Only one in six organizations is achieving measurable value from its AI investments. That is the conclusion of Harvard Business Review Analytic Services, based on research among 385 business decision-makers, conducted on behalf of Appian.
The picture that emerges from the data is contradictory. On one hand, AI has by now been broadly embraced: 59 percent of organizations already have the technology running in production. On the other hand, the impact remains remarkably narrow. While 64 percent of respondents report that AI has improved productivity and 58 percent see operational efficiency increase, only 30 percent say AI is contributing to new revenue streams. Fewer than 35 percent see an improved ROI. Productivity gains are plentiful, but the step change toward business value is not materializing.
The explanation lies in how AI is being deployed: only 18 percent of organizations have primarily integrated it into their business processes. On top of that, the infrastructure on which AI is supposed to run is simply not in order at many organizations. Nearly seven in ten respondents say that legacy systems are actively limiting their ability to scale AI. Additionally, 34 percent cite fragmented or poor data quality as a barrier, and 31 percent point to a lack of integration between systems. Process integration is therefore not only a strategic choice, but a technical challenge as well. “Enterprises have reached a tipping point,” says Matt Calkins, CEO of Appian. “Instead of using AI for productivity, organizations need to make the leap to business growth. The true potential of AI is only realized when it is no longer a standalone tool, but an integrated force that generates revenue.”
Integration makes the difference
The numbers confirm that integration makes all the difference. Only 16 percent of all surveyed organizations are achieving a high degree of measurable value from AI. Among organizations that have primarily embedded AI into their business processes, that figure rises to 71 percent. The research suggests that these two numbers are connected: the deeper AI is anchored in processes, the greater the return. Organizations that have also invested in modernizing their legacy infrastructure or integrating data sources perform even better: three-quarters of those organizations report strong results.
One of the most striking findings concerns AI agents. These are being deployed with increasing frequency, but primarily in relatively safe territory: software development (35 percent), IT operations (31 percent), and marketing and sales (26 percent). In the core of the operational business, in procurement, production, and supply chain, adoption lags far behind. That is precisely the domain where processes are complex, consistency is critical, and mistakes carry serious consequences.
Rules and guardrails are essential
And that is precisely where the real problem lies. Among organizations that are already using AI agents or are considering doing so, 92 percent say agents need rules and guardrails to function safely and effectively. Yet fewer than half, 48 percent, have actually documented those rules. According to the research, agents operating without clear frameworks can act unpredictably, creating the risk of unintended consequences. Mark Talbot, Area Vice President CS AI Incubation at Appian, points to an additional risk: if employees lean too heavily on AI recommendations without building their own process knowledge, meaningful human oversight becomes an illusion. “Without process knowledge, meaningful human control becomes a facade,” says Talbot. Calkins notes in a separate interview with Techzine that organizations are already aware of what is missing. “They understand what’s lacking. They know it’s still too early to deploy AI for strategic applications.” That awareness exists. The execution lags behind.
The researchers are clear about what it takes to make the next step. Organizations need not only to deploy more AI, but to deploy it differently: with more clearly defined rules, standardized processes, and better coordination across departments. Fifty percent of respondents say they are already actively working on this by defining better guardrails; 49 percent are focused on standardizing workflows.
“Organizations are adopting AI, but many have yet to integrate it into the core processes that drive business outcomes,” concludes Alex Clemente, managing director of Harvard Business Review Analytic Services. “Those who successfully embed AI in workflows are better positioned to realize true value.” The urgency is there. The intent is too: 86 percent of respondents say they want to extract more business value from AI. But intent without structure produces no results. As long as AI keeps running outside the processes, the promise remains exactly that: a promise.