Much like the retail and investment banking sectors, both the personal and commercial insurance domains are on the cusp of a seismic industry change that has been widely stimulated by technology innovation. For an industry sector that has been driven and underpinned by global trade, for centuries. Insurance is perhaps one of the industries most ripe for widespread digital transformation that needs to be powered by data, analytics, ML & AI.
For too long, practices such as policy pricing have been executed using spreadsheets and emailed across the organisation. Whilst underwriting approvals have been at the behest of a common few who have had a market monopoly or established relationships, as the result of decade long partnerships and agreements.
However, things are changing!
Much like the banking sector, Covid-19 dealt an instant hammer blow to insurance organisations overnight. At the flick of a government switch, businesses have had to adapt to get their back office and technology functions working remotely, upgrade their digital services to match customer demand and invest in new distribution channels to remain relevant.
Whilst many organisations did what they had to do, in order to navigate the unchartered waters of the pandemic. Many are now in a much better position to make use of digital technology and data driven solutions. Insurance has been an industry that was typically challenging to disrupt and change. However, now more than ever, heads of innovation, managing directors for lines of business and technology leaders, can be more confident that their workforces will embrace greater levels of change and internal disruption. To bring this subject area to life, we will explore the following opportunities for ML & AI across the insurance sector:
- Dynamic Pricing
- Claims Management
- Fraud Detection
- Lapse Management
- Recommendation Engines & Product Cross-Selling
- Automated Underwriting
Insurance organisations all want our business. They want us to insure our cars with them, our houses with them and our freight shipping containers with them! They also want to ensure that they are able to effectively price their policies competitively without adding undue risk onto their organisation. Therefore having an effective pricing strategy has often been seen as the gold standard for a high performing insurance service provider.
This has proven to be especially important in the last 18-months, whereby firms have had to understand market sentiment and respond to economic fluctuations brought on by events like the global pandemic and Brexit. And in response, launch aggressive campaigns to retain or acquire new customers with offers such as lower premiums, returning portions of existing premiums or extended coverage periods.
In the face of fierce competition, many organisations have sacrificed long-term profits to provide short term spurts of growth. This is particularly challenging in the personal lines sector where margins are small and the cost to serve customers is exponentially increasing owing to ageing technology stacks. Furthermore, the creation of, testing and validation process for publishing a new pricing model can often take between 6-8 weeks to push to production. This simply isn’t fast enough!
In a price-sensitive, consumer-driven market where pricing aggregators are the preferred channel for many personal lines customers, organisations need to know what their competitors are charging for the same or similar policy. And where possible, competitively price their product without onboarding risk to their business. In short, with an enriched data, analytics, ML & AI driven pricing strategy, firms can deploy pricing changes in minutes, not months.
A dynamic ML & AI pricing capability across the insurance sector can be a source of growth and create significant value for both personal and commercial providers. Insurance organisations are already sitting on a wealth of historical data that when combined with a modern technology strategy can enable the delivery of pricing changes that are more tailored to each customer and their specific needs and risk profile.
By integrating the analysis of internal and external data sources insurers can build a true 360 perspective of their customer. In time, by harnessing the power of self-learning algorithms and ML, insurers have the ability to generate price quotes that beat the market whilst still maintaining a healthy profit margin. Furthermore, by leveraging dynamic cloud-hosted infrastructure, organisations can scale technology operations up and down based on market demands. This results in the more optimal management of cost overheads for technology. Especially in the commercial sector where firms have traditionally invested in huge compute resources to support one-off insurance pricing events.
In addition, by enhancing their pricing capabilities firms should also be able to understand the policies they sell and perhaps more importantly the policies they don’t sell and why. Innovative leaders in the pricing space are able to utilise these insights to invest in initiatives that generate new leads for initial quotes. The result of this could be a targeted advertising campaign to customers who represent their most valuable segment to drive incremental revenue gains.
Insurance products are traditionally abstract in their nature and in essence, are acquired as a risk protection service with an upfront payment that provides little to no guarantee that your insurer will cover any losses or damage that are incurred as a result of a disruptive event! Insurance is by all intents and purposes a begrudging purchase. Customers buy it because they need to, not because they want it.
Specifically, the claims management journey, from prevention to loss notification, to assessment, to handling and settlement has traditionally been something of a black box and convoluted to customers. When customers submit a claim, it can be a game of cat and mouse to ensure that all the conditions and boundaries of their policy have been operated within. Ultimately, this can be a complex, cumbersome and time-consuming activity that has traditionally been fraught with annoyance and frustration for customers.
I mean, who hasn’t spent 30-minutes in a call queue waiting to speak with an agent about an existing or new claim submission, only to be handed off to another team once your details have been verified?!
On the flip side, insurers see claims as a cost centre.
In essence, it is a business unit that costs them time and money to operate. With significant technology and people overheads that have been underpinned by heritage systems and processes which have traditionally been decoupled and siloed in nature. As an example, Oliver Wyman reported that across the European insurance markets, the annualized growth of total benefits-and-claims spend is more than 4%. This translates to more than €350 billion per year, a number that is likely to continue rising without the right interventions.
