Artificial intelligence (AI), or the ability of a machine to solve complex problems and improve itself over time, has been on everybody’s lips for some time now. For once, the insurance industry is picking up quickly. In particular, artificial intelligence is stirring up an area that has long been rather devoid of major innovation: the claims process. While the start-up Lemonade announced that it is able to settle claims with the help of artificial intelligence within 3 seconds, news of major incumbents partnering with tech companies to improve their claims processes are circulating. But how can artificial intelligence really help to improve the claims process? What are its limitations? And what does it require?
Opportunity: efficiency gains along the claims process
In its most basic form, artificial intelligence analyses data, initiates rule-based actions and in doing so, replaces repetitive, manual work. What is more, it assesses much larger amounts of data than any human ever could and thus detecting logic correlations that were previously left in the dark. As new data is fed into the AI, it is able to independently adapt its models and refine its output. When we look at the claims process, it is clear that AI offers major potential for automation, namely through assessing, classifying, estimating and validating claims without human interaction. By that artificial intelligence allows not only for tremendous scale economies but also for a service that is neither bound by time or location.
In a nutshell, artificial intelligence can shorten the claims lifecycle, save costs and enhance customer satisfaction along 4 major opportunities:
- Customer interaction / chat bot
Based on natural language processing, i.e. the ability of artificial intelligence to understand and process human language, so-called chat bots can engage in human-like customer interaction. As for the claims process and in particular for the first notice of loss, chat bots are able to answer questions from the claimant, request the relevant data, retrieve and process evidence and initiate first actions. What is more, they can be given any identity to reproduce human facial expressions and increasingly, AI-driven chat bots are also able to understand their counterpart’s mood and adapt responses accordingly.
- Triage and claims routing
Through its ability to detect patterns in the data fed into it, artificial intelligence can classify claims and then take rule-based actions accordingly. For example, by scanning claims data for specific variables, AI can filter out claims that can be serviced in a straightforward manner from potential cases of fraud, litigation or subrogation, etc. After classifying a claim by its potential severity, artificial intelligence can determine which level of validation is needed, decide on claims routing and assign a claim to a claims handler accordingly.
- Claims estimation
Artificial intelligence systems are beginning to develop abilities to automatically appraise from images. This is, however, one aspect in which artificial intelligence is not yet very developed; still, this capability holds great potential for insurance companies. From photos of damages, AI systems are increasingly able to assess the damage and match it with previous claims data to then estimate repair or replacement costs in a fully automated manner.
- Fraud detection
Have the claimants shown fraudulent behavior in other contexts or have they exhibited characteristics that tend to coincide with fraudulent behavior? Through processing enormous amounts of data and detecting correlations, artificial intelligence systems are perfectly suited to fight one of the largest cost drivers in claims: fraud. Based on algorithms, AI can determine the fraud potential of a claim or claimant and automatically assign conspicuous claims to human experts.
Limitations and key prerequisites for the application of AI solutions
While the opportunities described above can be substantial, they all rely on a key prerequisite: training with adequate data. Artificial intelligence runs on data. Thus, regardless of how AI is developed or sourced, insurance companies looking at adapting artificial intelligence for their processes first have to consolidate their data and make it available for the AI. Often enough, this poses a fundamental challenge to insurance companies in itself. Secondly, Artificial intelligence is only ever able to do what it has “learned” from data that it has been trained with. In an incremental process on the road to full automatization, the accuracy levels of an AIs output will continuously rise with its training. In order to achieve high accuracy, models and algorithms have to be carefully chosen, tested and fed with increasing amounts of data. Operationally speaking, for now, artificial intelligence will have to “learn” alongside manual processes to eventually fully replace them. The point, or the accuracy level, at which an AI solution starts performing specific process steps by itself, depends on the insurance company’s risk appetite.
Overall, the benefits of AI for claims management are obvious: AI based processes can save costs for the insurance company, shorten the claims lifecycle and by that enhance customer satisfaction. But there is still a way to go – available artificial intelligence solutions are not yet “plug and play”. It takes time and effort to train them for high accuracy – a fact that opens up a race against the clock. The fastest provider will benefit from a competitive advantage that will be hard to catch up to. The ways to deploy AI are manifold: Artificial intelligence can be established independently based on existing AI solutions (e.g. Google’s TensorFlow), used from the cloud (e.g. Amazon Web Services), and sourced as “software as a service” from dedicated AI providers. It is thus about time for insurance companies to carefully assess their current business processes for AI based automation potentials, and to set a strategic agenda.