Scaling AI for Customer Experience in Insurance: A Leadership Playbook for SEA

As AI presents new opportunities, it also presents new challenges. CIOs are expected to balance risks against the benefits of innovation and transformation, such as technical debt, layoffs and IP loss. Beyond these pressures, leaders face structural barriers to scaling AI in the region.

Concerns about privacy and security risks
This risk remains as new regulations are introduced and cybersecurity threats create risks such as data breaches and unauthorized access to sensitive information. For example, in 2025, a ransomware attack on a Singapore data provider compromised the personal information of at least 146 policyholders, drawing attention to vulnerabilities in third-party data handling and operational partner risk management.

Data hygiene, data quality and integration constraints involving legacy systems
Many insurers still rely on legacy core systems with fragmented and low quality data, and lack of support for integration. These systems struggle to support real-time data access, modern APIs for integration, or scalable AI architectures. The result is complex implementations, longer TTV, and incomplete customer views. However, we observe that insurers are gradually prioritizing enterprise-wide data governance and readiness assessments before advancing AI programs.

Lack of AI talents and uncertain user adoption
AI transformation highly depends on talents and user adoption. Many CIOs are prioritizing the design of a human–AI hybrid workforce, where employees are augmented by AI agents and task-specific copilots. To do this, workforce planning is becoming equally important as AI reshapes role design and headcount models. Upskilling workforce with AI knowledge also becomes one of the key initiatives to consider. To sustain or increase user adoption, factors such as change fatigue, leadership alignment, concerns about role displacement, and unclear governance must be addressed accordingly.

There is no single blueprint for AI-led CX transformation in the region
Scaling involves expansion and likely, regional rollouts are involved. AI adoption is progressing at different speeds across SEA countries, shaped by market size, operating models, language structure, regulatory maturity, infrastructure, and workforce readiness. Notably, most SEA markets have published national AI strategies or draft governance frameworks, signaling that AI is a strategic priority. Comparatively, Singapore operates within a mature digital and regulatory environment, with the FEAT Principles established as early as 2018 and continued government investment under its National AI Strategy and Finance-focused AI initiatives. Indonesia, the region’s largest digital economy, is scaling AI across highly distributed customer and agency networks, supported by growing global cloud and AI infrastructure investment. In Thailand and Vietnam, language structure continues to add complexity in NLP deployment, even as national AI strategies accelerate ecosystem. These structural differences directly shape AI scaling potential across markets.

Finding the balance of human vs. AI for insurance moments that require empathy
AI initiatives struggle when insurers optimize purely for speed without recognizing the emotional nature of insurance moments. Claims tied to illness, accidents, or bereavement demand empathy as much as efficiency. In these situations, well-designed escalation pathways within a human–AI hybrid workforce are critical.

Rising customer expectations across segments puts pressure on insurers
Policyholders expect insurers to understand their needs and respond with timely, personalized interactions shaped by life events, medical history, financial context etc. Many benchmark their experience against digital platforms such as Grab and Amazon, yet most insurers are not structurally designed to deliver this level of consistency. At the same time, insurers must serve a wide generational mix, from mobile-native customers to older segments that still value human interaction. For example, Gartner states that “Gen Z and millennial customers are more than twice as interested in embedded insurance as other generations”^*1. Continuous effort is required to refine AI-enabled features in line with expectations.

Choosing the right partners is critical, as the broader insurance ecosystem adds complexity to scaling AI.
Customer journeys involve many parties such as agents, brokers, TPAs, bancassurance partners, and aggregators. Data fragmentation across these parties, especially where real-time integration is limited, frequently becomes a bottleneck that constrains AI effectiveness. Competition is also intensifying where mergers and acquisitions, partnerships, and digital-native entrants continue to raise expectations. Insurtechs, MGAs, and embedded insurance platforms often set new benchmarks for speed and convenience, forcing incumbents to respond.

Rather than spreading effort across the organization, insurers that successfully scale AI start with a few key areas where value is tangible, execution risk is manageable, and adoption can be sustained. Gartner stated that “98% of insurers plan to deploy AI by year-end 2027”^*2.

AI conversation is also moving beyond simple automation, where attention is on more complex and regulated processes. By year-end 2026, at least 25% of insurers will have implemented at least one multiagent system (MAS) targeted at back-office efficiency^*2. MAS allows orchestration of specialized AI agents across tasks such as intake, data validation, coverage verification, and fraud scoring.

Leading insurers start where the foundations already exist vs. enterprise-wide readiness. This means focusing on use cases that can deliver value within current constraints, while gradually strengthening data, integration, and governance over time.

Across SEA, AI adoption is most mature in claims, underwriting, conversational and 24/7 customer service, agent enablement, and fraud prevention. These areas tend to have high volumes, clearer risk boundaries, and quicker returns, Business outcomes are typically related to productivity, processing, and customer experience. They also use AI to support human judgment rather than replace them, which helps reduce resistance from frontline teams and regulators through better transparency and explainability. As a result, improvements are seen in areas such as claims, where insurers have reported processing time reductions of up to ~60% via AI .

