A customer opens WhatsApp to ask about an insurance renewal. Minutes later, the conversation moves to a voice AI assistant. Eventually, a human agent takes over. At each step, the customer is asked to explain the problem again.
It is a small frustration, and one that CIOs are spending heavily to eliminate. Yet the data suggests the investment is not paying off at the pace leaders expected.
Forrester’s latest Customer Experience Index shows how uneven the progress has been. After four years of decline, North America is showing early signs of recovery: in the US and Canada, 26% of brands posted statistically significant gains in 2026, against 7% that declined. Asia Pacific, which Forrester measures through Australia, India and Singapore, is largely holding in place. Of the 57 brands evaluated in both 2025 and 2026, only 7% improved, 19% fell and 74% were statistically unchanged. Banks in India and Australia were among the bright spots.
The backdrop is sobering. A year earlier, 37% of brands in those three Asia Pacific markets saw their CX scores fall, and only 5% improved. Forrester’s 2025 findings pointed to disappointing technology implementations, including AI, among the causes of the broader global slide.
The outlook for AI itself is far brighter. Gartner predicts that by 2029, agentic AI will autonomously resolve 80% of common customer service issues without human intervention, reducing operational costs by 30%. But the same analyst firm expects more than 40% of agentic AI projects to be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls.
That tension, between the promise of AI and the reality of deployment, is where Gaurav Anand, Vice President and Global Head of the Customer Interaction Suite at Tata Communications, focuses his attention.
Anand argues that enterprises are misdiagnosing their own problem. “In our experience, it is rarely the AI model that is failing enterprises,” he said. “The bigger challenge is closing the gaps between systems, data, intelligence, action and outcomes.”
The pattern, he explained, is familiar. HR, legal, finance, sales and customer service each run their own pilots, many of them unconnected. “Every pilot can end up building its own foundation, and organisations eventually reach a point where they ask, ‘Now what?'” Anand said. That is the point where many experiments fail to move into production.
The result is a set of siloed deployments in which copilots and agents cannot share context, collaborate across functions or execute work reliably across enterprise systems. “The outcome enterprises want lives in the handoffs between channels, agents, systems and people,” he said.
There is also a human factor. According to Anand, “Many organisations are comfortable with AI making recommendations but are still building confidence around allowing AI to take actions in real time.” That is where trust, security, compliance and governance become critical.
A context layer, not a rip-and-replace
Anand’s prescription does not involve replacing existing systems or launching a multi-year data transformation. Instead, he recommends adding an intelligence layer that connects and interprets data already spread across CRM, contact centre, communications, commerce and operational platforms.
The layer has three parts. An ontology creates a common business language, defining entities such as customers, products, interactions, channels and outcomes and the relationships between them. A knowledge graph connects relevant information across systems. A context graph adds the real-time dimension, mapping events across the customer journey and interpreting them against the customer’s history, intent and preferences.
Together, they let an enterprise understand not only who a customer is, but what is happening now and what action is most appropriate. Anand calls the goal a “Segment of One,” in which each individual receives an experience tailored to their own context rather than being placed in a broad segment.
Tata Communications has been building this capability following its investment in Commotion, he said, and combining it with the communication channels gained through Kaleyra. His advice to enterprises is to start small: focused co-creation initiatives and targeted proofs of concept on priority journeys, then scale as value is demonstrated.
From AI that answers to AI that acts
Context alone does not change outcomes. Anand describes the shift from reactive to proactive engagement through what he calls the “3 Ps”: proactive, personalised and predictive. The third ingredient, beyond unified context and real-time data access, is agentic orchestration.
“Customer experience is moving from AI that answers to AI that acts,” Anand said. “Anticipation alone is not enough. The ability to execute a meaningful next best action, update systems and complete workflows is what creates business value.”
He pointed to insurance, where customers traditionally call about renewals, upgrades or unused benefits. With predictive AI, the insurer can reach out first.
