The risks and rewards of hyper-personalised insurance models

Insurance, at its core, is a simple proposition – the pooling of risk to replace catastrophic personal losses with structured compensation payments from a group pot of funds.

Naturally, it makes sense to weight individual contributions to this funding pool by risk. A homeowner with a one bedroom house on a hill would not expect to pay the same property premium as a homeowner with a four bedroom house built on a floodplain, for example.

Now, thanks to advances in data enrichment and artificial intelligence (AI), the detail and accuracy of risk assessments underpinning pricing decisions is growing ever more granular – and many industry commentators agree that highly personalised pricing is a fairer reflection of risk profiles, rewarding careful policyholders with cheaper cover.

There is appetite from consumers for greater personalisation too.

For example, a 2023 survey of UK insurance policyholders conducted by global technology and management consultancy Capco found that 28% of respondents would share additional personal data in order to get more personalised services and products.

The consultancy’s associated report, entitled Hyper-personalisation: A new era of experience for insurance customers, noted: ”Hyper-personalisation represents a transformational shift in how insurers can interact with their customers.

”By harnessing the power of technology, data and human-centric design, they can move past today’s one-size-fits-all digital experiences to a future where every interaction reflects and resonates with an individual customer’s needs and aspirations.

”It is about forging deeper and more meaningful connections, especially with vulnerable groups, for whom tailored experiences can mean the difference between financial exclusion and empowerment.”

There is, however, a growing concern among insurance professionals that careful attention must be paid to this process in order to avoid two unintended consequences of personalisation – discriminatory profiling and the shallowing of the risk pool.

So, what are the risks and rewards of building data led underwriting processes and how can the industry navigate these challenges and opportunities?

Rewarding good risk

The benefits of accurate data are clear – ‘good risk’ can be rewarded with lower premiums, while ’bad risk’ can be appropriately quarantined.

Methods like vehicle telematics, flood risk mapping and smart home sensors provide individual risk assessments, where the behaviours and choices of a policyholder are directly reviewed and their actions then informing policy pricing.

Dan Cicchetti, associate vice-president of client engagement, insurance for UK and Ireland at data analytics company LexisNexis Risk Solutions, explained how data enrichment can benefit insurers and consumers alike.

He said: “Data enrichment is fundamental in supporting a great customer experience, accurate pricing and ensuring the product offered is right for the needs of the individual.

“If an insurance provider has had a past relationship with an individual, [it] should be able to better engage with that person at any stage of the customer journey, leveraging that knowledge.

”This is good for the insurance provider, as that knowledge might provide a unique insight that allows for the correct product, premium or journey but [it is also good] for us as consumers to feel more valued.

“There are also so many nuances when it comes to buying the right insurance product. Data enrichment is a key enabler to providing the right product for a consumer, regardless of their familiarity with those nuances.”

Segment-based bias

The main risk of hyper-personalised, AI-based underwriting is – as many see it – the chance of discriminatory policy pricing.

Segment-based risk assessment, in contrast to individual risk assessment, uses descriptive factors to place consumers into narrower risk buckets, without necessarily reflecting any one individual.

This may involve the use of information such as employment type, education level, political affiliation, social media activity, crime statistics in a policyholder’s postcode and even grammar and spelling in online applications.

Some market professionals report concerns that black box AI tools may use these proxy characteristics to segment customers by gender, age, education, race and other protected characteristics, ultimately leading to exclusionary or bias policy pricing.

James Daley, managing director at Fairer Finance, commented: ”As things stand, there’s no transparency around what goes into each insurer’s pricing algorithm and most industry participants believe it would undermine their competitive position if they had to disclose this.

”From the work we’ve done with insurers, I know that some have fairly simple algorithms, sticking to basics like postcode, age and vehicle. Others use a whole range of data – from occupation to credit history, to loyalty card data.

”By refusing to reveal what goes into their pricing algorithms, insurers leave themselves open to accusations of discrimination.”

For example, back in 2022, charity network Citizens Advice coined the term “ethnicity penalty” after its research found that in 2021, people of colour paid on average £250 more for their car insurance compared to white drivers.

At the time, Citizens Advice chief executive Dame Clare Moriarty said: “The use of algorithms has real-world implications for real people. They must be applied with caution, under the careful scrutiny of regulators.” 

However, Cicchetti told Insurance Times that the correct use of data could actually avoid unfair grouping – for example, in home insurance policies.

He said: “In home insurance, the granularity of data for perils risks, such as flood, and defining risk by address rather than postcode means that premiums can now be calculated based on a more precise understanding of the risk of the policyholder’s home, not their neighbour’s.

“This is crucial given a property in one street can have hugely varying degrees of risk – especially if they are on a hill when considering flood risks.”

Societal value

Although red flags around discriminatory pricing are clear, the march to hyper-personalisation raises another question for market participants – in an industry that prides itself on the societal value it provides, to what extent are insurers responsible for ensuring a service for all?

For the FCA, for example, the “pooling of risk is critical to delivering insurance for those most at risk”. The regulator therefore worries that hyper-personalisation could lead to “a world where those consumers are priced out of the market”.

Furthermore, Stere chief executive Dogan Kaleli emphasised that AI driven hyper-personalisation must be interlinked with “traditional underwriting models” in order to be successful.

He explained: ”Hyper-personalisation can be a game-changer for insurers looking to drive customer growth. But, of course, whatever we do in terms of customer engagement will have an impact on our underwriting and pricing models. Without the marrying these two concepts –implementing AI tools but still using traditional underwriting models – [this process] will lead to disappointment.”

The answer, as is often the case, may lie in the middle ground – data driven pricing that rewards customers’ care and attention, but with cross-subsidisation schemes between industry and government for those that risk being priced out of the risk pool. This should be delivered with a degree of transparency that prevents unfair segmentation.

Speaking at a conference in 2019, FCA consumers and competition executive director Sheldon Mills said: “Consumers should be able to see where their data is, understand with whom it has been shared and retain the power to easily revoke their consent at any time.”

Likewise, Cicchetti sees a greater risk in poor data, rather than more data. He explained: “The risks can lie in overreliance on data that may be incomplete, outdated, or incorrect, which can lead to flawed decisions.

“Algorithms used to support pricing decisions need to be explained and tested for fairness and models reviewed for discriminatory outcomes using fairness metrics. Human oversight remains crucial in decision-making, particularly as we see the explosion of artificial intelligence entering the insurance landscape.”