By Trevor Lambert, Senior Strategist and AI Specialist at Kanga Health Ltd
Vienna in November is always special, a bit cinematic even, but this year’s CX & AI centred NEXT Pharma Summit felt different. Standing in the cold night air outside the venue, the evening before it started, with my colleague Paul Allen representing Team Kanga Health Ltd, we talked about what we hoped to see.
Pharma has been awash in AI buzz for two years. I wanted to know are we still just talking about the potential, or are we finally seeing the practice?
After two days of panels, debates and use cases, I think I have my answer. We are edging out of the “wow” phase into the “how” phase. That’s good news. But that shift also reveals something awkward: a lot of our current engagement models are failing before AI even enters the room.
Below are my key takeaways, favourite bits, and a few simple frameworks you can steal for your own work. Session and speaker references are from the published agenda.
1. The HCP test: we are still failing the basics
If you want one unignorable truth from Vienna, it’s this: before we talk about AI, we have to fix the fundamentals of HCP engagement.
The opening panel on day two, “The HCP Test: Do Pharma’s Digital Solutions Actually Address Unmet Needs, and How to Fix Them?”, was one of the most pragmatic conversations of the summit. It featured Paulo Amaral (GSK), Dr Mercede Pannozzo (Viatris), and Dr Thomas Maggs (NHS Paediatric Doctor).
The consensus was blunt. Clinicians are drowning in noise. We keep mistaking volume for value. Many HCPs are receiving upwards of 50 brand messages a day, while screen time outside the US is falling. That’s a pretty good diagnosis of why “more digital” hasn’t automatically meant “better engagement”.
Paulo hit the nail on the head with a line that should be printed on every marketing brief in 2026:
“Louder is not better. You cannot interrupt HCPs into submission.”
That simple line captures a truth we often dodge. You can’t force attention. You can only earn it.
Dr Maggs added a critical layer:
“People’s view of normal is so elevated we have to work hard to exceed expectations.”
He’s right. The bar isn’t just “be compliant” anymore. It’s “be as useful as the consumer tech they trust in daily life”.
What struck me is how little of this is a channel problem. We keep “adding touchpoints” when the real issue is whether we are saying anything worth hearing. The sessions hammered home that successful engagement isn’t about more channels. It’s about better content and better interaction.
A simple fix framework
If your team is trying to improve HCP engagement, start here:
- Honesty and transparency – Clinicians want the risks, limitations and trade-offs, not just the upside. If you’re honest about what’s hard or uncertain, you gain trust.
- Peer delivery – HCPs trust other clinicians, not marketers or salespeople. We need more peer-to-peer formats, fewer brand monologues.
- Two-way conversations Small groups beat broadcast webinars. Not because they are trendy, but because there is pressure to engage and space to ask awkward questions.
Until those three things are in place, AI is just a faster way to produce noise.
2. The disappointment: the zombie MIT statistic
Now for the bit that really irritated me.
Across multiple sessions and hallway chats, I kept hearing the same claim thrown around as if it were gospel: “95% of AI projects fail (MIT).” It’s the corporate equivalent of an urban myth shared to put a shiver up people’s spines.
The problem is that the statistic is not reliable. Paul Roetzer has dismantled it expertly on his podcast The Artificial Intelligence Show, pointing out the methodological flaws and the way the number gets inflated in repetition. Roetzer was not at NEXTCXAI. Nobody was quoting his critique. They were quoting the headline.
Why does it matter? Because bad stats drive bad behaviour. If leaders think failure is almost guaranteed, they either panic-buy tools hoping for a miracle, or they freeze and do nothing. Both are wrong responses.
The real story is subtler and frankly more useful: failure is common when teams chase tech instead of value, and when they ignore the people and process work. That, at least, Vienna handled well.
3. Operationalising AI: Shaima Abid’s five-pillar blueprint
Perhaps the most useful “how-to” session I saw was Shaima Abid’s “From Pilot to Practice: Operationalising AI for Scalable Pharma Engagement.” She offered a checklist so clean you could drop it onto a slide for your next steering group.
Shaima’s five success pillars:
- Strategic clarity – Know precisely why you are doing it.
- Data readiness – Without clean, governed data from the start you have nothing.
- Workflow integration – AI shouldn’t be a bolt-on. It must fit the actual flow of work.
- Change enablement – If you don’t bring people with you, you don’t have adoption. You have a pilot cemetery.
- Measurement and iteration – If you can’t measure value, you can’t scale value.
What I liked most was that Shaima didn’t pretend this was complicated. It’s boring, disciplined work. Which is exactly why most organisations avoid it, then act surprised when pilots stall.
The biggest blockers are rarely technical. They are organisational: unclear objectives, weak ownership, poor leadership alignment, and shallow change management.
4. The “iceberg” reality: why adoption is a 70% people problem
While Shaima gave us the operational checklist, Maria Chiara Sbarra from Zambon gave us the reality check on why it is so hard to implement. Her presentation on AI adoption was one of the clearest at the summit.
She used a metaphor of the AI Iceberg.
