How the content supply chain works, and what AI is changing
by Rob Verheul 12 November 26A pharma company has two jobs: to make medicines, and to tell people about them.
It already knew how to run factories for the first, so when it needed to produce communication that would connect with clinicians in dozens of markets, compliantly and at a sustainable cost, it built one for the second.
These content factories were a logical answer to genuinely hard problems of scale: claims could be approved centrally instead of being rebuilt in every market, brands stayed consistent across languages, and smaller markets got material they would otherwise never have had. Most large pharma companies now run one in some form.
AI is now being applied to those factories to speed them up and reduce cost: more content, in more markets, faster, at a lower cost per asset.
At the same time, clinicians are already engaging less with what the factories produce, and more of it, faster, won’t change that on its own.
This article, originally published in the Designed for Impact newsletter, sets out how the factory works, what AI adds to it, and why the distance between what it produces and what clinicians value is growing rather than closing.
How the model works today
Most content supply chains share the same basic shape. At the top sits a global core, which holds the approved claims, key messages, references and imagery for a brand, signed off by medical, legal and regulatory review. That core feeds the factory, where content is assembled from approved modules and adapted for local markets, much of it through agencies and offshore production hubs. Out of the factory come the assets themselves (web pages, banners, emails, video, eDetails), rolled out across every market and brand, and delivered through CRM, marketing automation and field teams.
Underneath all of it sits a content platform that acts as the system of record, tracking what has been approved, where, and for how long.
The logic is modular: approve the parts once, then recombine them many times. Most of the efficiency the model delivers comes from that idea.
Where the factory falls short
The model was built to solve a production problem, and it solves it well. Its weaknesses sit in what production was never designed to address, and they were there long before AI.
The first is the content itself. Everything the factory makes is derived from the global core, so it can't be more useful than the core allows. Last week I argued that the lived experience of a condition is rarely at the centre of core brand messaging, and therefore of everything built on it. However efficient the factory becomes, it can only reproduce what it's given.
The second is the experience. The factory is organised around assets, while a clinician's experience is organised around their own needs, and the two don't always line up. Joining them together takes design work that sits outside the production line, so it tends to get less attention than the assets themselves.
What AI adds
AI is now taking on roles across the whole factory. It creates and adapts content from the approved modules, is starting to check it for compliance before it reaches formal review and, through next best action (NBA), recommends what each clinician should receive next, and when. The factory is no longer only making content. It's increasingly shaping where that content goes. The result is more content, in more markets, faster and more precisely targeted.
What it's genuinely good at
Where it's working, the gains are real. Cost per asset falls, and local teams stop rebuilding the same email for the fifth market. Reviewers see fewer assets with avoidable errors, so approval cycles shorten. Smaller markets get material they could never have afforded to commission. For an industry that has spent years buried in production, that matters, and nothing that follows is an argument for going back.
The question is what happens to the factory's two weaknesses when production gets this much faster.
What AI doesn't fix
Faster production carries both weaknesses forward. The gap in the core is reproduced in more formats, more quickly, for an audience that is already disengaging, and ‘next best action’ technology can put those pieces in a better order without turning them into an experience built around what a clinician actually needs.
AI also adds a risk of its own. When competitors in a therapy area use similar models on similar inputs, their content converges, in what it says and in how the brand comes across, towards the same bland middle. Compliant and quick doesn't mean compelling. And to the extent answer engines draw on brand material at all, there's little to tell one brand apart from another.
The organisational risks, around governance, quality standards and the expertise that has never been written down, I covered in September in "Three risks in the AI content supply chain transformation, and none of them are technical".
Why the gap grows
There's a reason these risks persist: production keeps improving, while what the brand says, and why, tends to stay where it was. Nothing in the system requires the core to change in response to what clinicians do. It changes when the label, the data or the brand plan does.
The obvious objection is that the core can’t move because it’s bound by the label. The claims are, but the proposition isn’t, at least not in the same way. What you choose to say around those claims, to whom, in which moment and why, leaves far more room than most cores use. While that stays fixed, every gain in speed widens the distance between what the factory produces and what clinicians value.
What would connect the two is the part of this I'd rather work through in a room than in an article. It's the focus of our quarterly roundtable with digital and commercial leaders, which is built around practical conversation rather than presentation, and I don't think anyone in the industry has fully answered it yet, including the vendors selling the factories. If you'd like to join, get in touch and I'll share the details.
AI will make every content factory faster. Whether it makes any of them better is a different question.
