The Target Story Isn't About Coupons. It's About Healthcare AI.

Twelve years ago, a Target statistician built a model that could predict pregnancy from 25 shopping items. The story usually gets told as a privacy parable. I'm telling it differently — as a preview of how healthcare AI is going to work for the rest of our lives.

Your smartwatch can already flag atrial fibrillation days before a cardiologist would. It can detect depression weeks before clinical scoring catches it. It can spot cognitive decline six to twelve months before you notice. The science isn't the question anymore. The question is who gets to see what comes out of the model — and whether we build the governance before the surveillance economy locks in.

In this episode I get into:

• Why Andrew Pole's 2012 Target model was a dry run for what's coming in clinical AI
• What the wearable accuracy numbers (70–95%) actually mean — and where they get softer than the headlines suggest
• The function-creep economy that's already running: CGM data sold to ad partners, period-tracking app subpoenas, life insurers bidding on de-identified wearable sets
• The 99.98% problem — why "de-identified" data isn't
• Habit-disruption windows: the real case for early-detection surveillance in healthcare
• Four policy moves that would change the data-broker incentive structure overnight

Full written companion with sources and citations: danmccoymd.substack.com

Watch on YouTube: https://youtu.be/LbE6TbGIzIY

I'm Dan McCoy. Ignition by RocketTools is the podcast for healthcare executives, physicians, and AI builders trying to think clearly about where this is all going. New episode every week.

