How can marketers ensure they’re targeting real people rather than shadows of activity? Karissa VanHooser of Epsilon explains how a malfunction in the AI behind Tesla’s on-board cameras illustrates the pitfalls of using data poorly.
Whether intelligent systems are dependable is down to data, explains VanHooser
Tesla’s cars have been seeing ghosts. Drivers have filmed their vehicles detecting phantom pedestrians wandering among headstones, an eerie spectacle that has become an unlikely Halloween story. The truth is not so supernatural; it's an AI malfunction, an algorithm trying to make sense of an ambiguous environment.
This is a vivid reminder that intelligent systems, however advanced, are only as good as the data that feeds them. And that holds a lesson for marketers who are also trying to interpret complex environments filled with incomplete signals.
Digital marketing is full of lookalike confusion. Every campaign depends on signals such as clicks, cookies, and page views: each treated as proof of intent. Yet, as with Tesla’s cameras, those signals are open to interpretation. I might be buying holiday gifts and end up with six pairs of headphones in my basket shortlist, looking like a die-hard audiophile. Minutes later, I’m on my phone buying a dress from a favorite brand, and the algorithm assumes I’m a brand-new customer.
This ambiguity is not an error in the system; it is a symptom of the environment itself. Consumers move fluidly across devices, channels, and contexts, leaving behind traces of behavior that are easy to collect but hard to interpret. When algorithms act on these fragments, they see movement without meaning. The result is wasted spend and false confidence in what is being measured.
Advertisement
Relying on one weak signal is like driving with one eye closed. You can perceive activity, but not depth or distance. Much of digital marketing still depends on this narrow view because surface-level data is easy to measure, yet it captures only the shadow of a person, not the person themself.
True understanding comes from combining diverse signals that reveal intention as well as action. Purchase history, loyalty participation, search behavior, and store visits together provide the texture that a single cookie never could. When these are connected into a consistent picture, marketers move from reacting to noise to recognizing genuine opportunities.
The introduction of AI will not automatically clear this confusion; in some ways, it may deepen it. New agentic and conversational systems depend on prompts that are, by design, simplified instructions. I might ask: ‘show me running shoes under $100,’ setting boundaries that determine what the AI can see. The system delivers options within that frame, but it does not know the broader context. It understands the request, not the reason behind it.
Marketers are beginning to recognize these limits. According to a recent report, model accuracy is one of senior marketers’ most cited technical challenges in AI implementation (46%), closely followed by integration with other systems. Both issues point to the same underlying problem: when AI operates in isolation, without full access to connected data, its view of the customer remains partial.
If marketers optimize campaigns around what AI agents surface rather than the full context of real consumers, they begin calibrating to artificial journeys. These are not complete paths but isolated moments defined by parameters the customer sets, often without realizing it.
Without anchoring those AI-led interactions in richer data such as loyalty engagement, purchase behavior and prior exposure, the connection between brand and person weakens. The algorithm begins to optimize to its own view of the world, which may be precise but disconnected from reality.
The lesson from Tesla’s ghost sightings is simple. Algorithms need reference points. When data lacks depth or variety, systems invent meaning that is not there. Marketing technology behaves no differently; it can only interpret the world through the signals it receives.
Advertisement
To see clearly and get full value from their campaigns, marketers need both identity and AI. When transactional, behavioral, and contextual information converge, the picture becomes sharper. Marketers can harmonize their brand’s conversations with consumers across devices as well as paid, owned, and earned channels. Campaigns stop chasing isolated metrics and start focusing on outcomes that represent genuine engagement. AI, when guided by this kind of clarity, becomes a practical assistant rather than an unreliable narrator.
Tesla’s phantom pedestrians may be a fitting Halloween story, but they also serve as a parable for marketing’s next chapter. Intelligent systems will continue to shape how people discover, choose, and buy. Whether those systems are dependable will hinge entirely on the quality and completeness of the data that drives them.
Algorithms trained on real signals are the first step, enabling recognition of real people rather than shadows of activity. But the real unlock comes when AI can use that understanding to decide how, when, and where to engage each consumer. That is how marketing moves from chasing phantoms to responding with the living world of real customers.
Epsilon
Epsilon is a global advertising and marketing technology company positioned at the center of Publicis Groupe. We connect advertisers with consumers to drive performance...
