In brief
- Before you go all out for AI, ask yourself, “What actually needs doing and what else could be done?” It might make more sense to explore the possibility of getting higher value from existing resources rather than diving headfirst into a novel technology with an epic degree of risk
- There's a great deal of innovation in the insurance space. However, across the value chain, few firms have gone on to become full-stack insurers (i.e., doing everything). Managed General Agencies (MGAs) are an example of insurtech businesses that support parts of the value chain with underwriting being handled by the primary carrier
- Pricing professionals are getting more involved with changing the tech stack, using new data and fresh approaches to modeling. But building an ML model or solution is only part of the issue. Embedding it into core business systems, getting it used and having it deliver value while complying with all the regulations and technical drift/debt, etc., is a much more convoluted affair
I've made a point of picking up my promotional backpack, branded thumb drive and novelty notepad while wearing my delegate/speaker lanyard with pride for more years than I care to remember.
I used to complain that wherever we gathered, it was the same old faces, the same old products and the same old innovation. The same old story, in fact. But I get the distinct impression that, nowadays, insurance get-togethers have really upped their game, especially since the emergence of ChatGPT and generative AI (GenAI).
GenAI and insurtech insights
In Zoreza Global’s latest podcast, I asked Sandeep Karkharnis about the 2 days he spent at March’s Insurtech Insights Conference in London. Did he witness a flood of relevant GenAI use cases in the insurance sector?
The Head of Machine Learning Strategy and Delivery for Allianz Personal Lines explained that GenAI could have a considerable impact on commercial lines insurance, adding that asset management companies are already using GenAI quite a bit. However, before insurers go all out for AI, there’s a case for asking, “What actually needs doing and what else can be done?”. Might it make more sense to explore the possibility of getting higher value from existing resources rather than diving headfirst into a new technology with an epic degree of risk?
He pointed out that a major airline was fined when GenAI or a chatbot gave the wrong information to a customer who had to be compensated. In a 2024 Insurtech Magazine survey, 34% of respondents said they’d be reassured by the option to refer Al decisions to a human. Therefore, most financial services companies are likely to follow suit, preferring to limit AI to internal applications where they can keep humans in the loop and retain control instead of favoring external-facing apps and all that it implies.
Experimentation is a question of ethics
I found that fascinating because I believe certain newspapers are suing GenAI platforms for copyright and plagiarism. There's a tricky challenge around ethics, too. How can you be sure that GenAI makes not just the right decision but also the right ethical decision? Is the insurance sector waiting for those issues to be resolved or plowing on regardless with experimental use cases?
Although there’s a lot of experimentation, Sandeep suggested he’d be surprised if the new initiatives were put into production and achieved value (saving or making money, saving time, etc., at scale). Executing a small POC is very different from trying to deliver benefits at scale. Top-tier strategy houses have done a fantastic job of generating hype around AI. Media noise has captured C-suite attention, but the reality is always different one way or another.
Delivering ML benefits at scale
He feels that we’re not at the level of Artificial General Intelligence (AGI) yet — in most cases, we’re really talking about ML. In fact, Sandeep redefines AI as “augmented intelligence” — solutions that keep humans in the loop. The claims, pricing and underwriting divisions differentiate insurance from most other industries, and ML solves real-world problems across these three areas.
Take automotive telematics, he said. “I haven’t seen much evidence of scaled benefits. In my experience, it's used to encourage better driving behaviors, especially in the UK. Insurance costs for young motorists are sky high, so a mechanism that rewards good driving will continue to pay back over time as the drivers mature and hone their skills.”
Too big data
Zoreza Global does a lot of work with the automotive industry and vehicles where onboard computers collect masses of data traditionally held in a black box. One of the challenges of preventative maintenance and so on is that the sheer volume of data is difficult to transmit and use in real-time. So, I referred my podcast guest to the issue, “From an ML perspective, isn’t having too much data the actual industry challenge to be addressed?”
Yes and no, he said. Unless the black box was added by an insurance company, a lot of driver telematics data are held by the car maker. I don't think you get a lot of usable information from it, but you can see why so many insurers are desperate to partner with motor manufacturers.
Better means bigger bills
That said, vehicle technology has improved. We now have front and back cameras, ADAS and a mass of innovative systems, which means cars are becoming more advanced and people receive more alerts. Consequently, having an accident costs more, partly because the infrastructure is built in such a way that replacement parts are more expensive. However, things like theft are being reduced — my new car sends me a warning message if I forget to lock it.
Driver and passenger safety has improved, and the likelihood of accidents has lessened, but aftermarket parts and running costs have risen considerably due to the complexities of war, COVID-19, inflation, the scarcity of semiconductor chips and supply chain problems. A one-time £100-£150 windscreen replacement could now cost up to £1000, a 600% increase because of the sensors embedded in the glass.
The cost of disruption
I reflected on the impact of new technology, such as Tesla’s electric vehicles or digital transactions. It was assumed that the raft of new disrupting banks would take over the banking world, too, but that doesn’t seem to have happened. So, I asked Sandeep about similarities in the insurance space. Were insurers concerned that even more disruptors could be on the horizon, or are the big insurance companies still agile enough to see off the competition?
Bracketing banking and insurance is like comparing apples and oranges, he said. For most of us, banking interactions are transactional. We login to our bank accounts regularly to check balances and transactions. And because they have to support frequent interaction, by default, the banking sector is streets ahead of the insurance industry in terms of digital transformation.
Insurtech support
In reality, there's a great deal of innovation in the insurance space. However, across the value chain, few firms have gone on to become full-stack insurers (i.e., doing everything). Managed General Agencies (MGAs) are an example of insurtech businesses that support parts of the value chain, underwriting being handled by the primary carrier.
