Clients No Longer Want AI Without ROI

Agentic AI is entering business processes at a moment when companies are trying to move from experimentation to real operational value. In that context, Infobip has developed AgentOS, a platform for building agents and enabling interoperability among agents developed on different platforms. Krešo Žmak, Chief Innovation Officer at Infobip, says the technology is still at an early stage and has significant potential, but the market no longer accepts AI simply because it is AI.

“AgentOS enables the development of agents that help brands communicate with end users,” Žmak tells ICTbusiness Media – ICTbusiness.info & ICTbusiness TV. Agents, as he explains, can automate business processes, replace or support people in parts of customer conversations and connect with tools companies already use. The technology is still new, so its full potential has yet to be realised.

Žmak warns that many global agent projects remain in the experimental phase. In his estimate, around 95 percent of projects begin as a pilot or proof of concept and do not progress further. That shows there is interest, but also that companies are still searching for the right way to scale. AgentOS was therefore developed early, while market rules, customer expectations and distribution models are still being formed.

The solution is not intended only for large organisations, although they are natural users of more complex implementations. Žmak says there are different distribution models: from services that allow smaller companies to build simpler agents themselves to complex solutions for large organisations with many users and demanding processes. The key difference is not company size alone, but the complexity of the problem being automated.

According to Žmak, Infobip has not changed its basic way of developing products. The company still works with clients and co-creates solutions with them. “We do not work closed in a laboratory in Zagreb, Vodnjan or Sarajevo; we go out into the field,” he says. Talking to customers and understanding their needs remain central to development. What has changed is that AI has entered daily work across Infobip over the past year, first among engineers and then throughout the business departments.

AI accelerates work in some areas and can take over tasks previously performed by people in others. But Žmak believes the market is now moving from fascination towards sustainable growth and return on investment. Companies adopted AI broadly in the previous phase, after which costs began to rise. Now the question is increasingly where the actual return lies.

“Clients no longer want AI only because it is AI. They want return on investment,” Žmak says. That applies both to Infobip’s internal processes and to the products offered to customers. If agents automate part of the business, the value has to be visible in better results, lower costs, faster response, or improved customer experience. A technology solution must have a business case, not merely demonstrate technical feasibility.

That makes the next phase more demanding than early pilots. AI and agents are nondeterministic, which means user experience and cost are not always as predictable as in traditional software systems. Žmak says frameworks and solutions are already being built around this issue to stabilise experience, control costs and make systems reliable enough for serious business use.

Technology is moving so fast that Žmak is reluctant to make precise projections even for the next week, let alone a longer period. He says the speed is increasing further because AI is beginning to participate in its own development. From the business side, however, he expects continued pressure to prove value. Companies will invest in AI where they can show sustainable growth and measurable results. For Infobip, that means building a layer where agents are not just a demonstration of new technology, but a tool for concrete automation of communication and business processes.