Looking ahead, insurers are preparing the best ways to manage “machine customers”, where AI agents act on behalf of human customers. This trend is observed in B2B market on tasks such as verifying dental insurance or employee benefits. Gartner stated a “strategic planning assumption: By year-end 2027, 35% of web interactions in insurance will be initiated by machines instead of humans”^*2. As a result, insurers need to be ready to handle AI agents in inbound interactions across both sales and service. For example, call centers and websites should be able to receive inbound calls and text messages from AI agents and process service requests accordingly.

Other emerging features that are seen in the market include the adoption of domain-specific language model (DSLM), to improve accuracy of outcomes. DSLMs, trained on insurance data and workflows, are increasingly viewed as better suited to complex interactions than generic LLMs. According to Gartner, “through year-end 2026, over half of insurers will have added an insurance domain-specific language model (DSLM) to their list of LLMs deployed, up from the 36% using one at year-end 2025”^*2.

Based on press releases and publicly available information, leading insurers across SEA are actively embedding AI across both the customer and operational lifecycles. Key use cases include underwriting, claims processing, exception handling, customer servicing, sales agent enablement, health claims adjudication, fraud detection, conversational AI, and chatbots. While some initiatives remain in pilot or experimentation phases, others have progressed to enterprise-scale production deployments and are delivering measurable business outcomes. The most common outcomes influence key productivity and CX indicators, including average handling time, first contact resolution (FCR) and overall CSAT.

The next step after selecting use cases is to translate prioritization into an effective AI roadmap that optimizes sequencing, scaling, and stopping decisions over time. Figure 2 outlines the key actions to consider when developing an AI roadmap.

Sequence prioritized use cases across a realistic time horizon, linked to key customer moments and business impact
Sequencing allows insurers to capture value early and avoid overloading the organization. In the first 3 months, the focus should be on high-value, high-feasibility quick wins. Subsequently, efforts can expand into secondary domains, broader transformation extending across products, channels, markets and ecosystem partners, and incorporating more advanced capabilities.

Fix foundations before adding AI
Poor data, unclear processes, or limited adoption can jeopardize AI returns. Here are a few core foundations to have in place prior to AI implementations:

• Data and Technology readiness: Data must be accessible, reliable, and reusable, with clear ownership and controls. Platforms, integrations, and APIs should support AI use cases as they grow into other facets.
• People, skills, and change readiness: Business users and technologists need to work together in hybrid operating models. Change management is key to address resistance and ensure new ways of working stick.
• Operating model alignment: Clear roles, processes, and decision flows are needed so AI outputs are trusted and acted upon in daily operations, especially in customer-facing teams where AI impacts customer experience.

Strengthen governance: Trust, transparency, and guardrails
AI governance works best when aligned with existing enterprise decision structures. Effective AI governance typically addresses people, technology, risks, and data elements:

• Clear ownership and decision rights: Define roles for model ownership, validation, approval, and audit across business, risk, and IT – whether governance is centralized, distributed, or hybrid. According to Gartner, “42% of insurers report that IT has primary responsibility for enterprise-wide decision-making regarding AI, but business leaders can make independent decisions within those standards”^*1. Collaboration between the CIO, CDAO, CISO, heads of compliance, and operations leaders will be essential in deploying effective AI programs that provide optimal business value while managing risks.
• Explicit scale and stop authority
• Embedded CX, fairness, and accountability metrics to ensure continuous safe use of AI and awareness of risks associated with AI
• Ongoing monitoring and auditability should align with standards and regulations

Develop scaling strategy and AI operating model: From pilots to enterprise platforms
Scaling AI demands a deliberate operating model that governs how AI is built, deployed, and managed at scale. This spans people, process, technology, and risk oversight. Leading insurers avoid rebuilding similar capabilities by investing in shared, reusable enterprise components such as document intelligence, conversational AI, fraud detection engines, and customer analytics platforms. These assets create consistency, reduce duplication, and improve cost efficiency, while still allowing localization for language, regulatory, and market requirements.

Define leadership KPIs and reporting
Executives should oversee AI performance via a balanced scorecard to align initiatives with expected outcomes, covering these dimensions minimally: Is AI creating measurable business value? Is it improving customer and agent experience? Is it meeting trust, risk, and compliance expectations? Is user adoption scaling beyond pilots?

Knowing when to stop is a key part of how sustainable AI advantage is built. Insurers should consider pausing or discontinuing AI initiatives when:

• CX or ROI targets are consistently missed within agreed timelines
• Persistent data quality issues undermine model reliability
• Regulatory or explainability risks become unacceptable
• Adoption remains low despite targeted training and change efforts