You can't AI your way out of a broken operating model.

Pharma is spending billions on AI whilst leaving the underlying problems untouched. This issue covers where the industry is actually making progress, where it's wasting money, and the four moves that separate leaders who'll get real value from those who'll just have expensive tools gathering dust.

The most honest thing I've heard about AI this year

I was presenting at a senior leadership offsite last quarter, and a Global VP pulled me aside afterwards. He'd just sat through two days of AI roadmap presentations from his own team.

You can't AI your way out of an inherently poor operating model.— Global VP, top-10 pharma

He wasn't being cynical. He was exhausted. His company had spent the equivalent of eight headcount on AI tooling in twelve months. The operating model was identical to what it had been before. Just more expensive.

And when I look across the work we do with leadership teams, I see the same thing playing out everywhere.

Here's what's actually happening

Every conference has three AI panels. Every leadership team has someone whose job title now includes 'digital' or 'innovation'. Every company has a partnership announcement with a tech giant.

And meanwhile, the same teams are running advisory boards that take three to six months from first email to final output. Sitting in meetings where nobody's sure who owns the decision. Asking field teams to fill in CRM systems no one reads.

The tools are new. The problems are old. This issue is about the gap between those two things, and what the leaders actually making progress look like.

The advisory board that took five months

Picture this. Your medical team wants to run a KOL advisory board. Good idea. You need external scientific input before a launch. Standard process.

Six weeks getting legal to sign off. Four weeks on contracts with the KOLs. Three weeks finding a date that works for everyone. Then the actual meeting, two hours, during which one KOL talks for 45 minutes, one checks their phone, and the others say variants of what you already knew. Then six weeks to write up the outputs in a format that satisfies compliance.

Industry guidance puts the minimum planning time for a pharma advisory board at eight to twelve weeks, with the full end-to-end process typically running three to six months. That's the baseline. Most teams don't beat it.

In hundreds of VP coaching conversations, this is one of the processes that comes up again and again. Everyone knows it's broken. Nobody knows how to fix it without blowing up the compliance requirements that sit around it.

A start-up called SynthioLabs (Y Combinator, 2024) is working on exactly this. They've built a platform that creates AI simulations of healthcare professional archetypes, trained on over 200 million medical publications and 400,000 clinical trials, that can run virtual advisory sessions. The idea is to generate and stress-test hypotheses first, then bring actual KOLs in to validate the most commercially relevant ones.

You're not replacing the human advisors. You're changing where they sit in the process. From generation to validation. It only works if you have clear questions to begin with. If you don't know what you're trying to learn, the simulation just helps you generate bad insights faster.

The four levels of AI maturity in pharma

Across our work with pharmaceutical leadership teams, we've watched dozens of companies go through some version of an AI transformation. What we've noticed is that there are four distinct levels. And most companies are stuck between levels one and two whilst pretending to be at level three. Here's how to tell which one you're actually at.

This is where most people are, even if they won't admit it. The team uses AI primarily to draft emails, summarise meeting notes, and rephrase PowerPoint bullets. This is fine. It's where everyone starts. But it's not a strategy.

You can tell you're at Level 1 when the main AI use cases in your governance meeting are about communication and admin. When the tools haven't changed how a single decision gets made. Level 1 is using a Ferrari to drive to Tesco.

This is where the danger is. Level 2 looks impressive but doesn't deliver. At Level 2, you have internal AI tools announced with fanfare. Partnerships with technology companies. Slide decks with diagrams showing your AI roadmap. Training sessions. People know what the tool is called.

But if you ask someone to name a decision in the last six months that was made differently because of AI, they struggle. The tools sit alongside the existing operating model rather than changing it. You've added cost and complexity without removing anything.

Novo Nordisk built something called NovoScribe, an AI documentation platform using Anthropic's Claude and Amazon Bedrock, that generates clinical study reports. The verified result: what previously took twelve weeks now takes ten minutes. That's not a demo. That's a process turned upside down. Most pharma companies have announced AI initiatives. Far fewer have a NovoScribe-style result to point to.

This is where things get genuinely interesting. At Level 3, you've stopped trying to AI everything and started identifying the specific bottlenecks in your operating model where AI creates disproportionate value. The key word is surgical.

