The Bottom Line: Business Users Are in the Driver's Seat
Ultimately, it's the business users who will decide whether a tool is good enough and who will dictate the pace of adoption.
This is in stark contrast to how most enterprise technology has been procured so far: through centralized IT departments.
Without falling into the trap of debating whether this is for better or worse, the undeniable value of this new paradigm is that business functions inherently know their own pain points, can size the commercial impact of a solution, and have the budget to “put their money where their mouth is”.
AI Agents as Business Companions
If Large Language Models (LLMs) represented the democratization of AI by making interactions natural-language-based, then AI agents take that concept to the next level. They are structured to solve for specific business needs from the outset.
LLMs lowered the barrier to entry by being “no-code” and “natural language-powered.” AI Agents, in turn, diminish the importance of “prompt engineering” (diminish, not eliminate) because they come pre-structured to solve a specific class of problems, for example, creating an in-depth report on a competitor or generating novel hypotheses for a research goal.
In practical terms, while an LLM can help summarize existing clinical trials and research papers, an agent can streamline much more complex, strategic, and valuable use cases, such as optimizing a clinical trial protocol from scratch.
It’s in this context that HCLS organizations should be looking at AI.
The question to ask is not "Can we use AI?" but rather, "What hard, high-value problems do we have that could be tackled with Agentic AI?"
You might ask: how am I supposed to know what’s a good problem for Agentic AI?
Here are some indicators:
- It’s Information-Intensive: A human would take months to review and absorb all the information you could potentially use to inform your decisions.
- It’s Codifiable: You can break the process down into logical steps, explain the end goal, and define the success criteria.
- It Requires Exploration: There are multiple possible—and potentially correct—solutions to the same problem, requiring creativity and discovery. This is usually where LLMs win versus more traditional AI or statistical models.
Concrete examples of these problems in HCLS include:
- Competitive intelligence
- Real-world evidence analysis
- Literature search, review, and synthesis
- Clinical trial protocol design
- And many others…
So, How to Go About It?
Although Agentic AI represents a massive opportunity to work more closely with business users, the translation between scientific challenges and technical capabilities is still very much needed.
Vendors and customers who do this translation well will innovate with AI faster and more effectively.
Here is a roadmap for a successful AI implementation journey:
- Introduction: Key stakeholders from both the technology provider and the customer meet to openly discuss needs and potential solutions. The expectation is that no one is selling or buying at this stage; both parties are listening carefully to match needs with capabilities. The people in this conversation must have both business and technical acumen to grasp the full picture.
- Showcase: The technology provider demonstrates its solutions, and the customer, in turn, showcases its current way of solving the identified problem. A word of advice for both parties: be creative. Technology providers may understand a sector, but it is the user of the technology who will ultimately identify its most valuable applications.
- Experimentation: Test the technology freely and iterate as quickly as possible. This is the step where most organizations fail, both on the client and vendor side. I will dedicate the next section specifically to this critical phase.
- Go/No-Go Decision: A successful AI implementation has time-bound milestones. The discovery and testing process must culminate in a clear business decision. This is precisely why business users should drive the process—the final verdict is about value, not just technical feasibility.
The Art of the Experiment: How to Avoid Pilot Purgatory
You've identified a high-value problem and a promising technology. Now comes the most critical phase: experimentation. This is where momentum is won or lost. Too many promising AI initiatives die in "pilot purgatory" because they are not structured for success. The most common "failure factors" are the following:
- The pilot is not led by the business. This often results in needs and success criteria that are too generic to be quantified and are disconnected from the company's bottom line.
- There is no IT partnership. This occurs when the technology enablers on the client side don't support the pilot, either because they lack the capacity or because they believe the problem can be solved with existing enterprise tools (which, in many cases, they built themselves).
- The KPIs are wrong for the stage. By setting KPIs suited for a production application rather than a pilot—metrics that could take months or years to verify—you run the risk of discarding a technology as "not a good fit," when the actual problem was the KPI, not the technology itself.
- You run the pilot without the technology provider's support. I might be biased on this one, but the reality is that only the technology provider knows how to navigate its internal product teams and can give you access to preview features that help you make the right investment before your competitors do.
Conclusion
In my role, I have the privilege of partnering with a diverse range of organizations in the Healthcare and Life Sciences space. Each has its own distinct culture, and each is moving at a different pace with a unique approach to AI adoption. The roadmap I've outlined above is a synthesis of the most effective strategies I have seen work firsthand—a playbook built from the best-in-class examples.
The common thread among the leaders is a fundamental shift in mindset. They see Agentic AI not as a complex technology, but as a powerful business companion to be wielded by the teams who know the problems best. Success in this new era is delivered through rapid, business-driven experimentation.
The journey doesn't begin with a multi-year, enterprise-wide rollout. It starts with a single, well-defined problem and a small, empowered team given the freedom to find the solution. The question is not whether Agentic AI will transform the HCLS industry, that is a certainty. The only question is whether you will be leading the change or reacting to it.
I'll leave you with one question to ponder upon: What hard problem will you solve first?
If you found this article insightful, leave me a comment, a like, and share with your network!
If you're testing Agentspace and NotebookLM and are looking for enablement/inspiration, check out this page which has HCLS-related examples.
Lucrezia
