A tougher environment in many insurance markets means that insurers must achieve more with the same – or less – resources. Simply cutting costs is not enough because customers are demanding more rapid, convenient, and tailored offerings as well as greater value for money. Meanwhile, insurers that fail to leverage new technologies such as GenAI could find themselves at a strategic disadvantage. 

Here we explore how insurers can respond to three top industry trends: the push for operational efficiency in life insurance; putting the fundamentals in place to allow personalization at scale; and identifying robust use cases to exploit the remarkable potential of GenAI.

An unsettled macroeconomic environment – in which customers are focusing on value for money and  insurers have made commitments to investors on cost cutting – is making it imperative for insurers to deliver operating efficiencies. This contrasts with recent years, when fast-expanding APAC economies and low insurance penetration rates led many insurers to focus on customer acquisition and top-line growth. 

To support that growth, many insurers near-shored back-office work to take advantage of labor arbitrage in the region. This use of lower-cost labor meant genuine process and technology transformations could be deferred as the costs of inefficiency and manual workarounds were minimized. 

This has stored up a wealth of opportunities for insurers to rethink their operating models, transform the way they do things, and exploit new technologies, automation and AI to cut operating costs and drive efficiencies. However, this kind of transformation is not necessarily easy because a tangle of legacy systems and processes must be teased apart, interrogated, redefined and rationalized before genuine straight-through processing (STP) can really be achieved.  

Simple solutions are sometimes available, but often the main challenge lies in understanding legacy undocumented processes and rules that are baked into disparate systems. Underwriting rules, for example, are often still housed predominantly within policy administration systems (PAS), hard coded over the years and often not revisited or documented. 

Understanding these rules, challenging their continued validity and analyzing trade-offs in risk appetite, customer experience and the cost of further investigation/adjudication can lead to significant improvements in STP and turn-around time. 

Surfacing and addressing this kind of issue is not just important for securing high rates of STP. In sales terms, it can be the difference between booking a sale while the agent and customer are sat together, or delaying for days and weeks while the customer loses interest and the agent grows disenchanted with the insurer’s performance.  Similar challenges appear when automating other insurer activities such as new business and claims adjudication. 

Life insurers can prioritize investments through 2024 by considering:  

• The kind of customer journeys they want to offer, particularly around key ‘moments that matter’ such as the point-of-sale experience and claims handling 

• The current state of their processes, including surfacing how underwriting rules and adjudication routines might impact STP initiatives (and how to mitigate this) 

• The balance between the cost of keeping potential exceptions within STP (e.g. underwriting risk, illegitimate payouts) versus the cost of managing exceptions outside STP (e.g. sales impact, poorer response times, staff time for investigations).

Insurers that do their homework around these key issues can identify the projects with the largest operational cost savings more accurately. They will also be able to select investments with the most impact on the ultimate customer and agent experience.

Capco’s recent insurance survey showed that Asia-Pacific policyholders are prepared to share the necessary personal data to gain more personalized products and services. For example, nine in ten policyholders in Hong Kong (94%) and the wider GBA (90%) would be motivated to share additional personal data with insurers in return for a range of benefits including more personalized services, cheaper premiums and enhanced claims processes. 

The results are similar across other surveyed markets such as Singapore, Thailand and Malaysia. Respondents also say they are open to new ways of sharing personal data such as wearing wireless smart devices and using smart devices in the home.

However, there is a wide range of maturity among insurers in terms of how prepared they are to tailor offerings around individual customer needs and lifestyles. While most insurers have altered their mindset towards becoming more customer-centric, the legacy systems they are using and their siloed distribution channels are often preventing them from making this strategic shift. 

For example, policy administration systems are predominantly still organized around policy numbers (sold products) rather than customers (individuals). Definitions of customers often continue to be unclear – for instance, does the insurer include policyholders, beneficiaries, and persons insured? Meanwhile distribution channels and associated compensation structures can promote the hoarding of customer data. As a result, insurers may even be uncertain whether multiple policies in similar names are owned by the same customer, or not.

Building a single view of the customer (SVC) is therefore a fundamental first step towards crafting great omnichannel experiences, gaining deeper knowledge about customer lifestyles and goals, and honing the ability to cross sell and upsell. It means pulling together the data from disparate systems, putting in place the right data de-duplication processes, and focusing on enhancing the quality of the data that the insurer possesses to make sure any personalization activities are based on verified, quality information.

