Pharma marketing has an AI challenge. It's not the AI.

The technology works. The adoption hasn’t. Here’s where it breaks, where it’s working, and what to do about it.

Walk into almost any pharma marketing meeting this year, read your LinkedIn feed or attend a conference and you’ll hear one word more than any other. AI. It’s in the agency pitch, the vendor demo, the keynote speech, the leadership town hall. The energy is real. The investment is real. And mostly, the work hasn’t changed... at least not yet.

MIT’s 2025 State of AI in Business study looked at more than 300 enterprise deployments and found that roughly 95% of them never delivered measurable business impact. This isn’t a story about bad technology. The models are good and getting better. It’s a story about how we adopt them. After 25 years in pharma, most of it spent building omnichannel platforms and the teams that run them, I’ve watched the pattern "rinse and repeat": the tool arrives with a parade, and the change management arrives never.

Nobody owns the change.

MIT was direct about the cause of that 95%: not model quality, not regulation, but the learning gap, the work of fitting a tool into how people actually do their jobs. The technology showed up. The new way of working didn’t.

In most organizations AI gets bought by procurement, championed by an innovation lead, and handed to brand teams with no change to their process, their incentives, or their day. So people keep working the way they always have and quietly use consumer tools on the side, what MIT calls the shadow AI economy, with no governance and no standard. A tool with no owner is a subscription, not a capability.

The fear nobody is managing.

There’s a human reason adoption stalls, and it rarely makes the project plan: people are scared. Not of the technology, but of what it means for them. An ADP survey of more than 39,000 workers across 36 countries found job insecurity running high even with unemployment low, and AI named as the cause. It isn’t only the front line. In a Writer survey, 61% of executives said they fear losing their job if they don’t lead the transition well.

And the fear isn’t abstract. In May 2026 Meta cut about 8,000 jobs, roughly 10% of its staff, in a restructuring it openly tied to funding its AI push, while moving thousands of remaining employees onto new AI teams. Tech has shed more than 100,000 jobs in 2026 alone. Pharma is cutting at similar scale: Novo Nordisk around 9,000, Takeda about 4,500, Bayer more than 12,000 since 2023. The main drivers there are patent cliffs and pricing pressure, but the stated rationale increasingly includes “efficiency,” “automation,” and “digital enablement.” Employees can read. When the same leadership that preaches efficiency rolls out an AI tool, people hear a threat.

The nuance matters: “AI will take my job” isn’t always the single biggest blocker, since distrust and “it doesn’t work” often rank higher day to day. But the fears that quietly throttle adoption are consistent: losing relevance, falling behind peers, being judged on AI use without ever being trained. The fix isn’t a pep talk. It’s framing and proof: present AI as a way to expand what people can do, then put it in their hands. People embrace a tool they’ve held; they resist one that’s been pointed at them.

We skipped the teaching.

The intent is there. ZS found that more than 60% of pharma and biotech tech leaders see AI literacy as essential, and about 70% planned to invest in training in 2025. But planning to invest and changing how people work are not the same thing. Too often “training” is a one-hour webinar and a recording nobody opens. That builds awareness, not a habit. You don’t learn a tool by hearing about it. You learn it by using it on something real, with help nearby.

And what you teach today is stale tomorrow.

There’s a deeper problem with treating training as an event: the thing you’d train on keeps changing. The tool that set best practice last quarter has three competitors and a new interface this one. And the real movement isn’t in any single model. It’s in the ecosystem around it, the agents, the connectors, the protocols, and the orchestration layers that turn a chatbot into a coworker. A curriculum built on “how to prompt ChatGPT” is obsolete before the deck is approved.

So the goal can’t be to teach the tool. The tool will change. The goal is to build the habit of learning the tool, in the organization and in the person.

It isn’t free, and you can’t tokenmax your way there.

There’s a myth hiding under the buzz: that AI costs less than the people doing the work. A 2025 MIT study found AI was the cheaper option in only about 23% of the tasks it examined; in the other 77%, the cost to buy, customize, and maintain it ran past the human wage. A senior Nvidia executive put it plainly this spring: for his team, compute now costs more than the people.

The trap has a name worth using: tokenmaxing, the belief that more AI activity equals more value. More prompts, more agents, more tokens, as if usage itself were the goal. But most AI is priced by the token, so cost climbs with use, and Goldman Sachs projects that agentic AI could multiply token consumption more than twentyfold by 2030. Volume is not a strategy.

