An oncologist asks a rep a narrow question: what do patients actually deal with in the first two weeks on therapy? By the time that question reaches the brand team, it has been filed as “interested in the brand.” The next email carries more clinical evidence. The question is never answered and the brand has spent another touchpoint proving it wasn’t listening.
Pharma has spent a decade building segmentation, personas, next best action and content engines. Very little of that investment answers the question a brand marketer actually faces on a Monday morning: we just learned something new about this HCP, so what do we do differently?
The failure is structural, not analytical. The insight sits in an analytics platform. The persona lives in a slide. The agency is working from a brief written six weeks ago. The field team knows something none of them do. Nobody in that chain is wrong; nothing in it connects. So the marketer becomes the integration layer with several meetings and an impressive number of open tabs.
Attention is already saturated. A ZS study of more than 25,000 physicians found HCPs averaged nine face-to-face touchpoints with pharma representatives every day, before counting the digital channels competing for the same attention.
Volume is not understanding. Research from Reuters Events Pharma and ZS found that only 10%–15% of launch insights were specific to an individual customer, and 57% cited a lack of differentiated insight into customer needs and preferences as a barrier to personalizing at launch. We are reaching HCPs more often and knowing them no better.
Every mislabelled question costs twice: the budget spent on the wrong asset, and the credibility spent on the HCP who asked.
At Lynx Analytics we built this end-to-end for a brand launching in a major oncology disease area: data to attributes to personas to content to response, each stage feeding the decision in the next. The lesson was not that the technology is hard. It is that personalization breaks at the handoffs — there are four — and each one breaks in a predictable way.
Engagement history, stated preferences, practice characteristics, field observations: it all counts as data, none of it counts equally. Webinar attendance tells you an HCP was in the room. A specific follow-up question tells you what they came for. Treat the two as equivalent inputs and you end up optimizing for attendance.
The discipline is to work backwards. Start from the decision the marketer has to make this week, then ask which permitted data actually informs it.
“Asked about the study population” is an observation. “Is unconvinced by the clinical results” is a theory about that observation. Both end up in the profile, and within a few weeks nobody can tell which was which.
So keep the source and the date attached to every attribute. It is unglamorous, and it is the only thing standing between a tentative read of one conversation and a permanent label on a physician.
A persona earns its place by changing what the agency is asked to make: which question we are answering, which evidence is relevant, which assumption we are testing. If the brief would have read the same way without it, the persona is a slide, not an input.
GenAI makes personas far easier to interrogate: you can hold a conversation with one. That is genuinely useful and genuinely risky. A simulated HCP response is a hypothesis, not evidence. AI can draft the hypothesis; deciding whether it holds is still a human call.
The HCP asking about study population and the one asking about treatment logistics need different material, drawn from approved assets, with adaptations routed through the review they require. That much is operational.
The harder test is whether anyone can reconstruct the reasoning afterwards: this signal indicated this need, which led to this asset, in this channel, in this order. If that trail goes cold, you are not personalizing. You are varying.
Generating content faster is useful. Generating it from an outdated understanding of the customer just gets you to the wrong answer sooner.
That is the real case for agentic workflows, and it is not a productivity case. The value is not that the system drafts quickly; it is that the system notices, surfacing a new signal, proposing an update to the profile, recommending a content change for the marketer to approve or reject. Say an HCP who has only ever asked about efficacy starts asking about dosing logistics. The system flags the shift, proposes an attribute update with its source attached, and suggests the approved asset that answers the new question.
Which makes personalization a set of recurring questions rather than a quarterly exercise: What changed? How good is the evidence? Should we act? What did the response teach us?
Sometimes the honest answer is a different message. Sometimes it is to wait. An email open is feedback, but it does not establish that anyone’s question was answered.
The oncology launch showed the connected workflow can be built. The next step is making it live inside the marketer’s week rather than alongside it, with evidence and human judgment inside the loop instead of bolted on afterwards.
Go back to the oncologist’s question. In a connected workflow, it is recorded as what it is: an observation, with a source and a date. It changes the brief. And the next touchpoint answers it, with approved material on what patients actually deal with in those first two weeks, instead of another round of clinical evidence.
For a brand team, this does not start with a platform. It starts with one recurring customer need. Follow it end to end: signal, attribute, persona, content, response. And mark every point where the information thinned out, the work stalled, or the engagement missed the need. Those marks are the roadmap.