Transcript

SPEAKER_00 0:00

There's a story about Target that keeps coming up in tech and healthcare circles, and it's the kind of thing that makes you immediately check your privacy settings. A father walks into a Minneapolis Target store and he's furious. His teenage daughter, still in high school, just received a mailer full of coupons for baby clothes, cribs, and maternity products. He demands to speak to the manager. Why ask, are you sending my daughter coupons for baby items? The manager apologizes, embarrassed, but a few weeks later, that same father calls back. Turns out his daughter was pregnant. She just hadn't told him yet. Target knew before her own father did. This wasn't magic. It was Andrew Pohl, a Target statistician using predictive analytics to identify about 25 products that, when purchased together, created a pregnancy prediction score. Uncented lotion in the second trimester, calcium, magnesium, and zinc supplements around week 20, larger purses, extra cotton balls, and hand sanitizer as delivery approaches and term gets near. The algorithm could estimate a due date within a small window and start sending targeted, and no pun intended, coupons at exactly the right moment to capture a lifetime customer. Here's what makes this story complicated. Target knew they were creeping people out, so they got sneaky. They started hiding baby product coupons among random items like lawnmowers, wine glasses, regular stuff, so it didn't look like they knew. Their surveillance continued, they just made it less obvious. Now here's the question that should make us all a little bit uncomfortable. If a retail algorithm can detect pregnancy from shopping patterns, what can healthcare AI detect from the thousands of data points we generate every single day? Let's start with something that most people don't know. Healthcare has spent decades trying to close care gaps, and it mostly doesn't work. Care gaps are the medical equivalent of target knowing you're pregnant, but sending you the wrong coupons anyway. It's when the data says you need a diabetes screening or you're overdue for a mammogram or your blood pressure medication isn't being refilled, and the system sends you a reminder letter that you ignore. Here's what the research actually shows. Only about half of those interventions actually improve healthcare outcomes. The ones at work, they have dedicated people managing your care, educating you about medications, and using actual tools to track populations, not just automated reminder letters. But here's the deeper problem. Traditional care gap interventions are reactive. They're trying to fix problems that already exist. You're already diabetic, you're already hypertensive, the horse has left the barn, and we're sending you a coupon for a barn door repair. What if, and this is where things get really interesting for me and maybe a little uncomfortable, AI could detect disease before symptoms appear. Not by looking at your medical record, but by analyzing thousands of subtle behavioral signals you don't even know you're broadcasting. This isn't science fiction anymore. Your Apple Watch or Fitbit isn't just counting steps. The latest generation of these devices can detect how your sleep patterns change over weeks, tiny alterations in how you walk, shifts in your heart rate during specific activities, and changes in your daily routine you wouldn't consciously notice. And when AI analyzes all this together, it can predict health problems with the accuracy that rivals or beats many traditional medical tests. Here's what that looks like in practice. Let's talk cardiovascular issues. The system can detect irregular heart rhythms days before heart attack, not from one measurement, but from patterns in how your heart responds during your morning walk, your afternoon meeting, or your evening Netflix binge. How about diabetes risk? AI spots insulin-resistant patterns before your blood sugar ever shows up as high on a lab test. It's looking at how your body responds to meals, exercise stress, building a picture your annual physical would actually miss. How about cognitive decline? This one's kind of eerie to me. Behavioral changes six to 12 months before you notice memory problems. Maybe you're walking more slowly. Maybe you're checking your phone more often because you're forgetting things. Maybe your sleep is more fragmented. Individually, these mean nothing, but together they create a pattern. Mood disorders are another thing. Sleep disruptions that combine reduced activity with changes in social interaction patterns can really indicate depression or anxiety before you'd even consider seeing a therapist. Remember my retinal detachment story? I saw wavy lines, a clear symptom of a retinal detachment. But an AI system analyzing my phone usage, walking patterns, and even how I held my head might have detected that problem earlier when treatment options were actually better. Had a good outcome, though. The research backing this up, a recent analysis of 49 different studies found that these systems are getting it right about 70 to 95% of the time. Another study showed 92% accuracy predicting chronic disease risk. These aren't theoretical numbers anymore. They're from real systems analyzing real people. Now we hit the uncomfortable part, and this is where the target story becomes a warning, not just an anecdote. Here's something most people really don't realize. The same privacy laws protecting your health data, we're talking HIPAA in the US and for our European friends, GDPR, also make it harder to detect when these AI systems have been tampered with or manipulated. Think about it. Privacy protections prevent hospitals and tech companies from comparing patient data across institutions. That's good for privacy, but it also means if someone wanted to poison the AI training data to make the system fail for specific groups of people, it would be much harder to catch. The back doors could pass all quality standard checks because nobody's allowed to look across enough data to spot the pattern. But the bigger concern, what researchers are calling function creep, you sign up for continuous glucose monitoring to manage your diabetes. The data helps you adjust your diet and insulin, that's great. But then your insurance company uses that data to adjust your premiums. Your employer considers it during health plan negotiations. Immigration officials request it as part of a visa application. A data broker sells your patterns from your glucose data to third parties, none of which you explicitly consented to, by the way. But once the surveillance infrastructure exists, it's incredibly hard to limit how it's used. Here's another problem. AI doesn't watch everyone equally. These systems create risk profiles and they tend to intensify surveillance on people who are already marginalized, often without any health benefit and with significant risk of discrimination. And here's the kicker. You can't see or challenge these profiles. Insurance companies, tech firms, data brokers, they're accumulating detailed pictures of your health behaviors and using them to make decisions about your cost, access, and your coverage. You're not in the room when those decisions are made. As one researcher described it, we're building a pan-optic healthcare system where your behavior is constantly nudged and influenced by algorithms you can't see, using data you didn't know was being collected for purposes you never agreed to. Here's where I'll probably lose some of the privacy advocates because there is a legitimate benefit to early intervention that we just can't dismiss. Think about target strategy again. They weren't just identifying pregnant customers, they were timing their interventions from moments when people are the most open to change. Pregnancy, moving to a new city, getting married, starting college. Behavioral psychologists call these habit disruption windows. Your routines are already changing, so it's just easier to add new behaviors. In healthcare, the same principle applies. One study looked at home health visits during transition periods, so think like right after hospital discharge, during medication changes, when warning signs first appeared, like my retinal detachment. The results significantly fewer ER visits, fewer hospital admissions, lower cost, and people who were better able to manage their own health. The timing was everything. Not after someone was already in crisis, but before the crisis happened. Same with weight management. Catching someone when they first start skipping the gym or hitting fast food more often, maybe me, much easier to course correct than after they've gained 40 pounds and their metabolism has adapted. Here's the thought experiment. What if Target, instead of sending baby product coupons, had sent information about prenatal care, healthy pregnancy nutrition, or resources for teen parents? Same targeting technology, different intervention. Clear health benefit instead of commercial exploitation. So here's where we are. AI surveillance systems can detect disease risk earlier than we've ever been able to before. We're talking about interventions that could genuinely save lives, reduce suffering, prevent chronic diseases from taking hold. But these same systems create unprecedented opportunities for discrimination, exploitation, and erosion of privacy that we're only beginning to understand right now. The research community is proposing some safeguards, public dashboards showing exactly what data is collected and how it's used, community advisory panels where affected populations have real input into the data, independent oversight bodies that can audit AI systems for bias, that's hard by the way, technical approaches that let AI learn without centralizing all your data in one place, mandatory testing of AI systems against deliberate manipulation attempts before they're deployed. But here's the political question we can't dodge. Do we want a healthcare system where AI watches our every move, potentially catching diseases early, but also creating a permanent digital health record that can be used against us for the rest of our life? Or do we want a system that respects our privacy and autonomy, even if it means missing early interventions that could have prevented real suffering and disease? The target story isn't really about pregnancy prediction. It's about the moment we realize that algorithms can know things about us, intimate, private things, that we haven't shared with anyone, including ourselves. In healthcare, the stakes are infinitely higher. An algorithm that spots your cancer risk six months early could save your life. The same algorithm in the wrong hands could make you uninsurable, unemployable, or subject to discrimination you can't even see happening. Both scenarios are real. Your smartwatch is already capable of detecting cognitive decline months before you notice symptoms. It can predict heart attacks days in advance. It can identify diabetes risk before your blood sugar goes clinical. These aren't future capabilities. They're happening right now, and they're backed by dozens of peer-reviewed studies. But we're also building the surveillance infrastructure that enables this: constant biomonitoring without meaningful consent, data repurposing for non-health uses like the target, informational power that favors corporations over patients, and systems that intensify surveillance on sometimes the most vulnerable populations. The question isn't whether AI can detect disease early, it demonstrately can. The question is whether we can build safeguards strong enough to prevent the same technology from becoming a tool of discrimination and control. Because right now, we're building their surveillance infrastructure first and hoping the safeguards will catch up later. That's not how this usually works out.