Access Is the Product: How AI Agents Are Changing Pharma Data Infrastructure

This article is part of solli’s  September series exploring The Effective Use of AI in Pharma Media.

Pharma media data rarely lives where it’s created. It moves into an agency’s planning system, a client’s warehouse, a measurement environment, a reporting deck, which is the defining trait of a data business: Unlike a platform that owns inventory, a data platform’s value depends on how easily its capabilities travel. 

For years, that delivery required a person navigating a platform, submitting a request, waiting on a technical team to translate a question into a query. Agentic AI is changing that: An end-user describes what they need inside the system they already work in, while an AI agent translates the request into governed actions across the stack, from pulling code sets and building audiences, to resolving identity or checking measurement. 

As agents act on someone’s behalf, infrastructure has to do more than store information. It has to make information understandable, governable, and usable wherever the work happens, starting with one of the hardest parts of pharma marketing: turning a business question into a clinically meaningful audience. 

Making complex data usable 

A marketer might describe an audience simply: patients with a given condition who’ve progressed to a specific line of therapy, excluding anyone only screened or evaluated. Underneath sits a harder clinical question, like which signals confirm the condition, mark progression, indicate active therapy, or need excluding screening or history rather than disease.  

Translating that into a usable, de-identified, HIPAA –compliant modeled audience typically requires someone fluent in clinical code sets, a valuable expertise that sits between the marketer’s language and the data’s structure, turning what should be iterative into a chain of handoffs. 

An AI agent can close that gap, translating a request into a candidate audience with the logic behind key inclusions and exclusions, giving marketers and end–users enough context to review and approve without becoming data experts. The core value here is bringing clinical sophistication into the workflow without requiring deep expertise in data analysis. 

Once an audience is defined, identity is the next test, with patients and consumers represented through de-identified, privacy-first methods including tokenization and eventually connected to exposure data so impressions and clicks can be evaluated against the intended population.  

Coverage can vary at every step, which makes a single summary number an incomplete measure of quality. The more useful question is where the gap occurred: incomplete underlying data, records that couldn’t be represented in the process, an audience definition that’s too restrictive or inconsistencies across systems. With that insight, marketers and agents can act immediately.  

Measurement on demand 

The same issue shows up in measurement. The industry isn’t short on outcomes data; it’s short on efficient ways to turn that data into answers. Outcomes-based analysis tends to become its own project: Define the question, submit it, wait, review, and decide what to do next. When that loop takes real time, teams default to whatever is easiest to pull, like impressions and clicks, until ease of measurement starts determining what gets measured.  

A more useful model makes measurement accessible throughout the campaign. If a marketer can ask about an audience, channel, cohort or time period and get an answer without opening a separate project, then those insights can inform in-flight decisions, shifting spend, refining audiences, adjusting channel strategy or changing activation while the campaign is still running.  

In relying on in-flight optimization rather than waiting for a post-campaign report, it’s possible to improve campaign effectiveness tremendously. An agent that can run those analyses on demand makes that feedback loop practical, giving marketers a way to optimize against outcomes rather than relying on non-industry-specific metrics. 

That ease raises the bar on rigor, rather than lowering it. More questions asked means more chances to use the wrong population or an incomplete definition, so every result needs to carry the population, exclusions and parameters behind it; a fast answer that can’t be traced back isn’t a better one. 

Infrastructure and governance that travel with the data 

Moreover, data should be designed around more than one destination. Clients have spent years building around APIs, scheduled deliveries, and established measurement, all investments that shouldn’t be rebuilt, only extended.  

The same capabilities need to be reachable through several channels at once: an API serving an existing workflow, a clean room for parties that can’t exchange raw records, a client’s own AI agent calling approved services directly, a provider’s first-party surface for clients who want it. These can’t be separate products with separate logic. They have to be different doors into the same governed capability, especially as organizations connect their own agents to external providers. The goal isn’t to eliminate platforms, but to make them less of a prerequisite for reaching what’s inside them. 

That only works if governance doesn’t depend on the interface. Minimum cell sizes, suppression thresholds, entitlement scopes, and permitted joins need to apply the same way whether a request comes through a human-facing application, an API, or a client-side agent.  

A control that exists only because a button is disabled in one interface isn’t a control at the data layer— it controls whoever happens to use that interface. That matters more as workflows automate: Rules need to be enforced by the infrastructure itself, so an agent operates within the same permissions as any authorized caller. Access can get easier without governance becoming optional.  

What this looks like in practice 

The shift toward agentic workflows doesn’t mean replacing the platforms and integrations pharma marketers already use. It means making the capabilities within those systems accessible in new ways.  

An agent can translate a marketer’s question into the underlying data operations, reducing the technical work required to define an audience, investigate identity or measure outcomes. For non-technical users, that can turn a process that once required multiple handoffs and days of work into something that happens within the workflow itself. 

The practical benefit is more than speed. When marketers can refine audiences, identify issues and evaluate performance while a campaign is still running, they have more opportunities to optimize rather than waiting for a post-campaign analysis. PurpleLab is one example of this approach: Our API-first infrastructure supports existing integrations alongside MCP (Model Context Protocol)–enabled agents, a first-party agentic experience and clean-room environments.  

The broader lesson extends beyond any one provider: Data infrastructure needs to be designed so its capabilities can be accessed, governed and put to work wherever the next question comes from. 

Access changes the definition of a data product 

The industry’s next phase won’t be defined by bigger datasets or dashboards, but by how well data and intelligence move into the workflows where decisions get made. For pharma marketers, that means defining audiences without losing clinical nuance, understanding identity performance, asking sharper measurement questions and doing it all within a governed framework. 

The technology will keep evolving, but the requirement is simple: Data has to be built for access, not just storage. When data, logic, governance and documentation are structured to work across multiple channels, the interface can change without forcing the infrastructure to change with it.  

The product is no longer just the dataset, platform or API— it’s the capability to deliver trusted intelligence wherever the next question comes from. Access isn’t just a feature of the product; increasingly, it is the product.

For more on The Effective Use of AI in Pharma Media, click here.

This piece was written by Colin Jacobsen, VP of Product, PurpleLab