A new dimension of personalization -- from 1to1 to 1to-SelfWe are living in an era where we adapt to the ``today's self'' and ``today's present self'' of the same customer. The structure of real-time contextual inference and how to implement it in pharmaceuticals.

01A three-step history of personalization

Marketing personalization has distinct developmental stages. The first stage is mass marketing. In the world of TV ads, newspaper ads, and radio commercials from the 1950s to the 1980s, the premise was to deliver the same message to everyone. The next step is segmentation. Divide people into groups such as male/female, 20s/30s, metropolitan area/region, and deliver a slightly different message to each group. Direct mail in the 1990s and email marketing in the 2000s were basically at this stage.

The third stage is One-to-One Marketing. Presented by Don Peppers and Martha Rogers in their book of the same name in 1993, it created the theoretical foundation for CRM. Their argument is to "break down segments down to the single person," and Amazon's recommendations, Netflix's viewing suggestions, and Spotify's Discover Weekly are the peaks of this third stage. It has the history of each person in the database and returns experiences specific to that person. Until the early 2020s, digital marketing was a competition to refine this 1to1.

As we enter the period 2024-2026, the outline of the fourth stage is starting to emerge. In this series, this 1to-Self is what we call this. Premised on the fact that multiple "selves" within a single customer rotate by time, physical condition, situation, and mood, the idea is to optimize for "me" at that moment.

The last column of the table shows why each stage is swallowed by the next stage. Massing misses individual preferences, segmentation misses variation within the same attribute, and 1to1 misses changes in the same person over time. Each was pushed to the next stage when finer-grained differences proved economically viable.

02Why 1to1 is not enough

The structural weakness of 1to1 isTreat history linked to the same customer ID as an averaged "you" There is a particular thing. Amazon uses my purchase history to make recommendations for you, but those are the average me drawn from a collection of purchases over the past five years, not the me at 7 a.m. today.

The observation that the same person has multiple "selves" has been known for a long time in consumer behavior theory. Presented by Mark Snyder in 1985 self-monitoring theory, discussed by Hazel Markus and Paula Nurius in 1991. possible selves concept. Additionally, neuromarketing studies in the 2010s repeatedly demonstrated that the same subjects' purchasing decisions were statistically significantly altered by fasting/fullness, morning/nighttime, and alertness/fatigue. The problem is this implementation The technology to do so did not exist for a long time.

In the 1to1 era technology stacks, it was standard to maintain one profile per customer and update it over time. The granularity of the profile is "category preference," "price range," and "brand tendency." Differences by time of day, differences in reactions depending on physical condition, and fluctuations in preferences due to context are averaged out as noise and disappear.Messages aimed at the ``average me'' will not reach ``the me at this moment.''。

Translate this in a pharmaceutical context. A medication support app for chronically ill patients sends notifications with the same tone at the same time every morning. The message remains the same whether the patient is feeling well at 7 a.m. or tired from working late into the night. 1to1 can only be personalized at this granularity. On the other hand, if you can use the heart rate variability, sleep score, and activity level that can be read from the patient's Apple Watch as context,"Today's tired patient" An encouraging message for"Today's healthy patient" The tone can be switched moment by moment to present a new exercise goal.

031to-Self concept

Strictly define 1to-Self.A design philosophy that assumes that a single customer has multiple "self-states" that change depending on the time, situation, physical condition, and mood, and implements communication that is optimized for the self-state at that moment.. While 1to1 uses "customer ID" as the unit, 1to-Self uses the combination of "(customer ID, moment)" as the unit.

I will explain the specific differences from three perspectives. Firstly,Profile granularity. In 1to1, there was one customer and one profile, but in 1to-Self, one customer has multiple self-states (self in the morning, self at night, self after exercise, self during stress, etc.), each with sub-profiles. Second,Update frequency. 1to1's profile is updated on a weekly/monthly basis, but 1to-Self's self status changes on a second/minute basis. Thirdly,layers of judgment. 1to1 operates on static rules (I like this category → recommend this category), but 1to-Self relies on dynamic contextual reasoning (what self-state this person is in at this moment).

``In the morning, I read long articles about world events on a news app. During the day, I skim short posts on social media. At night, I buy fiction on a bookstore app. I'm the same person, but I'm not the same person. Marketing has long treated me as an average person.Now, I'm finally trying to face myself in the moment.」── Sherry Turkle (MIT, The Empathy Diaries, 2024)

The core of 1to-Self isTreat customers as a dynamic process rather than an object That's it. It is understood not as a bundle of fixed attributes, but as a series of constantly changing self-states. Note that this is not so much a theoretical development in consumer behavior theory as it is a sequence in which the maturation of implementation technology has forced a conceptual update. The theory has been around for 30 years. Implementation finally caught up in 2026.

