How AI Is Rewriting the Rules of Precision Pharma Media

For decades, pharmaceutical media planning has run on a simple assumption: You only need historical data to know where to spend. Claims data tells you who’s prescribing. ROI data tells you what’s working. Digital engagement tells you who’s paying attention. But that assumption breaks down completely in rare disease and, increasingly, across precision medicine more broadly. Where patient populations are small, specialty pharmacy dispensing leaves claims data incomplete, and media budgets are too modest to generate meaningful digital signal at scale. 

At Addressable Health, we’ve spent the last several years building our agency around a different assumption: that limited claims data isn’t a planning obstacle, it’s a signal to plan differently. AI has become the foundation of that shift, as the engine that lets us understand rare disease audiences, predict prescribing behavior before it happens, and prove media value in categories where traditional measurement simply doesn’t work. 

I want to share how we’re putting that philosophy into practice: from mining unstructured EHR data for high-value healthcare decision makers that no claims database would ever identify, to building predictive audiences that outperform traditional specialty targeting, to compressing reporting cycles so our clients can move budget while a campaign is still live. This is the new reality of running precision media for pharmaceutical brands that don’t have the luxury of scale. We believe they are just a preview of where all of pharma media is headed. 

AI in planning and strategy 

How we are using AI to improve audience understanding, segmentation, and targeting 

A good portion of our client roster sits in rare disease or what I would call complex treatment pathways, and that space breaks the traditional media playbook. Patient populations are small, so visit volume and script volume are low. Claims data is incomplete or nonexistent because dispensing runs through specialty pharmacy. Digital signal can be directional at best because media budgets are modest relative to primary care categories. 

If we tried to plan and target the way we would for a cardiovascular or diabetes brand, we’d be flying blind. So, across our client base, we’ve built AI into the workflow in two distinct ways: audience understanding and dynamic media targeting. 

Audience understanding: 

• We use AI to process unstructured EHR data, particularly physician visit notes, to fill the gaps where claims and script data simply don’t exist. One of our clients in the rare respiratory space is a good example. There’s no available claims data, and media spend is limited to awareness-stage tactics, so visit notes are how we identify the HCPs actually shaping treatment decisions, track how their perception is shifting over time and surface treatment patterns well before they’d ever show up in a traditional data feed. 

• We’re also piloting AI-driven patient cohort tools that build patient audiences and generate journey insights predictively in addition to the standard, claims-based deterministic model. Today, we’re using this primarily for audience insight and message development. But the real opportunity is extending it into media activation and closed-loop measurement, so we can plan and measure against the same AI-derived cohort end to end. 

Dynamic media targeting: 

• This is where AI earns its keep for rare disease and complex treatment clients. Rather than building target lists from historic prescribing, which in a low-incidence category often simply doesn’t exist, we use predictive models to build audiences based on likelihood to prescribe going forward. The model isn’t looking backward at who has written scripts. It’s forecasting who will. 

• We can also refresh those targets as updated data becomes available, instead of the manual list pull most teams schedule once or twice a year. Today, because one of the underlying data feeds runs on roughly a three-month lag, that refresh is closer to quarterly than real-time. As data latency shortens, so can the refresh cadence. When an HCP’s behavior shifts, such as if they start writing more or a new referral pattern emerges, we realign media spend against that momentum on the next refresh rather than waiting for the next planning cycle. In categories where the addressable HCP universe might be a few thousand physicians, that responsiveness is a genuine competitive advantage. 

Using AI to create more human experiences 

How predictive audiences change what an HCP or patient actually sees 

There’s a fair worry in our industry that AI makes marketing colder. In our experience it does the opposite, as long as the model is doing the narrowing and people are doing the messaging. When a target list is built from a specialty code, every physician in that specialty gets the same launch message whether or not the therapy is relevant to a single patient in their practice. When the list is built from likelihood to prescribe, the audience is smaller and the message can be more specific, because we know more about why each HCP is there. 

