AI initiatives at many organizations are too small and too tentative. They never get to the only step that can add economic value—being deployed on a large scale. Testing the waters may deliver valuable insights, but it probably won’t be enough to achieve true transformation. A pilot program or experiment can take you only so far.
Sound familiar? There are usually many hurdles in the way of getting beyond the initial micro steps; data silos, multiple unconnected technology stacks, and inflexible core technology. You name it! But the fact is some insurers are surmounting these obstacles. Read this practical advice but first, lets look at those hurdles in a bit more detail before looking at the solutions.
"At many organizations, AI initiatives are too small and too tentative; they never get to the only step that can add economic value—deploying a model on a large scale. In a 2019 survey conducted by MIT Sloan Management Review and Boston Consulting Group, seven out of 10 companies reported that their AI efforts had had minimal or no impact. The same survey showed that among the 90% of companies that had made some investment in AI, fewer than 40% had achieved business gains over the previous three years. That’s not surprising: A pilot program or an experiment can take you only so far."
Thomas H Davenport & Nitin Mittal HBR Jan 0Feb 2023
Not to mention the need to face constant failure in order to relearn fast, iterate, possibly change direction and last the course to eventual success. One example of this comes from a different field altogether; see - US Marines defeat Pentagon AI test by hiding in cardboard box
Scharre writes: "They parked the robot in the middle of a traffic circle and the Marines had to approach it undetected starting from a long distance away." You're already on his- some Marines hid in a cardboard box and got to the robot without the algorithms having the faintest idea.
Even Tesla with the most mileage, event, and contextual data of any auto OEM is finding it a tough nut to deliver autonomous vehicles. But, they will continue the intensive efforts, and the rewards for those auto manufacturers that perfect AVs will be immense.
As Daren Rudd Vice President of Consulting - Head of Insurance Business Technology Consulting at CGI UK says " We are very much in the learning-by-doing stage on how to get the best out of these tools. You need to be out there testing our different approaches, seeing what works or doesn’t, and not be afraid to fail (or learn fast)."
One feels that insurance carriers and brokers are at this tipping point between experimenting with AI-powered software like Tractable, Solera, Shift, FRISS, BAIL, SLVRCLD, and Om:nius etc, and leveraging the full strategic benefits of AI. The huge amounts of data generated by sensors, connected property, vehicles, plant & machinery, and telematics are the fuel to power AI potential. And many companies have developed mature applications solving real problems but they need to join together in a strategic plan.
Successful AI today tends to be "narrow AI"- tight use cases for very specific tasks.
Yet too often the single application merely fast-tracks the prediction or decision to the next buffer of human decision-making unless these point solutions are planned in the context of deploying at a large scale and across the whole enterprise.
One factor that has held insurers back has been either legacy systems or the inflexibility of current core systems of record. It just takes too long, costs too much, and swallows too many scarce human resources and talents. What's the answer?
The new generation of No-Code/Low-Code, API Smart, and Micro-Services architectured CoreTech and ClaimTech like EIS, ICE, Instanda, Genasys, Ignite, and RightIndem are ideal platforms to start the plan to "Go All In" with AI.
Insurers such as esure, Liberty Mutual, The AA, and Simply Health have re-platformed to enjoy the competitive advantages of agility, faster innovation, and the ability to deploy technology ecosystems.
With these agile and inrerconnectable platforms carriers and brokers are better able to test short-listed applications in A: B trials with production-level claims and integrate the different packages chosen, the multiple and critical data sources, and do so a step at a time towards the ultimate goal of leveraging AI and data to the full.
All is not lost for those committed to the last upgrade of the core systems of course. RightIndem has integrated with Guidewire and other current core systems not to mention the new generation. You can still work on planning and deploying AI with the older architecture platforms; it is just a degree harder.
HBR describes 10 steps to success.
- Know what you want to accomplish
- Work with an ecosystem of partners
- Master Analytics
- Create a modular, flexible architecture
- Integrate AI into existing workflows
- Building solutions across the organisation
- Create an AI governance and leadership function
- Develop and staff centres of excellence
- Invest continually
- Always seek new sources of data
1) HOW WHAT YOU WANT TO ACCOMPLISH
This is a critical factor in whatever innovation and transformation you want to achieve. Why, and what clear objectives to achieve the company goals e.g.
- Improve process and settlement speed
- Reduce indemnity and operational costs
- Personalised products and services
- Build communities of customers
HBR recommends one well-defined, overarching objective that is a guiding principle for AI adoption; eg achieving improved process and settlement speed across a global insurer will require developing a global data model. Start with your national ones but aim to eventually deploy a single data model to achieve global competitive advantage.
2) WORK WITH AN ECOSYSTEM OF PARTNERS
A company needs strong partnerships to succeed with AI. Take carrier esure in the UK with its goal to be "The Insurer of the Future". It's partners include: -
- EY- SI and implementation partner
- EIS- CoreTech partner
- Amazon AWS- Cloud Infrastructure & Services partner
- Amazon Connect- Omni-Channel Call Centre partner
- RightIndem- Digital Claims Partner
Naturally, esure has also specified AI-powered apps to be integrated into this new technology stack.