Her argument was that technology and algorithms are just the visible tip. Roughly 30% of the challenge. The submerged 70% below the waterline is People, Organisation and Processes. The blockers are not code. They are things like:
- Lack of leadership alignment
- Inability (or unwillingness) to reimagine workflows
- Weak governance
- No clear business case for change
- Roles not redesigned to make use of the capability
That framing helps explain the “value gap” she highlighted. Many companies are stuck at “adoption” levels of ROI, but very few reach real “value extraction”. The difference is not tool sophistication. It is whether they do the submerged work.
A strategy for value: defend, extend, upend
Maria also offered a useful roadmap for where to start:
- Defend – Start with task-specific improvements that boost productivity. Get the basics working.
- Extend – Once stable, use AI to optimise existing processes. Improve what already works.
- Upend – Only then aim to reinvent by creating new products or core processes.
It’s a sober sequence. And it implicitly warns against the common pharma temptation to leap straight to “upend” while the “defend” layer is still weak.
5. Real-world use cases: where the ROI is showing up now
Beyond frameworks, it was good to see working examples of AI delivering value now, not as a PowerPoint wish.
Launch readiness at Takeda: Orbit
Daniel Stefecka’s AI-stage session, “The Insight Engine: Accelerating Pipeline Readiness through GenAI application,” showed how Takeda built Orbit, an internal insights agent supporting launch readiness.
Orbit tackles a painfully familiar mess:
- Launch insights scattered across SharePoint, Teams, emails and local drives
- Teams shrinking and more stretched
- Knowledge walking out the door with turnover
- Repeated file-hunting and re-summarising of information that already exists
The win was not flashy. It was operational, useful and genuinely impactful:
- Reduced dependency on “human routers” (people who essentially forward links and emails)
- Less rework
- More time spent on strategic thinking about positioning and competition
- Bottom-up adoption because it removed daily pain rather than adding a shiny extra step
The best bit was Daniel’s point about prompting. Early feedback was mixed not because the tool was weak, but because people were giving it vague, half-formed prompts. Training changed that quickly.
Takeaway: do not launch an agent without teaching people how to ask it competent questions. Otherwise, you will blame the tool for your own sloppy usage.
The ROI of time: small wins scale
There’s often a debate about whether AI has to “transform the whole business” to be worth it. Tom Parker (Shionogi) offered a refreshing counter-perspective:
“Even if AI only saves each person 10 minutes each day, in a company of 5,000 people that’s a massive ROI (520 days a week).”
That’s not a slogan. It’s a useful reframing. AI doesn’t need to reinvent everything overnight. It needs to remove real friction, consistently, at scale. Most companies will not win through one miracle use case. They will win through dozens of boring, well-governed time-savers that stack up.
Hyper-personalisation and CX at Zambon
Maria also tied adoption back to customer experience. Her circular CX model balanced two forces:
- Maximise content – Faster production and localisation, more responsive omnichannel output.
- Strengthen capabilities Better training, clearer autonomy, and organisational readiness so AI improves decisions rather than just output volume.
The implicit point was right: AI has to sit at the centre of business strategy, not on the edges of content production.
6. The cultural barrier: fear vs curiosity
Technology is easy. People are hard. That was the undercurrent of many conversations.
Vladimir Nimec (Daiichi Sankyo) gave the most profound quote of the event:
“Curiosity only kicks in when you’ve addressed fear.”
This is the unspoken KPI of digital transformation. If teams fear AI is there to replace them, to expose their weaknesses, or to put compliance risk on their shoulders, they will never use it creatively. You will get reluctant box-ticking at best.
We also heard evidence that warmth and competence are predictive traits for field rep access, yet only about a third of reps score high on both. That frames why AI coaching for field teams has traction: not as surveillance, but as a low-risk rehearsal space to build competence and confidence, letting reps practice without fear of failure in front of a customer.
If you don’t address those fears explicitly, adoption stalls no matter how good the tool is.
7. Build or buy: a question teams can’t dodge
The “Build or Buy?” session pushed another uncomfortable but necessary question. Do we build capability in-house, partner, or buy something already stable?
The speakers at Vienna didn’t pretend there is one answer. But the context matters. If you don’t have strategic clarity, data readiness, workflow integration and governance, building more tech internally will just create more surface area for confusion. If you do have those things, building can make sense. Partnering sits in the middle when speed and risk reduction are priorities.
The point is not to fetishise “build” or “buy”. The point is to make the decision on value and readiness, not fashion.
8. Final thoughts: leaving AI tourism behind
We are leaving the era of AI tourism, where companies ran endless pilots just to say they were “exploring”. The winners in 2026 will be the ones who focus on the submerged 70% of the iceberg: people, processes and culture.
The speakers who stood out most for me weren’t waving magic wands. They were doing the boring but essential work: clean data foundations, workflow fit, change enablement, role redesign, and real measurement.
At Kanga Health, we’re seeing the same patterns. The technology is ready. The question is whether organisations are willing to fix what’s underneath it.
A big thank you to the organisers of NEXT CX & AI for a sharp, well-structured few days in Vienna. And to the speakers who resisted the urge to sell the future and instead talked about making the present work. It was great to connect with so many smart people focused on staying sharp.
