EP36: From Hype to Everyday Impact: The Real AI Revolution in Pharma

From an AI Winter to Medical Communications

James's route into pharma AI is an unlikely one — he studied computer science in the early 2000s, at a time he describes as a genuine "AI winter," before moving into web development and eventually medical communications, or "medcomms." Camino Communications positions itself as a medcomms agency with a technology slant, applying AI and digital tools to content that has traditionally lagged behind in the highly regulated pharma sector — slide decks, presentations, emails, and websites — without compromising the scientific rigour required.

Why "AI" Means Different Things to Different Teams

James is candid about a recurring confusion in his client conversations: when marketing and medical communications leads say "AI," they usually mean generative AI specifically — unaware that their own organisation may have had dedicated machine learning and data science teams working on drug research for years. He shares a memorable anecdote about training a pharma company's communications team, only to discover mid-project that the same company had an entirely separate, established AI team — whose lead data scientist had mixed feelings about ChatGPT suddenly making "AI" fashionable, after years of doing genuine technical work in relative obscurity.

Solving Real Problems: The Conference Abstract Challenge

Rather than chasing abstract transformation goals, James's team focuses on concrete, everyday pain points. A standout example: major scientific congresses can publish up to 6,000 abstracts just weeks before the event, leaving pharma teams scrambling to identify which are genuinely relevant before meeting customers and competitors on the ground. Camino built a system using large language models to extract key details from each abstract — cancer type, trial phase, and sponsoring company — then rank them by relevance, compressing a task that previously consumed dozens of staff hours into a rapid, prioritised shortlist delivered within the first days of a two-week window.

Efficiency, Value, and Innovation — Not Just Cost-Cutting

James pushes back firmly on framing AI purely as a cost-cutting tool, describing frustrating conversations with procurement teams who assume AI simply means "the same work for half the price." Instead, he frames genuine AI value across three dimensions: efficiency (doing existing work faster), value (using freed-up time to do more valuable work, like prepping field teams ahead of a conference), and innovation — while cautioning that innovation for its own sake, without practical application, often amounts to wasted effort.

Why Vague Efficiency Targets Set Projects Up to Fail

One of the most pointed insights James shares concerns unrealistic ROI expectations. He describes a client told by a consultant that AI would deliver 40% efficiency gains — a target he considers essentially unmeasurable in practice, contrasting it with narrowly scoped, genuinely measurable projects like the conference abstract tool, where clear before-and-after metrics could be defined from the outset. Vague, sweeping efficiency targets, he argues, are a major contributor to the high failure rates often cited for AI pilots — not because the underlying work lacks value, but because success was never properly defined or measurable to begin with.

Building Confidence Through Hands-On Experimentation

To combat overwhelm, James runs hands-on workshops encouraging teams to experiment directly with real challenges they've previously struggled to solve with AI — rather than starting from abstract, hypothetical use cases. He notes that even personal, non-work use cases — like using ChatGPT to plan family meals — genuinely build the confidence that later transfers into more professional applications, describing this incremental confidence-building as central to sustainable AI adoption.

Real Applications: From Internal Training to Congress Booths

James shares two detailed examples of AI applied to genuinely complex communication challenges:

• AI-powered internal training — replacing traditional, disengaging e-learning modules and end-of-module quizzes with conversational chatbot avatars, significantly increasing genuine engagement with dense medical training content.
• A conversational AI avatar for a rare disease exhibition booth — allowing natural-language questions about a complex disease area, rather than rigid dropdown menus, while using pre-recorded, medically-approved responses to satisfy strict regulatory sign-off requirements. Notably, using an AI avatar allowed the content to be revised five separate times during the approval process — something that would have made a live-actor video approach practically and financially unworkable.

Building Safely on Foundation Models

James explains how Camino layers foundation models like ChatGPT with domain-specific context — shared team projects pre-loaded with relevant clinical papers — alongside clear internal data security policies, including a strict rule against uploading commercially sensitive client information. He notes that pharma teams are often unaware they already have secure, sanctioned internal AI tools available via IT, and that simply clarifying this "permission" unlocks significant use that fear alone had been suppressing.

The Accuracy Paradox: Why Better Models Create New Risks

In a particularly striking insight, James describes how AI accuracy improvements have created a subtler verification risk. Early models, around 50% accurate at extracting scientific content, forced rigorous manual checking by necessity. As accuracy has climbed to roughly 95%, the risk has shifted toward complacency — content "looking right" often enough that teams become less vigilant about catching the remaining errors, requiring active effort to maintain proper verification discipline even as the underlying technology improves.

Regulation Moving Slower Than the Technology

James highlights the pace mismatch between AI development and pharma regulation, noting that the UK's ABPI Code — the governing framework pharma companies follow — only added its first mention of QR codes this year. He suggests explicit AI-specific guidance within that framework may still be decades away, pointing to a broader opportunity for industry collaboration to help shape sensible principles ahead of formal regulatory catch-up.

"AI Is Normal Technology"

Reflecting on what excites him most, James expresses more caution than enthusiasm about flashy new tools like AI video generation, given pharma's conservative, highly regulated nature — and instead champions the unglamorous, incremental value of AI quietly saving people 10 or 20 minutes at a time. He favours framing AI as "normal technology" — not a dramatic, transformative force, but a practical tool that, broadly adopted, leads to happier, more productive working lives.