Enterprise AI in customer experience is moving, but not in the neat, linear way the hype cycle suggested. Health Insurance contact-centres are a complex system of people, processes, technology, and data, governed by complex and slow-moving regulatory and compliance regulations. Whilst 88% of organizations report AI use in at least one function, nearly three-quarters of projects remain in the experimentation or piloting phase.

Over the next 18 months, the initiatives that scale will be those that treat AI as an operating model upgrade, not a tech demo. To get this right, leaders must think organization-wide with AI deployed in the right sequence, tuned to the realities of workforces and regulatory environments.

These are 5 trends that we believe will separate the winners from the rest

1) Optimizing for outputs

Successful operational leaders are moving past pilots for the sake of innovation and asking a harder question: what changed as a result?

Over the next 18 months, AI initiatives will be judged on measurable operational outcomes, not potential. That means clear baselines, defined success metrics, and controlled rollouts. Did QA coverage increase? Did compliance exceptions decline? Did conversion improve? If the answer is unclear, the project will stall.

The organizations that make progress will treat AI as an iterative program, not a lab experiment. They will measure, adjust, and redeploy in cycles, focusing relentlessly on outputs rather than capability. Initiatives that remain abstract, overly dependent on perfect data, or disconnected from operational metrics will struggle to scale beyond the pilot stage.

2) Human involvement as a feature, not a stop-gap

In highly regulated industries with complex, often nuanced rules, the most successful AI implementations will be those that maintain humans-in-the-loop well beyond human-assisted controlled pilots. The best automation will balance speed and scale with experience and empathy, amplifying human processing and judgement, not simply replacing it.

Over the next 18 months, the winning implementations will see AI do broad processing at scale. Humans-in-the-loop will remain, to validate exceptions, calibrate scoring, handle disputes, and provide feedback to refine rule sets. This is not a temporary compromise; it is how organizations will continue to ensure trust and accountability, reduce the risk of errors, and produce defensible outcomes for audits and oversight. The most effective teams will formalize human-in-the-loop as a system, with clear workflows for review queues, calibration sessions, version control for scorecards and rules, and continuous improvement cycles.

3) Subject matter expertise beats generic capability

The best AI implementations won't be one-size-fits-all solutions, especially in regulated industries with a high degree of domain specificity. In highly regulated sectors like insurance, compliance is a complex, evolving set of rules, not a static checklist. Tools that understand and optimize for this complexity will win.

Medicare call centers operate under tight constraints that shape every interaction. Disclosures vary by call type, scripts change by plan, and not all deviations carry the same risk. AI only adds value when it understands that context natively. The most effective tools will interpret conversations through a Medicare-specific lens, distinguishing material compliance failures from acceptable variation and mapping outcomes directly to CMS expectations.

Specialized systems are also better positioned to establish best-practice quickly and accurately. By learning from compliant, high-performing interactions within a specific regulatory environment, specialized AI tools can reduce variance and standardize what "good" looks like. Once these insights are collected, they can be pushed back into operations through clearer scorecards, more consistent QA, and targeted coaching. In regulated environments, depth of understanding is what turns automation into efficiency rather than noise.

4) Data quality and workflow design become the real competitive advantage

Most contact centers already have data. The differentiator will be whether it's structured, consistent, and collected within workflows people actually follow.

Over the next 18 months, organizations that thrive will be investing more in:

This is the unglamorous truth of the next phase: AI's value compounds when the operation is disciplined enough to absorb it.

5) AI as connective intelligence, not isolated tooling

Even when organizations get measurement right, embed subject-matter depth, and design disciplined workflows, the impact of AI will still remain limited if confined to individual teams.

The best implementations will view AI as a shared intelligence layer across the organization with insights generated in one function helping inform decisions in another. For example, QA findings should influence training priorities, compliance patterns should shape script revisions, and sales performance signals should refine workforce planning.

In highly regulated environments, this horizontal alignment is especially powerful. When domain-specific intelligence flows across teams rather than staying within them, organizations reduce variance, tighten risk control, and move faster with confidence.

The real shift is from optimizing individual functions to strengthening the coherence of the whole system. Leaders who treat AI as connective infrastructure, not a collection of siloed capabilities, will create compounding gains.

Final thoughts: Operational discipline will win the day

In 2026, the difference between leaders and laggards will be as much about execution discipline as it is access to frontier AI technology. Leaders who choose the right sequence of adoption, who start with use-cases that are easy to measure, and who build operational trust before expanding, will see the best results. Investing in governance and change management will also be crucial, because even the best AI fails if teams do not adopt it or have the right process to train and act on its outputs.

If you want a simple way to stress-test an AI CX initiative, ask two questions: Can we prove impact on a metric the business already cares about, and do we have a trustworthy process for validation and continuous improvement? If the answer to either is no, it is probably still a lab project.