Artificial Intelligence adoption is accelerating across the healthcare and life sciences sectors, but poorly designed processes are emerging as a major obstacle to successful implementation, new research from Camunda suggests.

The enterprise orchestration company’s latest report found that 66% of healthcare and life sciences organizations have experienced an AI initiative fail because of process-related challenges. On average, those failed initiatives cost organizations $1.59 million.

The findings have particular relevance for drug developers and other highly regulated life sciences organizations, where AI increasingly intersects with complex workflows spanning research, development, manufacturing, quality and compliance.

Rather than redesigning those workflows around AI, many organizations are adding the technology to established processes. According to the AI Process Gap report, 75% of respondents said integrating AI into existing processes creates less internal resistance, while 82% expect it could take up to five years to adapt their most important processes for AI. That approach could leave organizations struggling to capture the expected value from AI investments.

The research indicates that life sciences organizations recognize the problem but are finding it difficult to address. Eighty-three percent of respondents said their AI investments could fail without additional investment in process redesign. Meanwhile, 71% said AI costs could spiral unless they gain greater control over their business processes.

The challenge is partly a question of timing. AI technologies and applications are evolving rapidly, while changing established processes—particularly in highly regulated environments—can require extensive planning, validation and organizational alignment.

Kurt Petersen, SVP of Customer Success at Camunda, says organizations are often taking the faster route of integrating AI into legacy workflows rather than rebuilding those workflows around new capabilities.

The report found that 59% of respondents believe process redesign cannot keep pace with the speed at which their organizations need to adopt AI. For drug development organizations, the issue can extend beyond productivity. AI-enabled workflows may need to operate within established quality systems, approval structures and compliance requirements, making poorly integrated technology a potential source of additional operational risk.

Process weaknesses are also becoming intertwined with AI governance. Forty-one percent of healthcare and life sciences organizations surveyed reported an AI-related compliance or governance issue within the previous 12 months. Process-related problems contributed to 80% of those incidents. The prospect of increasingly autonomous AI systems is adding another layer of concern. Half of respondents said they are worried about AI agents making unintended changes to processes.

Employees appear similarly cautious. Eighty-one percent said they fear their use of AI could result in a compliance issue within their organization, while 95% expect process-related problems to contribute to future AI-related compliance incidents.

For organizations operating under stringent regulatory requirements, the findings highlight the importance of establishing appropriate controls around where AI can act, what decisions it can influence and when human intervention is required.

There is also a notable disconnect between how organizations view their AI investments and how employees experience them. While 89% of healthcare and life sciences organizations said AI is making their teams more productive, just 55% of employees agreed. Nearly half of organizations surveyed—46%—said they had scaled back AI use because the technology negatively affected employees’ ability to perform their jobs. Forty-four percent of employees said poorly implemented AI could be enough to make them consider changing jobs. The research suggests that implementation decisions are frequently being made without sufficient input from the people expected to use the technology.

Seventy-seven percent of employees said they were not fully consulted about how AI would affect their day-to-day work. At the same time, 68% said they could use AI much more effectively if it were implemented differently. Some employees are already compensating for shortcomings in AI-enabled workflows themselves. According to the report, 43% said they had manually overridden AI outputs because the underlying process was not configured correctly. Another 39% said they had used AI simply to satisfy an organizational mandate when there was no practical need to do so.

For life sciences companies, these workarounds could be particularly significant when AI becomes embedded in workflows associated with research, manufacturing or quality operations. A technology intended to reduce manual effort can instead add new layers of review and intervention if the surrounding process has not been properly redesigned.

The issue is likely to become more important as organizations move beyond isolated AI pilots and begin integrating the technology into core operations. In drug development, where processes often span multiple departments, systems and stages of the product lifecycle, changes to one workflow can have consequences elsewhere in the organization.

That makes process visibility and governance increasingly important. Companies need to understand how AI fits into existing workflows, where decisions are being made, where human oversight remains necessary and how changes can be documented and controlled.

The findings suggest that the next challenge for life sciences organizations may not be simply determining where AI can be used, but determining how the processes surrounding it should change. Adding AI to an inefficient workflow may accelerate individual tasks without addressing the underlying bottleneck. In some cases, it could make an existing process more complicated by introducing additional review, validation or exception-handling requirements.

For drug developers, the stakes can be particularly high. AI applications are being explored across areas ranging from research and discovery to clinical development, manufacturing and quality operations. As those applications mature, the organizations that can integrate them into well-defined, controlled workflows may be better positioned to realize their potential without creating unnecessary operational or compliance burdens.

Camunda’s research ultimately points to a broader shift in how companies may need to approach AI implementation: technology deployment and process redesign cannot be treated as separate projects. For life sciences organizations, that means the AI strategy may increasingly begin not with the question of which model or agent to deploy, but with a more fundamental one—whether the process itself is ready for AI.

Download the full report.