04Structure of real-time contextual inference

To implement 1to-Self,“What self-state is this person in right now?” We need a mechanism to instantly infer. In this section, we will dissect the structure of this contextual inference by dividing it into three layers.

Layer 1 -- Signal collection. At the bottom layer, we collect raw signals from multiple sensors. Movement status from the smartphone's gyro sensor, heart rate variability and SpO2 from Apple Watch/Garmin, presence or absence of a previous meeting from the calendar API, and whether you are at home, work, or on the go from location information. The benefit of the generative AI era is that these unstructured signals can be normalized into a form that LLMs can interpret.

Layer 2 -- Self-state inference. Probabilistically infer current self-state from signals. If your heart rate is high and your calendar is empty, you're feeling relaxed. If you're on the move and your heart rate is fast, you're in a hurry. If it's late at night and your screen is displaying a long message, you're concentrating. This inference has been brought down to a granularity where LLM-based lightweight models (Claude Haiku, GPT-4o-mini, Gemini Flash, etc.) can be executed at the edge. This is the technological tipping point of 2024-2026.

Layer 3 ─ Content adaptation. Dynamically change content according to inferred self-states. The same ``medication reminder'' sends a detailed explanatory text to the ``relaxed self,'' a one-line notification to the ``hurried self,'' and only a silent notification to the ``concentrated self.'' The ability of generative AI to instantly rewrite the same information in different tones has made this layer practical.

These three layers are independent of each other; a rich signal in layer 1 is meaningless if the inference in layer 2 is coarse, and even a sophisticated signal in layer 2 cannot be reached if the content in layer 3 is fixed. Designing the three tiers to work seamlessly is a central challenge for 1to-Self implementation.

054 methods of contextual inference using AI

In reality, there are four typical approaches to layer 2 self-state inference. Each relies on different signal sources and is good at detecting its own state.

In actual operations, designs that combine the four methods are becoming mainstream. For example, method 3 (environment) takes the general context of time and place, method 1 (behavior) narrows down the subject of interest, method 2 (organism) checks the physical condition if necessary, and method 4 (dialogue) reads the final intention. Although the accuracy of each component may be incomplete on its own, the inference power of the combination greatly exceeds the sum of the components.

06Patterns of personalization failure

There are some typical failure patterns in 1to-Self implementations where over-optimization can backfire. This section lists three things observed in the field.

Firstly Failure of being too proactive. As the accuracy of contextual inference increases, the temptation to predict and present actions before the user is aware of them becomes stronger. An example of Amazon notifying you at 8 a.m., ``You usually order coffee beans at this time.'' Although it may seem convenient, it can make users feel like they are being watched, causing them to abandon the app in the long run. There is a trade-off between proactive accuracy and the user's sense of autonomy.

secondly “Context mix-up” failure. The inference that heart rate variability is high → relaxing → presenting long content may not be distinguishable from the patient's state of being deeply saddened. Biological signals are only the surface of the state, and their meaning depends on the context. The more accurate the inference is, the more serious miscommunication will occur due to a misunderstanding of the context. In pharmaceutical Patient Support Programs, a safe design is to fall back to neutral messages in situations where the inference is uncertain.

thirdly "Homogenization" failure. Applying 1-to-Self optimization to all users ends up creating a paradox: everyone receives the same optimized experience. For example, if the entire company sends out ``Morning Messages to Me'' all at once during the morning rush hours, users' mornings will be filled with messages. The phenomenon where the superposition of individual optima creates collective overload has been widely observed in the explosion of notifications on SNS. 1to-Self carries the same risks.

07Implications for pharmaceuticals - Delivering to patients' "today's selves"

In pharmaceutical marketing, 1to-Self is most effective in patient support programs for chronic diseases. Diabetes, hypertension, heart failure, rheumatoid arthritis -- these diseases are closely intertwined with patients' lives, and daily compliance with medication and lifestyle management determine long-term prognosis. There are days when the same patient is healthy and able to exercise and manage his diet, and other days when he is unable to do anything due to stress from work and family."Today's patient" Messages received significantly change long-term adherence rates.

Draw three specific scenarios.

Scenario A – App for diabetics. Get HRV and sleep scores from your Apple Watch. In the morning, when sleep is short and stress indicators are high, the system refrains from suggesting exercise and instead sends a short message saying, ``Let's start the day without overdoing it. Just do the basics today.'' On the other hand, in the morning when you have enough sleep due to holidays, it will suggest a new recipe or a 10-minute exercise goal. The same app shows different faces depending on the patient's "today's condition."

The second is Antidepressant medication support. The system infers the patient's level of mental fatigue using contextual reasoning, limits the notifications to short ones with low cognitive load during periods of high fatigue, and encourages self-monitoring diaries during the recovery period. What is important is that the design chooses to do nothing when the inference is uncertain. Patient Support Program designers in the field share the finding that silence is often best.