As a more complete picture of the customer emerges, insurers also need to map the type of data they have today against the data they require to become more customer-centric and fulfil their future business goals. What additional data variables does  the insurer need to collect, and what is the best way to do this, e.g. via customer outreach, innovative engagement strategies, or new partnerships? 

Finally, as data becomes more insightful and valuable, putting the right data governance in place becomes even more important. The kind of data that can improve customer outreach needs to be accurate, up-to-date, and secure as well as conforming to data privacy standards. With the right planning, however, 2024 could be the year that more insurers turn their aspirations to become customer-centric into a practical reality.   

Insurers are excited about the opportunity to use generative AI (GenAI) to automate portions of the insurance value chain including back- and front- office tasks. Examples include helping customers and advisors to understand which risks are covered under their existing policy, and supporting sales agents to sell the right products to customers. There is tremendous scope to use new GenAI tools to improve customer, employee and advisor experiences at the same time as reducing operating costs. 

However, the new technology arrives with potential risks and downsides, which must be balanced against potential rewards if insurers are to identify the best early use cases. GenAI risks include concerns around customer privacy and data protection, such as making sure GenAI assistants don’t learn from customers’ personal data without their permission. Insurers also need to make sure bots do not give users inaccurate or wrong answers due to biases in the data, or because the bot is ‘hallucinating‘ and making up answers that are nowhere to be found in the training data.

Governance frameworks are emerging to help insurers defuse these and other risks, and these can be combined with a handful of early industry learnings to help insurers identify the most tractable GenAI use cases on a risk/reward basis:  

• Pick early use cases where humans can easily validate outputs of the GenAI model/bot. Consider, for example, a project to use GenAI to help call center staff rapidly ascertain whether a risk is covered by the insurer’s policy documentation. This can be validated by making sure the model links each output to policy wording in the underlying policy documents. 

• Provide the AI model with a specific purpose, a relevant database to learn from, and guardrails. Bots should be designed around a specific purpose and data set. For example, the bot in our policy documentation example will need a careful definition of its task – including guardrails that define how it should respond and any questions that it should deflect. Clearly the bot will need the insurance policy documents to train on, however, the phrases that customers use to describe their claim may not be the same as those in the policy wording – underlining the need for thorough testing and model augmentation through techniques such as vectorization. 

• Consider the data set you will need for your use case and any associated challenges. Insurers need to impose  strict governance on the data employed in GenAI projects to ensure compliance with privacy regulations and other rules. Some use cases will be relatively easy to greenlight because they involve data that is inherently less problematic such as publicly available data (as opposed to sensitive personal data). Early in every project, insurers must review and understand the data that will be supplied to the bot, considering whether this data might introduce biases to the model output that need to be tested for and mended. 

• Always include prompt engineering and test, test and test again, to discover whether certain prompts can make the model hallucinate, cede control to users, or break through its guardrails. Prompt engineering should be conducted by the best developers in the field, who know how to use GenAI model logic to circumvent standard safeguards. Testing may also involve checking bot answers for evidence of bias, and cross-validation versus the outputs from different kinds of GenAI models.

• Finally, bake in your compliance and transparency approaches at the start of the GenAI project, rather than layering them in at the end. This is safer and means that successful early use cases can be scaled with more confidence. 

In their earliest use cases, insurers will want to keep a ‘human in the loop’. With the right training, skilled back-office professionals can apply their experience to identify unexpected model outputs and ensure any issues are reported and addressed. For customer-facing bots, the conversations (inputs and outputs) need to be monitored and reviewed to ensure the bot is performing as expected.

Care when specifying the purpose of the bot, choosing high-quality training databases, and ensuring excellent prompt engineering are often the keys to success. Implementing early use cases in a rigorous way will also help clarify where insurers need to improve fundamental capabilities – particularly around data governance and the automation of this – to take full advantage of GenAI’s remarkable potential. 

 
References

Applied Generative AI Governance: A Viable Model Through Control Automation, Gerhardt Scriven et al., Capco Institute Journal of Financial Transformation – Artificial Intelligence, November 2023, p.24. Link.