Conversely, customers aren’t prepared to hang on the phone for 30-minutes to speak with an agent any longer. Especially to be told their claim won’t be accepted and processed by the insurer!
Customers want to be serviced 24/7, through digital channels that give them the option as to when they need to speak with a human being or a virtual assistant. It needs to be simple, friction-free and integrated. Otherwise, customers will recall their previous challenges and simply opt not to renew their policies when it really matters!
On the flip side, Insurtechs are further disrupting the established players with digitally native propositions that are building integrated end to end digital solutions that address customer hassles across the claims process. As such, insurers know they need to modernise their claims processes to become customer-centric in order to retain and acquire customers.
Claims management continues to be a particularly manual, human driven review process. However, any workflow where there are multiple steps consisting of the following events is prime for the execution of ML & AI driven improvements.
1) Digitisation of information.
2) Data collection.
3) Data processing.
4) Analysis and review.
5) Review of terms and conditions.
6) Optimised decision making.
A typical claims process is made up of the following steps:
1) Claim is raised and started.
2) Associated information, data and evidence is collected from the claimant.
3) The claim goes into a review process and goes through a number of reviews and iterations.
4) A decision is made and communicated to the claimant and any associated 3rd parties.
5) The information and data is updated on the insurers systems, as well any industry and regulatory bodies.
For each of these steps there are potential improvements through the application of ML and AI. Namely;
1. Claim is raised and started
a) Digital interaction with the claimant through a virtual agent or a chatbot engagement can drive out the right levels of comprehensive information.
b) Typical evidence such as documents, photos and videos can be collected and stored competently in a self-service manner.
2. Associated information, data and evidence is collected from the claimant
a) New types of information and evidence such as IOT data, location data, links into government, police and health data could potentially be linked.
b) This data could be processed, catalogued and stored in modern data lake structures which allow the fusion across structured, semi-structured and unstructured data.
3. The claim goes into a review process and goes through a number of review and iterations
a) A wide range of features can be generated to feed into machine learning models which predict claim fraud, severity, and terms and conditions viability.
b) Then this can feed into recommendation engines to either provide automated decisions or support underwriter review.
4. A decision is made and communicated to the claimant and any associated 3rd parties.
a) Automated notifications and communications can be triggered to the customer via letter, SMS and email. With further human touch follow-ups scheduled to confirm the outcome of the claim with the customer.
5. The information and data is updated on the insurers systems, as well any industry and regulatory bodies.
The use of API’s can securely transfer the required data and supporting artifacts to the required 3rd parties with audit events captured demonstrating the sharing of the information, payment to the claimants and any other relevant information.
Like their banking peers, the insurance sector is awash with fraudulent activity. Whether this is customers lying or not providing full facts when they take out a new policy. Or submitting a fraudulent claim to underpin some form of illicit activity. Fraud is big business and it costs insurers a lot of money each year. These costs ultimately end up being passed onto honest customers with higher premiums. In 2020, The Association of British Insurers reported the following facts and figures regarding fraud:
1) 107,000 acts of fraudulent insurance were £1.2 billion were uncovered in 2019.
2) This equates to 300 fraudulent submissions every day being caught. Or one every 5-minutes.
3) £3.3m in fraudulent claims are uncovered every day. With the average value of a dishonest claim equating to £11,500.
4) In respect of dishonest applications. 760,000 were identified to not have been given the correct facts to correctly price and underwrite the policies.
5) 58,000 motor insurance cons were identified over the course of a year.
Did we mention that these are the fraud claims that U.K. insurers identified? Therefore, it is little surprise that insurers are actively seeking to increase their abilities to spot fraudulent activity with ML & AI.
There are a number of key interaction points where fraudulent activities occur and where machine learning can provide substantial uplifts.
1) New Applications or Renewals
Policies are priced on the information that a customer provides. If the customer provides incorrect information deliberately this can lead to significant lost revenue for the organisation. Machine learning techniques such as clustering and missing value prediction can be used to highlight anomalies.
The ML & AI enhanced claims process has been detailed in a previous section. To build on this there are two main areas of opportunity for machine learning:
- Individual claims fraud: An individual attempts to claim for a completely fictional claim or they attempt to alter the features of the claim to inflate the value.
- Organisation criminal gangs claim fraud: A particularly nasty type of fraud is being driven by organised criminal gangs which is then used to fuel further criminal activities. They use their network of gang members and affiliations to damage property, vehicles and other material goods to drive forward fraud at scale.
They use individuals and businesses to falsely claim or initiate criminal activity which they then claim on e.g. arson. A powerful technique that's already used for this type of activity is graph analytics where the complex interrelations between people, communications and property is mapped and analysed. However these approaches tend to be outsourced to suppliers and are limited by flexibility, scale and cost. Now that graph analytics databases and open source machine learning are widely available, the opportunity exists to build leading solutions bespoke to the organisation.