3) CREATING A MODULAR, FLEXIBLE ARCHITECTURE
This is a pre-requisite to effectively deploy data, analytics, and automation across an insurer and its brokers and supply chain partners. Many a claims transformation project has been stymied by the resistance of brokers to adopt rigid technologies deployed by carriers. Focus on real problems, simplicity, and getting buy-in from all parties becomes easier ( but it's still not easy) with modular architecture (MA).
MA makes it easier to extract key terms from legal documents, policies, invoices, medical reports, freeform text, and speech.
How many auto insurance carriers in the UK analyse the current, timely, and valuable insights from MOT data? Percayso Informs Kieren Fischer brings this to life in a recent article:
"Percayso obtains this data directly from the DVLA, Fisher said, which ultimately is a live system but what’s interesting to see is that while some insurers are already consuming the most basic MOT data, few are digging deeper into the wealth of insight it can yield.
To his mind, the real benefit is around understanding the condition of the vehicle. When insuring an asset, he said, understanding that asset as much as possible and how likely it is to be involved in a claim is what insurers should be doing. So, Percayso is taking it one step further and starting to examine the results of MOTs, looking for advisories and failures and categorising these into different tranches.
This enables Percayso to deliver an MOT ‘risk score’ based on the condition of the car and, crucially, its maintenance at the point-of-quote and at speed, he said. And be able to understand these trends on a vehicle-specific basis and deliver these insights in a meaningful, consumable way is an interesting proposition that is translating well across the insurance marketplace.
How willing are insurers to embrace new data variables?
Fisher and the team are seeing significant appetite from insurers to engage with new data variables, which he highlighted is unsurprising given that everybody is looking for the newest exciting data to give them an edge in today’s competitive market. Having access to these processed results on vehicle data is a key way for companies to stand out but it also allows them to price better and write better business."
With a modular and flexible architecture such as that chosen by esure, it is practical to embrace the data, applications, and platforms to leverage AI.
4) MASTER ANALYTICS
This requires a commitment to using data and analytics for most decisions embedding AI in products and services and conducting tasks and even completing entire processes in a more automated and intelligent way. Typically, most enterprises' biggest obstacle is acquiring, cleaning, and integrating the right data. Take the matter of broking and placing business from retail brokers in the US onto the London Markets. Data is typically a nightmare; the surfacing and normalization of unstructured data is a massive challenge and a priority to manage effectively before contemplating AI ventures.
"Greater Than" can analyse a carrier's claims and GPS data and model against its unique global database and states that its predictive risk AI can reduce the auto loss ratio by 10 to 15 percentage points. Which carrier wouldn't at least test this? Especially when UK carriers are currently reporting increasing loss ratios and worsening bottom-line results.
5) INTEGRATE AI INTO EXISTING WORKFLOWS
Inflexible work processes are as much of a deadweight as inflexible IT architectures. Working back from the customer ( policyholders, brokers, agents, and supply chain partners who are at liberty to work for you or not i.e. choose other insurers in a tight market) HBR recommends you choose which workflows are ripe for AI speed and intelligence. It is the workflows that generate enormous amounts of data and repetition that will benefit the most.
The claims handlers on the frontline will have an eye for and knowledge of processes that will benefit and how they can be explicitly improved. Again, modular and agile technology makes this a practical goal.
6) BUILD SOLUTIONS ACROSS THE ORGANISATION
Once you have tried, tested, and mastered AI across specific workflows, say accidental damage home contents claims with STP, you can become more aggressive across the organization and deploy it in motor insurance. Can algorithms applied in one process be applied across the company?
Take the data analytics mentioned above to predict risk and price risk in a granular fashion. Why not use the same algorithms used to predict risk to be utilised to claims volumes and plan claims resourcing i.e. number and skill profile of claims adjusters and strategy to outsource claims?
7) DEVELOP AND STAFF CENTRES OF EXCELLENCE
To go all in with AI will require sufficient funding and time and heads of AI and excellence that can evangelise and implement projects across the various business units. Claims and underwriting can both benefit from leveraging AI. Everyone knows that reducing claims by writing better risk profiles COMBINED with managing those claims with an optimal mix of automation and technology to enhance claims decisions will lead to considerable competitive advantage. That requires the upskilling and reskilling of people combined with the right technologies.
Set a goal for AI-powered underwriting that reduces loss ratios by 10 to 15 percentage points COMBINED with digital claims transformation to reduce loss ratios by a further 5 percentage points. Ambitious insurers may aim even higher.
8) CREATE AN AI GOVERNANCE AND LEADERSHIP STRUCTURE
It is one thing to have an AI leader to evangelise and be in charge of how AI is deployed throughout the organisation. It is quite another thing to create a culture that focuses on data-driven decisions and actions. That requires a CEO that is familiar with IT leading the project.
- Motivate and get buy-in throughout the organisation
- Support teams and collaborative partners through the rough as well as the smooth.