Orchestration also addresses the handoff problem. It determines which system, channel, AI model, workflow or human should respond next, based on real-time context and business rules. Done properly, a move from WhatsApp to voice AI to a human agent should not feel like starting over. “A seamless transition therefore involves more than transferring a transcript or recording,” Anand said. “It transfers understanding, responsibility and momentum.”
How much can really be automated?
Anand is candid that full automation remains a moving target. Many organisations begin with a goal of 20% to 40% of interactions handled by AI agents, depending on the use case. In some environments, he said, Voice AI agents have handled around 30% to 40%.
He does not expect that figure to reach 100%. “I don’t think we will reach a point where everything is automated,” he said. He suggests assessing tasks on consequences, complexity, reversibility, policy sensitivity and the value of human judgement. Low-risk, repeatable work suits automation, while high-impact or relationship-defining decisions should stay human-led, with AI providing recommendations and context.
“Hence, the future is really about collaboration, where AI handles routine interactions efficiently while human agents focus on more complex or higher-value situations,” he said.
The contact centre, in turn, is changing role. Quality teams have traditionally reviewed only a small sample of calls. AI can analyse 100% of interactions across channels, surfacing sentiment, intent, emerging issues and moments that call for supervisor intervention. In Anand’s view, this turns the contact centre into a source of real-time intelligence rather than a reactive support function.
Measuring what matters
Anand also wants enterprises to rethink how they judge AI. Productivity gains are not enough, he said. Success should be measured against outcomes such as first-contact resolution, conversion, retention, repeat-contact rates and customer satisfaction.
“Instead of asking whether the AI responded well, we should be asking whether it solved the customer issue and whether it achieved the outcome for both the customer and the business,” he said.
The stakes are commercial. Forrester has noted that even a minor improvement in CX quality can reduce churn and increase share of wallet. Anand adds that in competitive, thin-margin markets, a single poor experience can push a customer toward another brand.
Asked what he would tell a CIO and chief customer officer starting out today, Anand named three decisions.
The first is to resist scaling pilots. “The goal should be to scale a repeatable operating model,” he said. He advises starting with one or two economically important use cases with measurable outcomes, clear business ownership, baseline metrics and stage gates, then moving from human assistance to bounded automation and broader orchestration as confidence grows. The approach echoes Gartner’s advice to pursue agentic AI only where it delivers clear value.
The second is to architect for real-time execution rather than analytics alone, connecting insight to the systems that can act on it.
The third is to make responsible AI foundational. “Personalisation cannot scale without trust,” Anand said. Security, privacy, compliance and governance need to be built into the architecture.
Giving AI agents an identity
That last point leads to what Anand sees as one of the defining issues of the coming years: accountability. As agents take on more actions, organisations will need to know which agent made a decision, under what circumstances and with what authority.
“AI agents will increasingly need identities, appropriate access controls and audit trails, similar to the accountability mechanisms we have for human employees,” he said.
What the next five years look like
Looking ahead, Anand expects customers to stop repeating themselves, and agents to ask less and guide more. Engagement will shift from campaign- and transaction-driven interactions to relationship- and journey-driven ones, and from post-event analysis to predictive, real-time decisions. AI workers will coordinate with human employees and enterprise systems in a shared operating environment, and the focus will move from automating individual tasks to orchestrating end-to-end business outcomes.
Across industries, he sees the same direction. Insurers are exploring Voice AI for renewals and claims and for proactive outreach. Airlines are orchestrating voice, WhatsApp and booking systems into single journeys. In HR, AI agents are drafting job descriptions, screening profiles and scheduling interviews. And through a collaboration with Tata Tele Business Services, Tata Communications is working to bring AI-powered voice agents for enquiries, appointments, follow-ups and orders to small and midsize businesses.
For CIOs, the message from both the analysts and the practitioners is the same. The technology is advancing quickly, and the ability to connect it is not keeping pace. Customers will not reward companies for deploying the most pilots or the most sophisticated models. They will reward the ones that know them, anticipate them and finish the job.
The enterprises that close the gap between intelligence and action will define the next decade of customer experience. The rest, if Gartner’s forecast holds, will be among the projects that never make it out of the pilot.