds and highlighting known GPS errors, the AI blocks her “incomplete” payments and triggers automated credit hits. A resolution only came months later after the consumer filed state-level servicer complaints.
The Fix: Prioritize Dynamic Qualitative Data Collection. Design should allow real-time, contextual evidence to override static, biased datasets. True brand innovation requires systems to yield to the experts: their customers.
Leveraging ‘Data Intimacy’ while Neglecting Situational Accuracy
This trust paradox occurs when brands use private data, but fail to combine situational data, making personalization feel like needless surveillance.
The Experience: During January’s recent record-breaking New York snowstorm, a customer called a national pharmacy’s location in her neighborhood to make sure they were open. The AI-powered interactive voice response (IVR) recognized her number, asked for her birthdate, and greeted her by name. Yet, after performing this exchange, it provided a “default” confirmation that the store was open when asked. Without a car, the customer braved life-threatening conditions on foot only to find a handwritten note on the door indicating it had closed due to the storm.
The Fix: Add Good Friction. A term coined by MIT professor Renee Richardson Gosline, “Good Friction” requires that when external context (like a Level 5 storm) conflicts with standard scripts, the system pauses and verifies first.
Prioritizing ‘Recency’ But Erasing Loyalty
Recency bias in algorithms weights the last data point more heavily potentially resulting in algorithmic erasure.
The Experience: A 20-year elite status customer calls an airline, only to be greeted by the name of his niece (a nonmember relative for whom he recently booked a one-off ticket) and then is erroneously deprioritized in the automated journey as a nonmember. In many “growth audience” and immigrant households, economics are multigenerational and communal, with a single “lead user” facilitating purchases for extended family. This airline system’s “memory” was shallow, seeing only the most recent transaction and ignoring a decades-long relationship because a reservation shared the same contact number.
The Fix: Focus on Holistic Design. AI must be weighted to recognize the arc of the customer journey, ensuring that loyalty isn’t erased by a single data point or the nuances of communal purchasing.
To be sure, bad data is a universal problem, but the lack of situational intelligence in our AI systems hits growth audiences—like Black consumers—first and hardest. Because these audiences represent a disproportionate share of future consumption and have the most “cultural common denominators,” their frictions are diagnostics for markets writ large. We aren’t just solving for a niche by championing them as lead users, we are adopting more rigorous, empathetic, expansive, and effective standards that solve real problems for all people.