Lilly didn't build a generic AI strategy. They identified a specific problem: patients who needed GLP-1 therapies were being diverted to compounded alternatives, and pharmacy friction was part of the reason. So they built LillyDirect, a direct-to-patient platform combining telehealth, electronic prescribing, and home delivery.

By Q4 2025, LillyDirect accounted for nearly 50% of new Zepbound starts, up from a near-zero base when it launched in August 2024. Zepbound generated $4.3 billion in Q4 2025 alone, part of a tirzepatide franchise running at over $10 billion per quarter when combined with Mounjaro.

The AI infrastructure matters. But the thinking behind it matters more. Lilly identified one friction point and removed it. The rest followed. Level 3 leaders ask one question: where in our process does the current way of doing things create the most delay, cost, or guesswork? Then they ask whether AI can help with that specific thing. Not everything. That specific thing.

Very few pharma organisations are here yet. But a handful of smaller companies are showing what it looks like. At Level 4, the business model itself has changed. These companies didn't have the budget to deploy large field forces. So they had to think harder about where to deploy attention. AI wasn't supplementing a traditional model. It was enabling a different one entirely.

Take Verona Pharma, whose COPD drug Ohtuvayre became the first approved dual PDE3/PDE4 inhibitor. Merck acquired the company for approximately $10 billion in October 2025. Or Insmed, whose brensocatib became the first approved treatment for non-cystic fibrosis bronchiectasis, generating $144 million in its first full commercial quarter with 2026 guidance of at least $1 billion.

These companies had to punch well above their weight against organisations a hundred times their size. That constraint forced clarity about exactly where to deploy attention. It turns out clarity is worth more than headcount. Level 4 asks: what business are we actually in? And does AI change the answer?

What this means for you, practically

You probably can't redesign your entire operating model this quarter. But here are three things you can do right now.

First: find your five-month process

Every pharma team has one. A process that takes far too long, produces mediocre output, and everyone tolerates because no one knows how to do it differently. Identify yours. Ask whether AI can flip the generation-validation sequence, the way SynthioLabs is attempting with advisory boards.

Second: run an honest Level audit

Get your team in a room. Ask them to name one decision in the last six months that was made differently because of AI. Not faster email. An actual decision. If no one can name one, you're at Level 1 or 2. That's okay. But call it what it is.

Third: borrow from peers, not from demos

The most useful AI learning isn't in conference keynotes or vendor presentations. It's in informal conversations with peers who've actually deployed something. The use cases that matter are the ones colleagues describe over dinner, not the ones that appear in case studies six months later.

The turning point

A cross-functional leadership team at a mid-sized biotech we've been consulting for had been running country-level brand planning the same way for years. Each country submitted a brand plan, commercial collected them, and the global team spent weeks trying to synthesise across markets.

They were preparing to launch in twelve countries simultaneously. Getting meaningful insight across all twelve in time to inform launch decisions would have taken the better part of a year.

This was their different approach: they sent a request to each country for existing brand plans and franchise plans. Three to four hundred slides in total, from markets the team didn't know deeply. They used AI to analyse patient journeys across each market, identifying where their programme filled a genuine clinical gap in each country context.

When it went back to the country leads for validation, the feedback was consistently: yes, this is right. Not because AI is magic. Because the questions were clear, the data was real, and the humans were used to validate rather than generate. Launch planning started several months earlier than it would have. That time mattered.

Collaborator's Corner

Try this this week: if your organisation has an enterprise AI tool, Copilot, Claude, or Gemini, run this experiment. Pull the last six months of email communication with one key internal stakeholder and ask the AI to summarise their main concerns, priorities, and likely objections. Then compare it to what you'd have said unprompted. Most leaders are surprised. The gap tells you something useful about where your perception of a key relationship might be out of date.

This week's action

Name your five-month process. Every team has one. The process that takes far too long, produces output everyone tolerates rather than values, and hasn't changed in years because 'it's just how we do it.' Write it down. Map the steps. Ask where humans are doing generation work that could be done by AI, leaving the humans to do validation instead. You don't need an AI strategy to start this. You need one honest conversation with your team about where your time is actually going.

P.S. Working out where AI can genuinely move the needle is exactly what we focus on in our leadership coaching. Not strategy in the abstract but the specific decisions and processes where getting clarity creates real value. If you're heading into a new planning cycle and want a sharper view of where your team's time is going, drop a note to ben@thepharmacoach.co.