How does a small or mid-sized biotech or pharma company make sure AI delivers on its promises and moves beyond the pilot phase? At a corporate level, there are two perspectives to consider. First, the technology itself. Second, the people: how to train, support, and enable the workforce so that AI helps the company achieve its goals. You need both. A technically sound deployment can still struggle if employees don’t understand why they should use it or what it means for their work.

Let’s start with the technology. Two priorities deserve particular attention: security and access controls, and data integrity. These aren’t the only technical considerations, but they are foundational. Permissions and workflows need to be clearly understood, documented, tested, and audited so that an agent, or someone using it, cannot access information or take actions they shouldn’t. You also need to establish where the agent can act independently and where a person needs to review or approve its work. In a life sciences organization, those boundaries matter, especially when the work could affect patients, regulated records, or important business decisions.

The second priority is the integrity of the data going in and the reliability of the information coming out. Those are related, but they aren’t quite the same thing. AI can produce an incorrect answer because it is working with bad data, but it can also generate an unsupported answer even when the underlying data is sound. Better data quality reduces one source of error; it doesn’t eliminate hallucinations. That means the technology team needs to consider both the quality of the source information and how AI outputs will be checked before someone acts on them.

Data integrity is certainly not a new concern. What AI changes is how quickly information can move into an answer, a recommendation, or an action—and how readily someone might trust the result. An employee may not know enough about the source material to challenge an answer, or may become less inclined to question it as their confidence in the tool grows. We’ll come back to that in the culture discussion. From a technical perspective, the point is straightforward: if AI and the people using it are going to influence decisions, they need reliable information and a way to trace important claims back to their sources.

The good news is that AI can also help improve data quality. It can help data teams identify potential issues earlier and focus their attention on records that need investigation. Take clinical trial data, for example. It often depends on busy people entering information into multiple systems, which creates opportunities for manual entry errors, missing information, and inconsistencies. Other issues, such as enrolling ineligible participants or deviating from the protocol, can affect participant safety and the reliability of the results. Unclear endpoints create a different problem: the trial may not be designed to answer its intended question clearly in the first place.

AI can help flag some of these issues, such as inconsistent entries, missing assessments, or patterns that warrant a closer look. Some checks may be better handled by straightforward validation rules; others may benefit from AI. The goal is to catch problems closer to the point of entry, while the people involved can still investigate and resolve them. But a flag is a reason to investigate, not proof of an error. These tools need to be evaluated for their intended use, and any corrections need to preserve the appropriate review and audit trail. AI can support the people responsible for data quality; it doesn’t remove that responsibility.

Now let’s get to what I think is the more interesting corporate perspective: the people and culture side. When discussing a successful AI deployment, notice that I haven’t defined success as “high adoption.” I’ve defined it as delivering on its promises. That starts with being very clear about what the deployment is meant to achieve for the business.

I often hear that a company wants to deploy AI to “increase workforce productivity.” That’s a reasonable starting point, but it isn’t a fully developed objective. What organizational outcome is driving the desire for more productivity? Is the company trying to manage costs and extend its financial runway? Is there a temporary increase in workload over the next six to twelve months that doesn’t justify permanent hiring? Is there a capability gap that AI could help employees address? Those are different situations, and they call for different decisions about where to use AI and how to measure success.

If leadership doesn’t explain the reason, employees will come up with their own explanations, often starting with the worst-case scenario. Being explicit gives people a clearer understanding of what the organization needs and why their participation matters. Another objective might be reducing avoidable delays in development. Perhaps new competitor data has increased the urgency around a program. Where could AI help the organization work faster while maintaining the necessary quality and review? Or perhaps recurring data quality issues are creating too much rework. In that case, success might mean fewer unresolved discrepancies and less time spent correcting them. The objective needs to be specific enough that employees can understand it and the company can tell whether it has been achieved.

For a small or mid-sized company, this also means being realistic about who will own the work after the pilot. Who is responsible for the business result? Who will support the tool, resolve issues, and decide whether it is still performing well enough to use? A promising pilot needs an ongoing owner, resources, and a place in the actual workflow. Otherwise, it can remain an interesting demonstration that nobody has the time or responsibility to turn into a working capability.

The next step is explaining what the business objective means for the employee using AI. If you expect someone to become more productive, what do you want them to do with the time they gain? Imagine a manager saying, “We want to reduce the time you spend checking routine data issues because we need more of your judgment. We want you to spend more time interpreting the data, finding useful ways to analyze and present it, and helping colleagues understand what it means. Too much good information is being misunderstood or ignored, and you could help change that.”

How would you feel if your manager described that future for you? For many people, that sounds exciting. It gives them a clearer purpose and an opportunity to grow. It also makes the productivity goal concrete: the organization wants more time spent on interpretation, communication, and better decisions. Of course, that conversation has to be honest. If the goal is cost reduction or a change in staffing needs, leadership should explain what is known and acknowledge what is still uncertain. A motivating message only works if employees believe it.

The third cultural priority is helping employees recognize AI risks and feel comfortable raising them. This needs to go beyond telling people, “If you see a hallucination, let someone know.” Employees need to understand what to look for: unsupported claims, missing context, outdated information, inappropriate access, or outputs that look plausible but don’t fit the evidence. They also need to know what to do next, who can help, and when to pause the work.

What does a culture that supports this actually look like? It comes back to what good People and HR teams have been saying for years: culture has to show up in everyday behavior. It needs to be reflected in meetings, management practices, performance discussions, and the actions leadership recognizes. That means thanking someone who raises a credible concern, even if the investigation finds no problem. It means being willing to say, “We have a tight deadline, but we need to pause and check this.” It means leaving time in a planning or technology meeting to ask, “What could go wrong here, and what might we be missing?” People should see that careful judgment is part of doing good work.

As I’ve mentioned in previous articles, the partnership between the People or HR team and the technology team is becoming more critical as organizations deploy AI. Technology teams need to care about how people understand and use the tools. People teams need to help employees build the skills, confidence, and motivation to work effectively with them. Together with the business leaders responsible for the outcome, they can connect the technology to a clear purpose and help employees see their place in it. That’s the partnership that gives an AI deployment a real chance of moving beyond the pilot and becoming part of how the company works.