And this is where enablement goes wrong. The plan, stated or not, is to hand everyone a powerful tool and let usage sort itself out. You don’t teach a bird to fly by throwing it off a tree. A fledgling learns in stages: short hops, a lot of flapping, a parent close by, the branch still in reach. Capability comes from supported practice on real work, not a launch email and a license key. Maximize the support, not the tokens.

Where it’s working, and why.

The upside is real, which is what makes the waste sting. Early adopters who get past the pilot stage report meaningful gains, on the order of 20% improvements in marketing effectiveness in some analyses. And some teams are already there. Pfizer, for example, built a generative AI platform, Charlie, with Publicis, aimed squarely at the marketing content supply chain: drafting, fact-checking, and the legal and regulatory review that slows everything down.

What pharma and its partners can do better.

First, pick one real problem and put one person in charge of it. Resist the platform that promises everything. Choose a single workflow that hurts today: MLR cycle time, content versioning, or segmentation. Solve it completely before adding the next. Then measure that owner on outcomes, not on logins, prompts, or a signed contract. The test is simple: is the work better, faster, or cheaper?

Second, lead with the people, not the technology. This is the priority most teams skip, and it decides the rest. Tell people plainly that AI is here to make them better, not fewer, then prove it with hands-on training on their real work and a culture where admitting what you don’t know is safe. Framing and practice are the antidote to fear.

Third, demand partners who stay past the demo. The agencies and vendors worth keeping co-own the integration, the training, and the measurement, and they price honestly, so “more tokens” never gets mistaken for “more value.” If a partner’s involvement ends at go-live, the 95% is where you’re headed.

The honest version

The buzz isn’t the problem, and neither is the technology. We keep treating AI as something to buy instead of something to adopt, and adoption is a people problem wearing a technology costume. Pharma marketing doesn’t need more pilots, more tools, or more tokens. It needs fewer, better bets: owned by someone, taught properly, priced honestly, and honest with the people who are afraid of it. Do that, and you won’t just keep up. You’ll quietly pull ahead while everyone else is still talking.

Aaron Uydess is the founder of Kairos Meridian, an independent consultancy bringing big-agency thinking with boutique agility to pharma marketing, data, and innovation. If your team has bought the tools but not the change, that’s a conversation worth having.

---

Sources

•     MIT NANDA, The GenAI Divide: State of AI in Business 2025 — ~95% of enterprise GenAI pilots fail to deliver measurable impact; sales and marketing hold the most budget and the lowest ROI; the “learning gap” and “shadow AI economy.”

•     MIT CSAIL (2025) — AI was the cheaper option in ~23% of tasks studied; humans cheaper in 77% once build, customization, and maintenance were counted.

•     Fortune / Axios (2026) — Nvidia executive on compute costing more than employees; Goldman Sachs projection of a ~24x rise in token consumption by 2030.

•     Meta layoffs (CNBC / NPR / Bloomberg, May 2026) — ~8,000 jobs (~10% of staff) cut in a restructuring tied to funding AI; ~7,000 employees redirected to AI teams; 2026 capex guided to ~$125–145B. Tech sector ~110,000 layoffs in 2026 (Layoffs.fyi via CNBC).

•     Pharma/biotech layoff trackers (Xtalks, Fierce Biotech, BioSpace, PharmExec, 2025–2026) — Novo Nordisk ~9,000; Takeda ~4,500 (FY2026); Bayer 12,000+ since 2023; primary drivers are patent cliffs, pricing, and restructuring, with “efficiency,” “automation,” and “digital enablement” increasingly cited (e.g., Pfizer’s Seagen cuts).

•     ADP (via Fortune, 2026) — 39,000+ workers across 36 countries; widespread AI-linked job insecurity despite low unemployment.

•     Writer survey (via CIO Dive, 2026) — 61% of executives fear losing their job if they fail to lead the AI transition.

•     ZS, Pharma trends 2025 — >60% of tech leaders see AI literacy as vital; ~70% planned to invest in training in 2025.

•     Pfizer “Charlie” (with Publicis); AstraZeneca ~12,000 AI-certified; Moderna custom assistants — industry reporting.

• ~20% marketing-effectiveness gain — secondary analyses.