Insights
The Segment of One
Why AI Personalization in Healthcare Ends with the Individual, Not the Segment In January 1989, two consultants at The Boston Consulting Group described a future in which companies…
Why AI Personalization in Healthcare Ends with the Individual, Not the Segment
In January 1989, two consultants at The Boston Consulting Group described a future in which companies would stop serving markets and start serving individuals. They called it the segment of one. The technology of the time could not deliver it, so marketing settled for the next best thing: segments, personas, and campaigns aimed at groups of people who resembled each other. Healthcare marketing inherited that compromise and still lives inside it, which is strange, because healthcare is the one industry whose core discipline never accepted it. No physician has ever treated a segment. Thirty-seven years after the idea was named, the technology has arrived, and patients, trained by daily conversations with AI assistants, have arrived with it. This paper examines what individual-level personalization actually consists of in a clinic, where the boundary between relevance and surveillance runs, and why the unit of patient engagement is about to change from the group to the person.
Medicine never had segments. Marketing did.
Consider what happens in every consultation room. A patient describes their history. The physician examines, asks, narrows, and decides. The result is a diagnosis and a treatment plan built for one person, adjusted to their anatomy, their comorbidities, their circumstances, and their goals. The clinical method is individual-level personalization, refined over centuries and applied without exception. No clinic would dream of reusing one patient’s treatment plan for another because the two belong to the same demographic.
Yet the same clinic sends both patients the same newsletter. The communication that surrounds care, the education before a decision, the preparation before a procedure, the follow-up after it, has almost always been produced in batches, because the tools could do nothing else. Healthcare became the most individual of industries at the point of care and one of the most generic in everything it says around it.
For decades this gap was invisible, because every clinic communicated the same way and patients had no other reference point. That reference point now exists, and it sits in their pocket. According to Rock Health’s December 2025 survey of 8,000 American adults, 32 percent had used an AI chatbot for health questions, double the share of a year earlier, and 64 percent of those users engage at least weekly. A KFF poll fielded in early 2026 found the same magnitude independently. OpenAI itself reports that more than 40 million people ask ChatGPT health questions every day.
Inside those numbers sits the detail that should concern clinic leadership most. Among patients who use AI for health, 74 percent use general-purpose tools such as ChatGPT. Five percent use a chatbot built by a healthcare provider. A patient can now ask nuanced, personal questions about their upcoming cataract surgery at eleven at night and receive an answer shaped to exactly what they asked, in the tone they asked it. The next morning, their own clinic sends them the same email it sends everyone else. The comparison is no longer between your clinic and the clinic across town. It is between your communication and the most individually responsive conversation your patient had that week.
A thirty-seven-year-old idea whose technology finally arrived
The segment of one is not a new ambition. Richard Winger and David Edelman coined the term at BCG in January 1989, describing the marriage of two capabilities: a proprietary database of individual preferences and behavior, and a service operation disciplined enough to act on it for each customer. They wrote this before the web existed. The idea was clear; the infrastructure was not.
What followed was thirty years of approximation. Mass marketing gave way to segmentation, segmentation to personas, personas to lifecycle stages. Each step was real progress, and each was a compromise. Segments were never the destination. They were the finest resolution the tools of the database era could afford, and an entire industry mistook the limitation for the method.
The distance between idea and infrastructure closed only recently. In 2024, the same David Edelman, writing with Mark Abraham, published Personalized, the playbook for customer strategy in the age of AI, thirty-five years after his original paper. The timing is not incidental. Three shifts finally made the segment of one operational: patient data can now be unified into one continuous profile instead of living in disconnected systems, generative AI can produce individually adapted content at a marginal cost near zero, and orchestration platforms can act on individual signals in real time rather than on campaign calendars. The ambition of 1989 met its tools in the 2020s.
What it takes to answer one patient
Stripped of vendor language, individual-level personalization, often sold as hyper-personalization, comes down to three capabilities in a clinic, and the absence of any one of them collapses the whole.
The first is a unified, consented signal layer. A patient’s journey generates signals in three places that were never designed to talk to each other: the website where they explore, the inbox and phone where they communicate, and the clinical system where their care is recorded. As long as those remain three silos, the clinic literally cannot see the person, only fragments of them. Individual response requires one first-party profile, built with explicit consent, in which a question asked on the website, a reply to a message, and a completed consultation are events in one story rather than entries in three databases. Whether a clinic’s data foundation is ready for this step is the question we examined in How Ready Is Your Clinic for AI Marketing?.
The second is content that adapts to the person without leaving the clinic’s voice or escaping clinical oversight. This is where generative AI changed the economics. A 2025 comparative study found that preparing a set of patient education materials manually took 14 hours; generative tools produced comparable drafts in under one hour. The evidence on real-world use is equally instructive for its honesty. A study of 75 clinicians at a large New York health system found that AI-drafted patient message responses were used in about one interaction in five and shortened message turnaround times modestly. Adoption is gradual, and it should be: every AI-drafted word in patient communication belongs under human review. The gain is not automation of judgment. It is that individually adapted communication stops being a luxury of time the clinic does not have.