Scenario C ─ Content for medical professionals. Information that MRs deliver to doctors is also subject to 1to-Self. In-depth clinical data can be provided to doctors during the intensive period before a conference, concise key points can be provided to doctors on a Thursday afternoon when they are exhausted from outpatient work, and case review content can be provided during review time on the weekend. Multiple reports have appeared in Pharma Marketing industry publications in 2025 that content delivery timing tailored to doctors' ``today's self'' increased the open rate of content by 2-3 times.

08Boundary with privacy

The most essential tension with 1to-Self is with respect to privacy. High-resolution contextual inference requires continuous observation of more biological, behavioral, and environmental data. From the perspective of the observer, this Increasing the density of monitoring There is also. How pharmaceutical companies draw this line will have a decisive impact on trust formation over the next five years.

This section presents five questions that pharmaceutical companies should be aware of when implementing 1to-Self. Firstly,Which signal to take. It captures heartbeats but does not capture continuous location information, and retains interaction logs but immediately discards raw data from biosensors. Signal screening is also a privacy design. Second,where to store. Edge AI design that infers on the user device rather than cloud-centralized reduces the risk of data exfiltration. Apple is pushing hard for on-device processing of medical data starting in 2024, and there is scope for pharmaceutical companies to adopt this design as well.

Thirdly,Granularity of consent. Rather than a general consent such as "I consent to the acquisition of medical data," a fine-grained consent design that separates the purpose, data type, and period, such as "I consent to the acquisition of heartbeats for two weeks for medication management," respects the patient's right to self-determination. Article 7 of the GDPR, the revised Privacy Act opt-in requirement, provides legal support in this direction. Fourth,Inference transparency. Explainability that allows patients to understand why they received this message at this time. Trust is supported by a design that allows the patient to see the self-state inferred by the AI (“I have determined that you are currently experiencing high levels of fatigue”).

Fifth,ease of withdrawal. It should be easy to withdraw consent, delete data, and leave the service. The “right to be forgotten” in Article 17 of the GDPR gains weight once again in the 1-to-Self era.1to-Self must be designed to observe the patient and at the same time ensure the patient's right to withdraw from observation.。

094 steps to implement 1to-Self

We offer four steps for pharma companies to start implementing 1to-Self, with a granularity that can be started in the next week. What is important is the principle of not trying to move everything from the beginning, but proving it in the smallest unit and then expanding it.

The priority order of the 4 steps is STEP 4 → 1 → 2 → 3. Privacy governance should precede technology introduction. This is an area that would have to be completely reworked later if the order was reversed. Previous companies that implemented 1to-Self in their Patient Support Programs (Sanofi's Onduo successor, Eli Lilly's digital assistance app) all invested the first 3-6 months in governance design.

10Next time: Connecting to your AI-generated content strategy

This time “Today’s me” among the same customers We dealt with designs that are optimized for individual contexts. Next time (part 5), we will reverse the perspective and look at how brands Generate and distribute large amounts of content Step into the side design.AI-generated content strategy ─ How to mass-produce brand voice。

The fact that 1to-Self requires "messages optimized for a single moment" also means that the total amount of messages will increase explosively. 3 tones x 4-6 self-states x 10-20 products = hundreds to thousands of variations of the same information are required. It is impossible to write this manually and requires mass production using generative AI. Next time, we will deeply dissect the design, quality control, and regulatory compliance methods for mass producing content using AI while preserving the brand's "voice" in the pharmaceutical context. The three elements that we covered in Vibe Marketing (part 3), 1to-Self, and AI-generated content strategy, form the core of AI Marketing.

1to-Self, which we covered in this article, is a good example of how technological advances have updated the concept. The observation that the same person has multiple selves has been around for 30 years, and the limitations of One-to-One Marketing have been known in theory. Even so, until the early 2020s, implementation was stuck at 1to1. The reason is simple: there was no technology to capture the changing self-state from moment to moment. Biometric data from wearables, instant inference using edge AI, and dynamic content writing using LLM -- when these are all gathered between 2024 and 2026, 1to-Self will transform from a theoretical concept to an implementable design concept for the first time.

For pharmaceutical companies, 1-to-Self means both great opportunity and great responsibility. It is possible to improve the long-term compliance rate of Patient Support Programs by 20-30%, and it is possible to increase the open rate of content for medical professionals by 2-3 times. On the other hand, continuous observation of biological data, transparency in contextual inference, and guaranteeing patients' right to self-determination are areas that need to be addressed before technology can take off. Companies that choose to put technology first and governance second will be forced to make major setbacks later on. In the next installment (Part 5), we will delve into how to use AI to mass-produce the "explosion of content" that is the premise of 1to-Self, while preserving the brand's voice.