Tech Explained: The invisible AI work powering better CX for firms in Simple Terms

BUSINESSES are spending more than ever on artificial intelligence (AI) agents, yet customer frustration is rising, not falling.

The technology itself may not be the problem. In many cases, organisations focus first on customer-facing AI, when the greater value lies in designing AI systems that support service teams behind the scenes.

These can lift productivity by helping teams respond faster, with better data, and in ways that meet clear business objectives.

Defining those objectives first, rather than rushing to deploy AI without enough thought, is critical, says Thomas Laboulle, founder and chief executive officer of Toku, a Singapore-headquartered, AI-powered customer experience (CX) technology provider.

“Simply putting an AI agent in front of customers without equipping it with the knowledge that data provides could end up making them annoyed,” he says.

“The conversation around AI is surfacing something many enterprises have avoided for too long: The AI ambitions being set at the board level are running ahead of the infrastructure beneath them.”

Intelligent automation, he says, will not work well on disconnected on-premise systems, but it is an opportunity that can push for the digital transformation organisations need all along.

For Laboulle, the priority is not a flashier interface, but the work behind it: ensuring AI has the right data, context and guardrails before it ever faces a customer.

Toku’s approach is rooted in its 360° CX Platform, a modular ecosystem that unifies telecommunications connectivity, cloud communications, AI and professional services into a single architecture.

Rather than bolting AI onto legacy systems, the platform was designed from the ground up to embed intelligence in every layer.

The efficiency of invisibility

The toughest part of enterprise AI is integrating it into day-to-day operations, a challenge many organisations still struggle with.

It is a process that involves multiple fronts and could take years, says Laboulle – from getting data ready to redesigning workflows so teams can use AI seamlessly daily.

That is why the most effective efforts, he says, often start away from the customer interface. One example is using AI to transcribe customer interactions and surface a clear summary, so an agent can grasp the issue faster without asking customers to repeat themselves.

This way, customers get quicker, clearer answers through the touchpoints they already use to reach the brand. Even when communicating with a human operator, the conversation is faster, more contextual and less repetitive.

These improvements may be invisible to customers, but they directly shape the quality of the service they receive – without forcing them to interact with more technology, says Laboulle.

Toku’s Core AI Suite supports service teams through transcription, summarisation, sentiment analysis and conversation analytics, helping agents respond faster with the right context and flagging when service recovery may be needed before issues escalate.

In some organisations, Laboulle says, these AI tools have resulted in over 30 per cent productivity gains by reducing the time spent on note-taking and context retrieval.

The most effective enterprise AI does not replace human expertise; it amplifies it. AI handles the repetitive, time-consuming tasks, so that human agents can focus on judgment, empathy and complex resolution, the areas where they add the most value.

Customers may never see the AI at work, but they feel it in smoother, faster, more consistent service, says Laboulle.

More than just speed

For regulated sectors, however, speed is only half the job. Trust, control and accountability matter just as much, if not more, when mistakes carry outsized consequences.

Just imagine an AI system giving incorrect investment advice or taking a grave misstep – such as wrongly approving a loan. The damage to the customer is immediate.

“Giving AI autonomy is not the answer. Giving it structure is, says Laboulle. “Our enterprise AI capabilities are designed around three principles: control, reliability and compliance. As capabilities expand, governance standards must be maintained.”

Regulatory scrutiny is also rising alongside adoption, he notes. Banks may be asked to explain how an AI system arrives at a decision to approve a loan, for example – which makes transparency, auditability and governance as important as speed.

In highly regulated industries, this also raises practical requirements such as data sovereignty – keeping sensitive information processed within the country to meet local regulatory standards.

Toku’s platform supports public and government cloud, and private data centre deployments, giving enterprises the flexibility to meet local data residency requirements without sacrificing capability.

This is increasingly important as regulators in Singapore, the Middle East and Europe continue to raise the bar for AI governance.

Content errors get the headlines, but in enterprise AI, the more dangerous failures are silent, Laboulle says. He refers to “process hallucinations”, where an AI confidently skips a mandatory compliance step, executes the wrong workflow or claims to have completed an action it never performed.

In enterprise environments, where each step may carry compliance or financial implications, these silent failures can trigger legal and reputational damage, and they demand a fundamentally different approach to AI architecture.

Customers feel the fallout through delays, reversals and loss of trust, adds Laboulle.

This, he notes, is why many enterprises focus on ensuring AI operates within clearly defined processes rather than relying solely on autonomous decision-making.

In Asia-Pacific, these pressures are compounded by real-world complexity, which can serve as a proving ground for enterprise AI.

Customer conversations in the region often involve multiple languages, code-switching and uneven call quality, which makes it easier for a weak AI system to misread intent or miss key details.

Designing AI for these environments requires systems that can adapt to real-world operating conditions and linguistic diversity.

Toku’s Core AI Suite is trained on regional data and designed to handle such realities, including calls with poor audio quality due to slow networks, says Laboulle.

Trust is not manufactured; it is built through reliable issue resolution and a deep understanding of customer needs, stresses Laboulle. For this, he says AI efforts today need to learn from early missteps and integrate more meaningfully into operations.

“What will separate the leaders from the rest is not who adopts AI fastest, but who has the courage to address the gap between their AI ambitions and the infrastructure underneath,” he adds.

“AI is not a technology purchase; it is an operating model decision. The enterprises closing that gap now are the ones that will lead in three years.”

The real test of any technology rollout, he notes, is whether it reduces friction or creates more of it.

He adds: “The best enterprise AI is invisible to the customer but unmistakable in the experience it delivers.”

Find out how Toku can enhance your CX journeys.