Member Spotlight: Sprout.ai
Could you briefly introduce Sprout.ai?
Sprout.ai was established to make every claim better. Sprout.ai has built an AI-based claims automation engine that helps insurers automate aspects of the claims process.
What is your role at Sprout.ai?
I am the CEO of Sprout.ai. I started my career in software engineering and most recently worked in the leadership team of Tractable.ai, where I learnt about how technology can help insurance companies automate the complexities of the claims process. I joined Sprout.ai at the start of 2022.
What challenges does the current claims process pose for claims handlers?
Even for a simple claim, claims handlers need to read and rekey lots of data, validate supporting documents and identify coverages and exclusions in long policy documents. For complex claims, there are many more relevant factors and data points that need to be considered. This information is often stored in multiple legacy systems.
At the same time, claims handlers are working with customers who are often under stress, impatient and pressed for time. They need to show empathy and understanding when communicating with customers.
When claims handlers spend so much time reviewing handwritten documents, analysing images and manually searching, this often does not leave them enough time to devote the right care to their customers and creates risks of human errors that may result in an incorrect settlement decisions.
Other than making claims handlers more efficient, what other challenges do the insurance companies themselves face on the claims front?
I think there are two big additional issues that most insurers face. The first is that it is often difficult for them to manage the volume of claims that they receive. Depending on the line of business, there can be spikes in volume, especially after natural disasters or in seasonal insurance like travel. Even insurance companies that have the most efficient processes can take weeks or months to handle spikes in claims.
The second is that currently, it is difficult for insurers to recruit claims handlers, there is a shortage of experienced professionals, job retention is low and new recruits take time to train.
What do customers want from the claims experience?
Customer expectations of the claims experience are different from before; people are expecting more from their insurance provider than previously. Speed has been a key driver over the past few years and there is a lack of willingness to wait weeks or even months for a claim to be settled. Sprout.ai conducted a survey last year that found one in five customers expect claims to be resolved within hours. Tech-first insurance companies like Lemonade and Hippo are providing more immediate claims experiences that push customer expectations higher for other insurers to follow this trend.
Other factors are also at play. Customers also want visibility into the claims process. They want to understand what information they need to provide, the status of the claim, when they will be paid, how much and why. Customers may be willing to wait longer for payment if they have more clarity around the process.
How can implementing AI change what it means to handle claims in the future?
With the right solution in place, AI and other technologies can save time for claims handlers. For example, Sprout.ai’s solution sifts through claims forms, related policy documentation, receipts, prescriptions, referral letters and other supporting documents. The relevant information is extracted and provided to the claims handler in an easy-to-use way. The AI can flag what aspects of the information need the most attention from the claims handler.
Some insurers may choose to automate certain claims entirely. This frees up time for claims handlers. They can process other claims faster and focus more time on complex claims or customer communications.
Overall, Sprout.ai’s AI solution can significantly reduce the time invested in claims processes by automating data extraction, policy checking, data validation and sometimes even the final decision.
Does this solution use generative AI, such as large language models (LLMs), or other forms of AI?
Sprout.ai uses both generative AI and other forms of AI. Our solution involves several different types of AI, including visual AI, deep learning, LLMs and generative AI. At Sprout.ai we have been using AI for around five years, long before the current hype around generative AI in particular, that has brought AI to the forefront of conversations.
Talking specifically about generative AI, one way we use it is to generate synthetic claims data that we use to train our other (non-generative) AI models. As a result, we do not need vast amounts of data from our customers to build accurate models.
Another example is that we use LLMs to interrogate policy documents and understand the coverage.
What advice would you give to insurers looking to adopt generative AI?
Many organisations want to implement generative AI, but the question should not be, ‘What can I do with generative AI,’ it should be ‘What problem do I want to solve in my business.’ Once a company has identified a problem, it should investigate what is the right tool to solve it. That might be using generative AI, other forms of AI or it might be something else.
Earlier this year, Sprout.ai surveyed over 120 insurance companies. 61% were using some form of AI to improve the efficiency and productivity of their team. 48% were using AI to improve their customer service. These are both key areas where AI in general can help.
Is AI ‘hallucination’ a problem for insurers when using generative AI?
In May 2023, a US lawyer famously generated a legal briefing for a court case using ChatGPT. It looked convincing, but it referred to historical precedents that did not exist (this phenomenon is called hallucination). This mistake was caused by using the wrong tool for the task and because the lawyer didn’t validate the results.
Similarly, in the insurance sector, insurers need to use AI tools that are adapted to the specific problems they are solving, rather than generic tools like ChatGPT. Sprout.ai has adapted our AI models specifically for insurance claims.
How do you manage other risks when implementing AI into insurance processes?
It would be possible to train an AI model on all the information about a claim and ask it to decide whether the claim should be paid, but this would create a black box; the model would give an answer without any explanation of how it made the decision.
A better approach is to break down the problem into several separate questions: ‘What is the item in question?’; ‘What happened to it?’; ‘When did it happen?’; ‘How much does it cost?’; ‘Is it covered by the policy?’; ‘How much of it is covered?’. These are separate questions that we can answer with AI models. You can then use the output of each to make the decision, while you retain clarity on how the decision was reached.
Finally, insurers should also put safeguard mechanisms in place. For example, insurers could choose to have claims handlers look at the details of every claim over £1,000. Alternatively, they could have a system where a claims handler looks at 5% of all claims and their decision is compared to the results of the AI. Insurers that work with Sprout.ai usually start with a soft launch like this where the AI process works in parallel with claims handlers. This helps identify and correct mistakes.
Is bias a problem when using AI for insurance claims?
AI models learn from the data they are given. The most effective way to avoid bias is to avoid providing models with data that may create biases. Insurers have lots of different information on claimants, such as address, gender and age. Where some of this information is not a relevant factor in the claim, it should be excluded to avoid creating bias.
How can insurers choose where to make decisions with AI, as opposed to humans making decisions?
The question is usually not choosing between humans and AI. In most cases, it is a collaboration between the two that drives the right results. Technology can help people do their job better, save time and free them up to work on areas where they add value.
For insurance claims, it is worth considering the circumstances of the claim. For claims that involve long-term illnesses, customers may need to talk to a person. If the claim is a chipped windscreen, automated communication over email may be sufficient. Insurers need to decide when they want to talk directly to their customers and when a fully automated solution is better.
What lines of business does Sprout.ai support?
Sprout.ai’s vision is to make every claim better across all lines of business. So far, we have seen the most success in life, health and property claims. We are also working on use cases in travel, pet and dental insurance. Our solution is currently well-suited to reimbursement claims.
What clients is Sprout.ai working with that you can name?
Sprout.ai’s clients include AXA, MetLife and Generali.
Why is Sprout.ai a member of InsTech and what sort of companies would you like to connect with?
Partnership and collaboration are important for Sprout.ai. We are looking to collaborate with system integrators, claims management systems, insurers, third-party administrators and brokers. We are working with InsTech to find potential new partners and customers, learn from the community and contribute to the network. We would like to connect with others who share the vision of making every claim better. You can reach out to Sprout.ai on our website and connect with the team.