As enterprises accelerate investments in AI, many are discovering that the hardest part of adoption is no longer experimenting with large language models but turning them into trusted systems capable of running critical business operations. They are also baffled about what return on investments (ROI) will AI deliver.
According to Manik Narayan Saha, Managing Director for SAP Labs, East Asia, that may no longer be the right question. Speaking to iTNews Asia at the ATxSummit, Saha argued that traditional ROI models struggle to capture the impact of AI because the technology is fundamentally changing how work is performed rather than simply making existing processes faster.
Instead of relying solely on conventional financial metrics, organisations should focus on the broader business value AI creates through productivity gains, faster decision-making and new ways of operating, he added.
Saha said traditional ROI models assume that technology simply improves existing processes. AI, however, is changing how those processes are designed and executed in the first place, making it increasingly difficult to compare performance against historical benchmarks.
"As organisations redesign workflows around AI, the baseline itself changes. You're no longer measuring incremental improvements to the same process but measuring entirely new ways of working,” Saha said.
He added that enterprises should therefore look beyond conventional financial metrics and instead assess how AI contributes to faster decision-making, greater productivity, improved customer experiences and new business capabilities.
Look at business value, not financial metrics
While assigning a dollar value to AI remains challenging, Saha said many productivity improvements are already evident within SAP's own engineering teams.
Over the past year, some development teams have reached a point where between 30 percent and 40 percent of the software development lifecycle is now AI-assisted. The technology is helping engineers accelerate coding, testing and software releases while allowing more product features to be delivered within the same development cycles.
Rather than pursuing AI projects with narrowly defined ROI targets, Saha believes organisations should evaluate AI based on the business outcomes it enables.
He said technology generally creates value in one of two ways either by reducing operating costs or helping businesses generate additional revenue, but AI's impact often extends beyond those traditional categories.
As AI becomes embedded into enterprise software, it can shorten decision cycles, improve employee productivity, accelerate product development and streamline business operations in ways that are difficult to isolate through conventional ROI models.
- Manik Narayan Saha, Managing Director, SAP Labs, East Asia.
Trust becomes the competitive advantage
For enterprise software, reliability is becoming as important as intelligence.
Saha said organisations need AI systems that consistently produce explainable results, protect confidential information, minimise bias and operate within clearly defined security boundaries.
This becomes even more critical as autonomous AI agents begin making business decisions independently. "Every action an AI agent takes should be logged, traceable and auditable," he said. "If something happens, you need to understand why it happened so you can fix it." He added that agents should also operate under strict authorisation controls rather than having unrestricted access to enterprise data.
Further, technology alone will not determine AI success, Saha said, pointing instead to workforce capability as a critical factor. He believes many organisations will need similar initiatives as enterprise AI adoption expands.
Focus on value, not hype
Instead of trying to automate everything immediately, Saha recommends organisations prioritise business value and develop phased adoption strategies.
Early adopters have already begun embedding AI into functions such as travel expense management, HR recruitment and performance management, where repetitive workflows can be streamlined without compromising governance.
Looking ahead, Saha sees autonomous enterprises becoming an achievable long-term objective, but only if organisations first build the necessary foundations. "My key advice to companies would be to start now because the train is already moving," he said. "Companies that don't get onto this AI bandwagon right now will be left behind in the next two to three years."
