4 Questions pharma leaders can’t skip before letting AI act

There is a moment, somewhere in a pharma content workflow, when a draft stops being a draft and becomes an action – sent, routed, acted on. Agentic AI is pulling that moment earlier and earlier, sometimes past the point where a person would naturally review it. In pharma, that shift is one I have been watching closely. When a generative tool drafts a claim or suggests a layout, someone typically reads it before it goes anywhere. When an agent plans a task, selects content, and passes it downstream, an error can travel further before anyone catches it.

Pharma is adopting this technology in one of the most tightly regulated environments in business. Outputs that influence healthcare professionals or patients must remain controlled, reviewable, and traceable to appropriate evidence. So, the question for our industry is not which AI model to choose, but what teams must do before letting AI act on their behalf, and how that groundwork becomes a competitive advantage.

A different category

Much of the current conversation treats agentic AI as generative AI with more horsepower. It is, in fact, a separate class of system. Given an objective, an agent breaks it into steps, uses tools, and makes decisions to reach it. Governance designed for a drafting assistant, where someone reviews the output, does not fit a system whose output is a sequence of actions. That is where I have seen the job of leadership shift. The task is no longer to approve a tool, but to define what an autonomous system may and may not do, what it can touch, and who is accountable for the result.

Gartner's research suggests that more than 40% of agentic AI projects will be cancelled by the end of 2027 due to escalating costs, unclear business value, or inadequate risk controls. At least two of those causes can be addressed at the organisational level.

McKinsey's 2026 survey adds a different angle. Among respondents from organisations with annual revenues above $1 billion, 40% report scaling AI agents, up from 27% a year earlier, while the share at smaller organisations has stayed flat at 22%. Larger organisations have more capacity to fund and staff the work that scaling requires. Smaller ones often have the same ambition on a thinner operational and financial foundation. Their agents are more likely to stay in pilots, because the cost of getting it wrong is harder to absorb. In pharma, where customer-facing content goes through Medical, Legal and Regulatory review, that foundation matters even more. For many companies, regardless of size, ambition is moving faster than foundation, and the companies that close that gap first will pull ahead.

Readiness is a data question

In my experience, the organisations moving fastest with agentic AI are the ones that sorted out their data foundation first: knowledge and assets that are structured, governed, and traceable. The reason is speed. A person handed duplicated, outdated, or ambiguous information will stop and ask a question. An agent, most likely, will not. It will act on the ambiguity at machine speed, with every downstream step inheriting the error.

For any piece of content, an agent needs three answers: whether it is approved, whether it is current, and where it came from. If those answers live in people's inboxes, chats, and even memories, rather than in the content itself, the agent has no way of knowing. It will simply use whatever it finds.

In my conversations with pharma leaders, I often find that their teams are closer to having those answers than they realise, and the reason is modular content. Modular content is, in effect, structured data: claims, references, visuals, and messages exist as tagged, versioned, approved components, rather than finished documents. When content is modular, approved, and traceable, the agent works from cleared material, and every output can be checked back to its source, so a mistake is caught at component level, rather than multiplying across assets and markets.

Compliance is architecture

In many AI-assisted workflows, the principal control is still review of a draft or finished asset. A person reads the draft, catches the problem, and sends it back. Agentic workflows require more: controls that apply to the sources an agent can access, the decisions it makes, and the actions it initiates at every step. Those actions happen in sequence, often faster than a review queue can follow. One unapproved claim does not stay a single claim for long: it becomes part of an asset, then a campaign, then several markets, and each step makes it harder to trace and more expensive to pull back.

So, what must be true before letting software act on a team’s behalf? Compliance has to be part of the architecture, and four things must be in place before deployment.

• Guardrails are defined. Which content the agent may use, which codes it must meet, and which actions always need human sign-off are written down.
• Traceability is built in. A reviewer can reconstruct what the agent did and why, without relying on anyone's memory.
• Every output links to an approved source. Nothing that reaches a healthcare professional or patient goes out without cleared material behind it.
• A human stays in the loop, with real authority. An accountable expert reviews at defined points in the workflow and can stop the process, not just approve the end result. They are trained to recognise when an agent has gone wrong, and that training keeps pace with the technology.

Get all four right, and compliance becomes what permits scaling with confidence, because every action the agent takes can be traced and defended.

These principles are shaping how we develop eVa AI within eWizard. Today, eVa AI can help teams find and reuse approved content from the digital asset management (DAM) system and generate new material with clear labelling, while eWizard applies compliance checks against industry codes such as EFPIA and ABPI. Our direction is to extend these controls across increasingly agentic workflows, while preserving source traceability, defined human decision points, and formal MLR authority.

4 Questions to answer before an agent goes live

Before any agentic system goes live, I would ask four questions:

1. Is the data structured, approved, and traceable?
2. Are the guardrails defined (approved sources, applicable codes, and where a human must sign off), including what the agent may not do?
3. Is auditability built in from the start, so every action can be traced back to its source?
4. Does the team have the operational literacy to work alongside the agent and spot failure early?

If the answer to all four is yes, much of the risk behind the cancellations Gartner expects by 2027 are already being addressed. It is time to move from pilot to production.

The unglamorous work comes first

The pharma companies that gain most from agentic AI will be those that start from the same premise: what must be true of an organisation before letting software act on the team’s behalf? It is unglamorous work: structuring content, embedding compliance in the organisation’s architecture, and growing human expertise. But that work is what moves agentic AI out of the pilot stage.

At Viseven, we have spent years building these foundations, from modular content to MLR automation to governed AI in eWizard. If you are deciding where your own organisation stands, start with eWizard and see how eVa AI works inside a governed workflow.

About the author

Nataliya Andreychuk is the CEO and co-founder of Viseven, a global MarTech services provider for life sciences and pharma industries. She is one of the top experts in digital pharma marketing and digital content implementation and has more than 14 years of solid leadership behind her belt. Andreychuk is among the strongest female leaders in the marketing technology world. She has been contributing her vast expertise to developing smart digital solutions and software, which are now serving clients in over 70 global markets, delivering intelligent, personalised content across channels, platforms, and countries.

About Viseven

Viseven is a global MarTech company specializing in digital content solutions for the Life Sciences and Pharma industries. With over 15 years of expertise, Viseven empowers pharmaceutical companies and their production agencies with AI-driven content management and automation solutions.

Our flagship eWizard Platform streamlines content planning, creation, distribution, and management—enhancing efficiency, reducing operational costs, and accelerating brand time-to-market. Designed for omnichannel and multichannel engagement, eWizard optimizes campaign management, data collection, and performance analysis, ensuring continuous message improvement for Brand Managers and Content Operations teams.

Visit us at viseven.com or follow us on social media: LinkedIn