What will decide the future of AI agents in APAC?

APAC’s challenge isn’t adopting but advancing AI, with many brands stalling at entry-level agentic AI.

We discussed with Ruslana Reznikova, VP, General Manager, APAC & Eurasia, Infobip, the difference between brands stuck in entry-level automation and those graduating into orchestrated journeys that actually move ROI. 

Ruslana Reznikova (RR): APAC brands show strong adoption of agentic AI, with our 2026 CX Maturity Report putting adoption at 57%, ahead of the 53% global average. With WhatsApp usage sitting at 91% and in-app messaging at 60%, the region also has the strongest adoption of messaging channels that shape customer experience.

This has enabled AI adoption to move fast as agents need an existing channel to operate on, with customers meeting them there at scale. On paper, APAC looks like the most advanced market in the world.

That impression doesn’t hold up once you look at advanced agentic adoption, where agents operate autonomously across multi-step customer journeys rather than simple tasks. Here the figure drops to 34%, compared to the global average of 42%. Brands have built the channels and deployed the chatbots, but 54% still point to integrating AI with their existing tech stack as the barrier, well above the 41% global figure.

The reason comes down to how a brand’s tech stack operates. The various tools they use, whether it’s the contact centre platform, the chatbot, or the messaging channel, tend to operate within their own slice of customer data. An agent can answer accurately in isolation but may not have visibility into what a customer said five minutes earlier on a different channel.

For brands to graduate past the entry level, the starting point is connecting that data. Brands need a unified view of the customer at every channel, paired with an orchestration layer that decides what happens next in the interaction. Once that layer is in place, brands can trust an agent’s autonomy rather than managing it channel by channel.

Why do you see the move from pilot to production as fundamentally a data and orchestration challenge rather than a question of model capability, and where should enterprises start when rethinking fragmented data foundations?

RR: The decision of which model matters, but it’s not what makes or breaks a deployment. An AI system stalls regardless of the model if it can’t access the right context and data, because its infrastructure is still operating in silos, the same silos a pilot never has to deal with. Pilot conditions are already controlled for, the data is cleaned up, the scenarios handpicked, and success measured by only whether the agent got the answer right.

Systems like CRM, contact centers, and billing platforms weren’t built for real time information sharing, so an agent that performs well in a demo often falls short when it needs data spread across different systems, especially when a customer expects the brand to remember something they mentioned days earlier.

APAC’s ceiling isn’t about capability so much as connection. The region holds the highest System Potential score globally at 61, ahead of the Americas at 58 and Europe at 54. That score measures how much more a brand’s existing technology could deliver to its customer journeys if those systems were properly linked. Closing that gap is what orchestration means in practice, bringing the systems together so an agent can act across them.

Enterprises rethinking their data foundations should start with an honest audit of three things:

• Data availability is where the audit needs to start. A lot of enterprise data still sits behind interfaces built for people, dashboards, reporting tools, systems someone needs to log into. Agents need to reach that same data through APIs instead, and today only 54% of APAC tools are API ready. If the only route to a piece of information is a human clicking through a screen, an agent can’t act on it.

• Training agents to understand what they’re looking at matters just as much, because access alone doesn’t mean an agent knows what the data means. Enterprises need to give agents context on what each field represents, how reliable it is, and how different pieces of data relate to each other, a customer record connecting to an order, an order linked to a support ticket. Without that structure, an agent can retrieve accurate information without understanding the story it’s telling.

• Governance closes the loop, defining what an agent can and cannot access, and under what conditions. AI agents can be confidently wrong, so without clear boundaries on what data an agent is permitted to see or act on, that confidence becomes a risk rather than an asset.

RR: As a mobile-first region, customers expect messaging from their favourite brands as a baseline and that shifts market to market. RCS usage in APAC sits three times higher than other regions, but baseline adoption doesn’t mean the same channel wins everywhere, preference still varies sharply by country.

Brands in Malaysia and Indonesia lean on WhatsApp, Japan runs largely on LINE, South Korea on Kakao, Vietnam on Zalo, and the Philippines relies heavily on Viber.

Regulation adds another layer, with countries across Asia writing their data protection laws, and those rules keep shifting. Many require customer data to stay within their borders, and also govern how personal data can be processed, shared, or erased. For example, a bank in Singapore and one in Indonesia can face different requirements for verifying a customer’s identity, capturing consent, or triggering an escalation, even when the underlying customer problem is the same.

Translation only solves part of the problem. How a customer explains a complaint depends on where they’re from, how direct they tend to be, how much detail they naturally give, how urgent they sound. An agent trained on how one market phrases a complaint will often misread intent in another, even though the complaint hasn’t changed.

That matters most at escalation, where a system built to detect urgency, frustration or a compliance sensitive keyword in one language has to perform just as reliably in the next, or the handoff to a human agent happens too late.

An agent built around one channel and rolled out unchanged across the region will fail in most of these markets. Building a separate agent for every market isn’t the answer either. What works is the same orchestration and governance already needed on the data side, keeping every market consistent, while still leaving room for the local flexibility each market needs, so an agent can recognise intent and hold service standards everywhere.

Through this, enterprises can get the balance right and autonomy will be able to scale cleanly across markets.