That changes the creative brief. Instead of one message for cardiology, we can build for the physician who’s seeing early referrals in a rare pathway, the one whose writing is starting to grow and the one whose visit notes show they’re evaluating alternatives. Three audiences, three messages, one campaign. On the patient side, we’re earlier in that journey. Today, our predictive cohort work informs audience insight and message development, so the creative reflects where someone actually is in a long and often frustrating diagnostic path. Carrying that same cohort into media activation is the next step, not something we claim yet. 

The guardrail is simple, and we hold to it. AI decides who is most likely to need to hear from a brand; our strategists and our clients’ medical and regulatory teams decide what they hear. Automation that tries to do both is where trust breaks, and in pharma, trust is the whole game. 

Measurement, optimization and proving value 

How AI-driven insights and optimizations have improved media performance, engagement, and business outcomes 

We see strong results utilizing AI-enabled predictive identification models. Instead of buying media against a broad specialty, such as cardiology, we buy against an audience the model has identified as most likely to start or grow prescribing in the near term. We’ve run this head-to-head against traditional specialty-based targeting, and the predictive segments outperform on audience quality, meaning our clients’ media dollars are reaching the physicians actually positioned to act, not just physicians who share a specialty code. 

We’ve also used AI to fundamentally compress our data ingestion and reporting cycle. AI-assisted data cleaning and anomaly detection mean we’re identifying issues, as well as opportunities, in near real time rather than at the end of a reporting period. That’s allowed us to optimize campaigns while they’re still in flight, moving dollars from underperforming tactics before the budget is spent rather than after. For our smaller pharma clients, where every media dollar has to work hard, catching wasted spend in week 3 instead of week 12 isn’t a marginal improvement. It’s often the difference between hitting launch targets and missing them. 

None of that works, though, unless measurement is designed in before the first dollar is spent. Dashboard reporting and measuring ROI are very different things, and in precision categories the second one is hard. So, we build it into the plan from the start, and it looks like this: 

• Define what media ROI means for the brand before channel selection. NBRx, TRx, infusion enrollments, facility-level group sales. Pick the number the business actually runs on. 
• Choose the measurement methodology up front— holdout, test versus control, or marketing mix modeling— and select channels that can support it. In most specialty categories, the large majority of the HCP universe is reachable through channels that return engagement tied to individual HCPs, so we push the bulk of budget and impressions into those channels on purpose. 
• Contract for the data. Vendor agreements have to cover the timing, quality, and depth of NPI-level engagement data, or the model has nothing to learn from. 
• Traffic with discipline. A strict tagging taxonomy and QC across every channel is unglamorous, and it’s also the difference between a clean read and an argument. 
• Be transparent about the methodology. Every model has flaws. The analyst reading the results has to understand the brand, the strategy, and the channel mix in the context of the therapeutic area, or the recommendation isn’t worth much. 

WHAT TO ASK YOUR MEDIA PARTNER

If you lead a brand in rare disease or another precision category and you’re evaluating how your media is planned, these are the questions I’d want answered: 

• When claims data is thin or missing, what does your audience actually get built from? 
• How often does the target list refresh, and what triggers it? 
• How was ROI defined for this brand, and which methodology is measuring it? 
• Which channels in the plan return HCP-level engagement data, and what share of budget sits there? 
• What did you move, and when, the last time a campaign underperformed in flight? 

The answers tell you quickly whether AI is a slide in the deck or the thing running the plan. 

Precision medicine didn’t just create smaller audiences. It created a set of brands that can’t afford waste, can’t wait for a quarterly read, and can’t plan off data that doesn’t exist. AI is how we’ve made media work for them anyway. Our view is that the rest of pharma media gets there too, just later.

Lea Wester is Managing Director at Addressable Health, a media agency exclusively focused on precision pharmaceutical marketing. She leads the agency’s work applying AI to solve one of pharma’s toughest challenges: reaching the right healthcare providers and patients in categories, like rare disease, where traditional data sources fall short. Her work spans predictive targeting, AI-driven audience insights, and in-flight campaign optimization.