- Deal with the ethical and regulatory aspects that may limit what a company can do
- Knows that this will take time, money, and commitment and will not be a short-term deliverable.
9) INVEST CONTINUALLY
Ambitious insurers that choose to be aggressive with AI should know that is not a decision to be made lightly. The move will have an impact for decades and require large budgets. They can, however, plan and deploy on a stage-by-stage basis and seek to make returns on investment for each stage so funding further stages.
The HBT article describes how CCC Intelligent Solutions has spent and continues to spend more than $ 100 million a year on AI and data. Its machine-learning models are based on more than a trillion dollars' worth of historical claims, billions of images, data on vehicle parts, repair shops, collision injuries, and regulations. I am sure the same may be said for Verosk, Solera, LexisNexis, Mitchell International et al.
Here's the thing- carriers can leverage this investment for a fraction of the cost especially as AI is itself making it less expensive to model events and predict outcomes.
Today there are many data, analytics, and AI service providers that can deliver the benefits of AI at a pro-rata lower cost to insurers. But, it still takes a continual and large investment as you add in operational, change management, and upskilling/reskilling costs.
10) ALWAYS SEEK MORE SOURCES OF DATA
Carriers generally collect and harness telematics data but how many collect and store GPS data? This is a rich source of insights to add to a data lake and enable better underwriting and claims management. Don't forget that if it comes from a policyholder's phone they must give permission for you to use it for the purposes you plan.
It is no good just collecting the data from your policyholders- you need data from the wider population if you are to predict more accurately, be selective on whom and what you cover. Practically real-time pricing data across vehicles new and old, parts and materials for home and motor, labour rates, and replacement costs for gadgets. That is if you want to manage to reserve effectively.
Data is often a nightmare- how many in underwriting and claims still rely on Excel? Try and get central IT to help manage data and when they find out it's Excel they will probably roll tgeor eyes and put your project at the bottom of the pile- into the notorious "pit of despair".
Luckily there are data and analytics consultants who will extract, and load data including Excel into data lakes and have the analytics platforms to transform the unstructured & structured data into usable and actionable insights and decisioning engines. It will likely take a long time, and will not be cheap but it will be effective. And if you want to be a top quintile insurer that not just survives and grows you will need to invest in this.
Google's recent announcement of its Insurance claim processing reference architecture | Google Cloud Blog is a major initiative to help carriers join all the dots though it does mean committing to a Google ecosystem. Amazon AWS and MS Azure can be expected to be on the same track.
The virtuous circle of more data, better models, more business, and more data is what will eventually make an insurer's application of AI powerful. New data will continue to flow into the company, and it will be used to improve estimate predictions and other functions. That will help insurers make better decisions, which will most likely bring insurers more business and more data.
"We also believe that AI—applied strategically and in large doses—will be critical to the success of almost every business in the future. Data is increasing at a rapid pace, and that’s not going to change. AI is a means of making sense of data at scale and of ensuring smart decisions throughout an organization. That’s not going to change either. Artificial intelligence is here to stay. Companies that apply it vigorously will dominate their industries over the next several decades. "
Here is a small selection of data sources insurers can leverage today
- CCC- featured in the HBR article; 50 billion miles’ worth of historical data through telematics and sensors in vehicles
- CoreLogic- manage property data for selling, financing, and protecting property
- Hazard Hub- property risk data
- Greater Than- a global database of vehicle/driver GPS and telematics data
- ICEYE- global flood earthquake and CAT damage data in near real-time
- KETTLE- house-by-house risk assessment across the USA
- LexisNexis- vehicle, ADAS, and home data
- McKenzie Intelligence Services Ltd wide range of data sources and data management
- Mitchell- Auto data
- SAFEHUB- building-specificS earthquake damage
- Synectics identity, financial, and fraud data
- Terrafirma property risk data
- Verisk- auto and property data
- WeatherNet- granular and near real-time weather data
- WhenFresh- wide range UK property data
Looking for AI solutions able to leverage this data?
Policy Admin, Claims Validation, and Triage
- Open GI
Claims Damage and Cost Estimation (often combinations of these)
- Be Valued
- Claims Genius
- Symbility (CoreLogic)
- Value Checker
Property Repair & Restoration
- Next Gear
- BAE NetReveal
And Claims Platforms?
Ecosystem Claims Management Platforms
- CoreLogic for property repair and restoration
- Verisk for property and auto repair, restoration, and total loss decisions
New Claimstech Management Platforms
- Claims Genius
- Claim Technology
- Salesforce Industries (Insurance)
Despite progress in AI-powered claims handling, human judgment remains ‘key,’ Mitchell says
Claims automation: separating fact from fiction.
From Prediction to Transformation across Claims
Nine customer types defining the next wave of insurance
To get substantial value from AI, your organization must fundamentally rethink the way that humans and machines interact in work environments. You should focus on applications that will change how employees perform and how customers interact with your company. You should consider systematically deploying AI across every key function and operation to support new processes and data-driven decision-making.