Then there are disruptors who simplify elements of the value chain, transcribing PDF documents and reading/extracting information, with another company providing the underwriting work ventures, post-claims, etc.
Most insurtechs (especially in Europe and the UK) focus on the distribution element of the value chain. Comparison sites, a massive disruption, have no equal in banking (in many markets, they don't exist).
What about “uberization”
I wondered why we hadn’t seen the uberization of P&C insurance. Sandeep reminded the listeners that very few disruptors have progressed to full-stack carriers. He believes that those that have, have great marketing engines, but if you look at their loss ratios, etc., they’re nowhere near as successful as they claim to be.
Great timing is the prime factor required for significant disruption. Doubtless, it will happen for insurance someday, but timing is one of the industry’s most stubborn challenges.
Daily or annual customer contact?
I’ve always been puzzled that insurance has a 12-month sales cycle while banking is sold based on daily transactions. Could Sandeep ever see insurers changing their product cycles to generate more customer action?
He replied that there are instances of usage-based insurance (UBI), but customers don't seem ready for that kind of change in any significant numbers. Insurtech Digital’s survey shows that UBI adoption decreased from 14% in 2023 to 10% in 2024, in direct contrast to the number of people who recognize the value of UBI, rising to 57%. UBI is probably most suited to the gig economy where a freelance driver is engaged in moving items from A to B and needs motor cover for the time it takes him.
Sold not bought
Going for the empathy angle by showing somebody a picture of their future self is an interesting move for selling life insurance. In daily life, people take selfies and, with the help of AI, either age themselves 20-30 years for a laugh or make themselves look younger to boost their dating app profile. AI is cropping up in unlikely social areas. I asked my guest where he thought AI’s greatest impact on insurance would be.
One universal truth is that insurance is sold, not bought, he pointed out. You're selling a product whose price you won't know for some considerable time. So, insurers could gain value from using ML to optimize or augment decision-making at major touchpoints, like first notification of loss (FNOL). There are numerous decisions that could help human assessors and claims handlers.
He went on to say that a marriage between the two works well. His company often sits team members next to claims handlers to listen to customer calls. They gain invaluable experience in how agents skillfully manage the distressing fact that somebody's house is flooded, assuring anxious owners that they’ll receive a payment in line with their cover. It’s crucial because although they're working with legacy system restrictions, they must develop a practical coping mechanism for the flood of stimuli and information.
Respect the human touch
Human touchpoints must be embedded throughout your ML solution, allowing employees to override decisions based on extenuating factors that the algorithm might not be aware of, such as distress levels and vocal characteristics.
ML produces a set of personalized questions that follow fairness and regulatory guidelines. Asking no more than a couple of relevant questions instead of the usual 20-odd eases the customer journey (insurers can retrieve most of the essential facts from the stack of factual and visual information freely available online).
Slick payouts pay off
Here's an example of a much-improved customer experience (CX). Suppose you needed to claim for an injured pet, and the vet’s practice had developed an ML solution that knew precisely what you were covered for. Almost immediately after posting the claim online, you’d receive a notification detailing the payout due and an assurance that the money would be in your account by the end of the day.
Similarly, an Irish healthcare business pays outpatient claims immediately through an app, using ML to read the receipt, identify the amount claimed and pay for a prescription or medical invoice that the patient settled previously. Of course, the patient might be seeing a specialist. In that case, they’ll probably want reassurance that the insurance company has investigated the situation thoroughly, ensuring they receive appropriate advice and the best service possible.
Working together
Nowadays, insurance company employees are much more in tune with technology and how data analytics works. I asked Sandeep if that would encourage greater collaboration between claims, data science and underwriting experts.
He acknowledged that we're seeing the pricing professions getting increasingly involved with changing the tech stack, using new data and fresh approaches to modeling. But building an ML model or solution is only part of the issue. Embedding it into core business systems, getting it used and having it deliver value while complying with all the regulations, technical drift/debt, etc., is a much more convoluted affair.
Many firms end up building models then struggling to get buy-in from the business. Other insurers are still wrestling with their data. For traditional insurance companies, the overarching challenge is how to manage old, dubious-quality data with legacy systems.
Attend to the details
We talk glibly about using GenAI as a quick fix as if it can solve all our data quality problems in the blink of an eye. Ultimately, it’s “garbage in, garbage out”. However, many incumbents are still working through the transformation process with only a handful of data scientists and engineers to complete the task.
So, how do insurers grow into new age segments or other dimensions in a static market? Or is growth by acquisition enough? And if all insurers use similar algorithms, ML-based pricing and risk calculation tools, how can they differentiate themselves?
The answer is in the detail. “Deliver great service” — a pain-free claims journey with great cover, a replacement car and fewer intrusive questions to answer. Successful insurers support customers in their time of need, whenever, wherever and however.
And finally
In answer to the question, “What are your hopes for 3-5 years’ time?”, Sandeep said he’d love to see ML teams treated as strategic partners within the business. It would be great to discuss new technologies with the leadership and make decision-makers aware of their capabilities beyond the hype, ensuring they understand the full scope of the business and operational potential.
Another vital commitment would be to move away from legacy tech. And while we’re at it, we should improve data literacy, so that the algorithms and ML decisions are as free from bias as possible, enabling insurers to make better-informed, more accurate decisions.
As usual, there are more questions than answers. For instance, “How can improvements be embedded along the value chain?” “In 3-5 years, how will distribution (the main focus for insurtechs) have evolved?” “Will we see an abundance of insurtechs along the various elements of the value chain?” “How will funding have changed?”
I guess that, by then, decision-makers will have brought in the right technologies to change lives and improve the value chain.
In an ideal world, anyway.
Still looking for answers?
If you’d like to discover more about how Zoreza Global can make insurtech disruption work for your insurance company, visit our website or contact us.