Companies are becoming more mature and more nuanced in the way they roll out various types of AI - and in the rewards they expect to reap, according to a new report from the IBM Institute of Business Value (IBV). The initial euphoria about the transformational benefits that gen AI can deliver has given way to a more realistic understanding of what is possible.
The report "AI at the core: From AI projects to profit" is based on the findings of two surveys of business executives conducted in partnership by the IBM IBV and Oxford Economics.
Overall it suggests that companies are becoming more strategic in deploying AI initiatives. They're moving away from the types of short-term AI projects they initially explored when gen AI was still very new and untested.
Now they're focusing on more considered deployments, targeting core business areas that can potentially have a longer-lasting bottom line impact. There's also a growing focus on agentic AI.
1. A reset for AI
In 2023, many early gen AI pilots promised and initially delivered returns as high as 31% according to the research. However, the ROI from those initial short-term initiatives has now settled at a less exciting 7% (lower than the typical 10% cost of capital goal for capital expenditure spending).
In fact, over the past three years, only 25% of AI initiatives have delivered their expected ROI, although the top 10% of AI projects still achieved ROI of approximately 18%, so higher returns are still achievable.
An underlying trend is that 64% of AI budgets are now directed at core activities (compared to 36% for non-core) with only 6% of enterprises now deploying AI in an ad hoc way (down from 19% a year earlier). Since transforming core activities tends to be more complex and likely to take longer, this pattern may go some way toward explaining the lower returns we're now seeing.
And while running ad-hoc pilots with AI can be helpful for initial learning, they won't deliver the systematic benefits that can come from coordinated, enterprise-wide strategic approaches, the report suggests.
Nevertheless, many companies are taking a cautious approach. Less than a quarter are "reimagining their workflows with AI at the center" or using AI "as a core engine for growth in their products", or "fundamentally reimagining their business models". This lack of ambition could mean that they miss out, according to the report's authors. They state that AI's greatest potential lies in finding entirely new approaches to value creation and delivery rather than making existing processes marginally more efficient.
2. Two-tier AI adoption
The report defines enterprises with a more mature, strategic approach to AI as "AI-first" organizations. They have "robust foundational AI capabilities" and "a transformational mindset" and tend to perform noticeably better than those that use AI in tactical, fragmented ways.
Around 25% of organizations qualify as AI-first adopters, who can demonstrate gains in revenue and operating profits attributable to their AI initiatives. In fact, over the past year, AI-first organizations have been able to attribute more than half of their revenue growth and operating margin improvements to AI initiatives.
These organizations use AI in ways that will make a dramatic difference in the future. For example, they're more likely to create entirely new opportunities that did not previously exist. So they aren't just using AI to optimize the present, but to reimagine what will be possible in the future. They're also more likely to use AI to try to target new customers and markets.
Therefore, AI-first organizations are more likely to use AI to expand their business models rather than simply operate more effectively within current limits.
Another trait is that more AI-first organizations (68% versus 54%) are treating AI initiatives as an "innovation portfolio". This means they explore short-term, lower-risk projects, as well as attempting to use AI for more ambitious, potentially transformative initiatives that may involve higher risk.
Having a portfolio of different projects like this means they're willing to experiment with riskier transformative initiatives while at the same time stacking up a steady stream of revenue earning smaller 'wins' from less risky initiatives, which helps ensure ongoing support from business stakeholders.
68% of the AI-first organizations also report having more mature and robust data and governance practices with well-established frameworks, compared to 32% of other organizations. They recognize the importance of maintaining good data hygiene and ensuring that data is trustworthy and accessible to both AI agents and humans.
According to the report, the enterprises that are going to see the best performance from AI will need to demonstrate "both technical sophistication and strategic clarity in their AI approach". They're the ones who have created an AI-enabled operating model that improves efficiency, enhances customer relationships, and accelerates innovation all at the same time.
3. Interest in AI agents soars
Businesses are extremely optimistic about the potential of agentic AI. 70% of executives consider agentic AI important to their organization's future, while 76% of them are actively encouraging experimentation with agents. And AI-enabled workflows are expected to increase 8-fold (from 3% in 2024 to 25% by 2026).
By 2026, 83% expect AI agents to dramatically improve the process efficiency by handling repetitive, rule-based tasks at speeds exceeding human capabilities, leading to a significant transformation in how work is done. Since agentic AI is not merely about automation but about "intelligent orchestration", many workflows could potentially be redesigned with AI providing a better understanding of what is optimal and efficient.
The two concrete benefits of agentic AI that executives rated the most highly are: AI agents can improve decision-making by enhancing access to data and insights (69%); and they can reduce costs through automation (67%).
Interestingly, echoing the earlier report findings, the research suggests agentic AI deployments also tend to be targeted at enhancing areas that can be described as "core" activities. They include research and innovation management, supply chain, sales and customer service.
Among the main challenges and barriers cited for agentic AI are "Concerns about intellectual property" (50%), "Data accuracy or bias" (49%) and "Trust issues" (46%). These underline the fact that AI agents are highly dependent on the quality, accessibility, and governance of the underlying data and information and stress the importance of human oversight.
The report suggests that success in agentic AI is not just about deploying technology correctly. Enterprises will need to rethink their approach on multiple levels. Agentic AI, say the report's authors, "requires a deep reconsideration of how work is structured, how decisions are made, and how humans and machines collaborate. And organizations "must develop new governance frameworks that balance autonomy with accountability, speed with safety, innovation with reliability".
Conclusion
Overall, the report's findings suggest that as AI adoption matures, companies are learning to move beyond fragmented projects toward more integrated, strategic, deployments in core business areas. Enterprises that combine strong data governance, clear objectives, and a willingness to use AI as an opportunity to rethink current ways of working are likely to see more measurable results. While not all initiatives will deliver immediate ROI, the research suggests that a strategic, long-term approach is key to using AI to achieve sustainable ROI and a lasting competitive advantage.
This blog was originally published on the IBM Community.