The third is journey logic that responds to signals rather than the calendar. Segment communication answers the question “what month is it.” Individual communication answers the question “what just happened to this person.” A patient who has just read a page about lens options three times is not in the same moment as one who booked a consultation and went silent, even if both sit in the segment “cataract, over 65.” A journey built for the individual notices the difference and responds to it, at the moment it happens, through the channel the patient prefers.
Everything in these three capabilities runs on patient data. Which raises the question every clinic owner should ask before any of it: where is the line?
The relevance patients welcome and the surveillance they punish
Personalization has a documented failure mode, and pretending otherwise would make this paper marketing. Researchers call it the personalization-privacy paradox: people value individually relevant experiences and are simultaneously unsettled by the data collection that produces them. A 2026 study in Psychology and Marketing describes the result plainly as creepiness, and it drives disengagement, not loyalty.
But the research also locates the cause with useful precision. Accenture’s consumer research has reported that when people describe a brand experience as too personal or invasive, roughly two thirds trace it to one thing: the company used information the person never knowingly shared. The discomfort is not triggered by relevance. It is triggered by surveillance, by the sense of being known without having agreed to be known.
For healthcare, this distinction is everything, and it is an advantage. Patients already differentiate whom they trust with their health data; in PwC’s 2025 survey of 4,030 US consumers, they drew clear distinctions between pharmacies, insurers, and fitness companies when asked whom they would allow to use that data for personalization. Trust is not granted to a category. It is granted to a relationship, and few relationships in commercial life carry more earned trust than the one between a patient and their clinic. That trust, as this series has argued before, is built between visits, not during them, which makes the communication around care the place where it compounds or erodes. A clinic that builds its segment of one on data patients knowingly and willingly shared with it, and says so, is not operating near that line at all. It is doing digitally what it has always done clinically: knowing the patient because the patient chose to be known.
That is why first-party consented data is not a compliance detail in this argument. It is the boundary condition that makes the entire model legitimate.
What this means for the clinic
The commercial evidence, with honest caveats, points one direction. BCG’s 2025 Personalization Index finds that personalization leaders grow revenue roughly ten percentage points faster annually than laggards, a pattern this series first examined in Personalization as the New Standard in Healthcare Experience. McKinsey’s synthesis across sectors puts the typical revenue lift at five to fifteen percent. In healthcare specifically, PwC found 28 percent of consumers willing to pay extra for personalized treatment. The caveats belong in the open: most of this survey data is American, and European expectations may trail it by a step; the outcome evidence for AI personalization is younger than the marketing claims built on it; and as the clinician adoption data shows, this is a practice change, not a switch to flip.
The practical consequence for a clinic owner is not a technology shopping list. It is a change in the question used to evaluate any tool, agency, or platform. The old question was “how precisely can it segment our patients.” The new question is “can it respond to this one patient’s signals, at this moment, in our voice, with our clinicians in the loop, on data the patient consented to share.” Any system that cannot answer yes to all five parts is a batch tool with better vocabulary.
The window matters more than the technology. Patient expectations are being reset right now by tools outside healthcare, at the pace of the adoption numbers above. Clinics that move early will define what individually responsive care communication feels like in their market. Clinics that wait will have it defined for them.
Your patients were never a segment
The segment of one spent thirty-seven years as a promise because the tools were not ready. They are ready now, and healthcare is the industry with the least distance to travel, because it never believed in segments where it mattered most. Every clinic already personalizes at the level of the individual in the consultation room. The shift this paper asks for does not require a new belief, only a wider application of an old one: treat the person in every message the way you already treat them in person. Marketing is not learning this from Netflix. It is catching up to medicine.
The clinics that close that gap first will not just communicate better. They will set the standard of feeling known that every other clinic in their market is measured against.
Carely was built for exactly this shift: journey-level personalization on consented first-party data, with the clinic’s voice and clinicians in the loop.
FAQ
What is the segment of one?
The segment of one is a marketing concept coined at The Boston Consulting Group in 1989 by Richard Winger and David Edelman. It describes serving each customer as an individual market, using a database of their preferences and behavior to tailor the service they receive. AI has now made it operational for clinics.
How is individual-level personalization different from segmentation?
Segmentation groups similar people together and sends them the same message on a schedule. Individual-level personalization responds to one patient’s signals, stage, and history at the moment they change. A segment answers “what month is it.” An individual journey answers “what just happened to this person.”
Does AI personalization in healthcare threaten patient privacy?
Research on the personalization-privacy paradox shows that discomfort is driven by data people never knowingly shared, not by relevance itself. Personalization built on consented first-party data, and disclosed openly, avoids the failure mode that makes personalization feel invasive.
What does a clinic need to personalize at the individual level?
Three capabilities: one consented first-party profile spanning the website, communication channels, and clinical systems; AI-generated content that stays in the clinic’s voice under clinical review; and journey logic triggered by patient signals rather than the campaign calendar.
Does personalization pay off commercially?
Personalization leaders grow revenue roughly ten percentage points faster annually than laggards, according to BCG’s 2025 Personalization Index, and 28 percent of consumers told PwC in 2025 they would pay extra for personalized treatment.