3 Ways to Move AI from Pilot Mode to Performance

• July 28, 2026

By Banwari Agarwal

With decades of customer and risk data at its fingertips, the insurance industry is positioned to be among the biggest beneficiaries of artificial intelligence (AI), according to BCG. But when new AI tools are layered onto fragmented legacy processes that were never designed for speed or scalability, AI can end up amplifying the same operational friction that the insurance industry is trying to eliminate.

Integrating AI into existing applications is proving difficult across industries. A Gartner survey found that 77% of engineering leaders consider embedding AI capabilities into current systems a major challenge. In insurance, where processes often span underwriting, distribution, service, claims and multiple legacy platforms, embedding AI into that operational complexity is especially difficult.

Applying AI to isolated tasks often creates localized improvements without improving the overall experience. AI may speed up customer routing, for example, but if requests still move through disconnected service processes afterward, delays and friction remain. The result is isolated gains that struggle to scale into meaningful service and growth improvements.

Over time, this disconnect limits both the impact and scalability of AI initiatives. Although AI can accelerate individual tasks, if the broader operating model remains fragmented, business gains plateau.

A better strategy is to focus AI efforts on a handful of workflow chokepoints where delays directly affect business outcomes. Narrowing the scope makes it easier to identify whether AI is actually improving performance before a broader rollout.

Hiscox recently released the results of its AI implementation. The company focused on understanding where brokers and customers were experiencing friction, then redesigned those workflows before layering in AI. It reported improvements in how information moved through quote-to-bind processes and in using customer interactions to inform service and distribution decisions. The insurer achieved gains in quote speed and first-time quote quality.

Here are three ways your organization can take a more disciplined approach to implementing and scaling AI:

1) Start with the business problem, not the technology. While it’s easy to get excited by new AI capabilities, launching pilots without a clear connection to workflow bottlenecks makes it difficult to assess return on investment (ROI). The priority should be to look at processes end-to-end and identify where manual work creates friction. Only then should AI be applied against clearly defined business outcomes with measurable success metrics.

Starting with business realities—what customers need, where brokers experience delays and which workflows teams can realistically absorb change within—makes it easier to tie AI initiatives to measurable outcomes.

2) Automate repetitive tasks, preserve human expertise. Insurance still runs on trust and expertise. Automating every interaction can create as many problems as it solves, especially in claims and complex underwriting scenarios where customers expect human judgment and empathy.

The better use of AI is behind the scenes, reducing operational friction while keeping people central to relationship-driven interactions. By automating submission handling and information routing, underwriters can spend less time organizing information and more time focused on judgment-driven decisions and broker responsiveness.

When captured and analyzed effectively, those customer interactions can become a valuable source of insight, helping insurers better understand customer needs and make more informed service, underwriting, and distribution decisions over time.

Faster handoffs and cleaner workflows don’t just improve efficiency, they improve the outcomes customers and brokers ultimately care about: quicker decisions, better service, and greater confidence in every interaction.

3) Scale through controlled experimentation. One of the fastest ways to stall an AI initiative is scaling it across multiple departments before teams trust the process or governance is fully in place. When insurers move too quickly, it becomes difficult to determine which changes are actually improving performance.

Avoided large-scale rollouts in favor of tightly scoped use cases, testing high-frequency, low-complexity workflows before expanding them further. Along the way, pause to refine processes and measure operational gains before scaling into additional workflows and lines of business.

Just as importantly, governance should be built in from the start. Legal, compliance, operations, data and technology teams need to be involved early, keeping humans in the loop so oversight evolves alongside the AI rather than after deployment.

The goal isn’t to launch the most AI projects. It’s to scale the ones that have already proven they can improve workflows in measurable, manageable ways.

Banwari Agarwal is CEO of insurance, banking and financial services at Sutherland Global.