Lapse management refers to when policyholders stop paying premiums and when the account value of the insurance policy has already been exhausted, the policy lapses. As an example, if an insurer has 1000 policyholders who are approaching the end of their policy and 700 of those customers renew their policy, then the insurer has a 30% churn rate of lapsed policies. Insurers are legally bound to give a grace period to policyholders before the policy falls into a lapse. However, many insurers want to ensure they retain customers and avoid losing them to competitors.
Furthermore, in the U.K. the FCA has launched a new set of legislation that is aimed at ensuring loyal customers are not penalised for remaining with their existing home and motor insurance providers. For example, it was estimated that in 2018, 6 million loyal policyholders would have saved £1.2 billion had they paid the average price for their actual risk. As such, lapse management and customer retention are going to become an increasingly complex overhead for personal insurers. Which if not administered correctly, could result in significant customer churn and run the risk of regulatory penalties.
Insurance organisations can apply enriched analytics to better identify the customers who are soon approaching their lapsed policy period. Furthermore, this can be integrated with RPA capabilities and notify customers that their lapse period is indeed approaching and they should act.
In addition, by integrating this with their dynamic pricing efforts, firms will be able to demonstrate that their customers have indeed been charged the correct amount for their policy without merely jacking up premiums. In turn, ensuring that they are complying with the FCA’s regulatory requirement.
Finally, sentiment analysis can once again be used during telephone interactions with customers who are entering their lapsed policy period. Ensuring that call centre agents and team leads are aware of which customers are likely to leave based on cost, service experience or other challenges. This can drive tailored responses on a customer by customer basis. Whilst the data can be used to better understand why customers did not renew their policies to remediate these issues earlier in their customer journey.
“It takes my organisation 14 days to get access to a curated list of customers for a targeted marketing campaign to launch a new product or bespoke deal”. Insurance organisations are missing out on launching “in the moment” policies and products because it takes them too long to identify potential customers. This often results in cold product selling, with products not hitting their most relevant segment. Meaning insurance organisations spend a lot of time and money, selling to customers who have a lower propensity to acquiring their services.
This one size fits all approach to marketeering, results in a lack of empathy amongst their messaging. Meaning customers might opt out of any future campaigns.
Firms can integrate data and analytics from their own systems with trusted 3rd party data sources. As an example, social media channels. This could result in key word searches, search engine optimization and targeted advertising being used to sell products to “lookie-likes” who demonstrate the same segmentation and demographic of customers that have previously purchased an insurance product from them.
Indeed, when customers opt to acquire a new insurance policy there are other possibilities in tracking certain data types (post code, vehicle type, age, crime rates) to provide recommended product bolt-ons that similar customers have previously purchased in addition. This demonstrates further opportunities for the use of ML & AI.
“Sorry, our Chief Underwriting Officer doesn’t work on Saturdays and I am only able to approve a certain value. We will have to get back to you in 5 working days with a revised quote”. Customers want to have their policies underwritten 24-7, 365 days a year. As such, insurance organisations are losing out on potential cross-selling opportunities because their underwriting processes are built for a generation of customers who are content with Monday to Friday, 9-5 services.
More often than not, these underwriting practices are conducted in spreadsheets…. Or on notepads stored in a cabinet or a desk!
Firms can use ML & AI to establish automated risk based models that operate during windows when the Chief Underwriting function is not available. Risk thresholds can be applied with multiple data points cross referenced to understand the premiums that need to be applied to a policy.
Each policy that is underwritten can then be scored, stored and processed to understand how each incremental policy affects the firm's risk ratio. This data can be cross-referenced across products, regions, countries and institutions in real-time.
All of this allows firms to shorten the underwriting process from days to minutes. Resulting in a friction free process that is entirely auditable and improves underwriting profitability.
Over the course of this blog, we have discussed the personal and commercial insurance sectors highlighting the potential use cases for ML & AI. Having discussed dynamic pricing, claims management, fraud detection, lapse management, recommendation engines & automated underwriting we can conclude that;
1) ML & AI is critical to establishing a dynamic pricing strategy and capability that enables insurance firms to remain competitive in an increasingly competitive market place.
2) ML & AI can streamline business processes and introduce cost & time savings in the claims management domain.
3) The underwriting process can become highly automated and more transparent, with effective use of ML & AI to handle enterprise risk management practices.
4) Alignment of technology choices and implementation strategies across pricing, underwriting and customer acquisition can shorten the feedback loop for enriched customer segmentation. This can drive further revenue growth for firms if implemented effectively.
5) The demands of customers, especially in the personal lines sector, requires firms to alter their digital experiences. In doing so, intelligent document ingestion and RPA can be used to effectively streamline the claims management process.
In our next blog, we will delve into the blockers that are preventative measures in scaling ML & AI adoption across financial services. Whilst we also explore what firms need to have in place to successfully implement ML & AI both safely, securely and in a compliant manner.
Interested in seeing our latest blogs as soon as they get released? Sign up for our newsletter using the form below, and also follow us on LinkedIn.