Increasingly sophisticated fraud is exposing insurance firms to growing risk — and it's also exposing the shortcomings of the corporate data systems expected to help AI combat the epidemic.
Insurance fraud is costing companies
But AI is only as effective as the data feeding its machine models. The sector may be awash with information, but bringing it online quickly and effectively to meet the demands of AI and get ahead of resourceful criminals is a challenge for nearly half of the market.
It's time for insurers to rethink conventional data processing strategies before pressing ahead — to ensure its designed reliable and effective AI-driven fraud detection and prevention.
A renaissance of digital opportunities in insurance
The insurance industry is undergoing a phenomenal transformation with new markets and a growing network of new partners and providers. Digital is powering both: products delivered more quickly and cheaper than ever before are attracting customers to buy online.
But fraud is evolving as well. The simpler days of opportunistic customers and misleading brokers taking advantage of claims have been overtaken by networks of savvy individuals and groups using sophisticated techniques to sniff out and exploit opportunities.
We've seen existing activities weaponized and new patterns emerge like the creation of "ghost" patient profiles created by insiders for fraudsters to exploit in healthcare, for example. The latest twists include things like the use of synthetic data fraud, which brings identity thieves into the mix. Thieves are using stolen and traded identities to create fictitious accounts. More sinister still is the growing co-ordination between individuals and organized crime: staged accidents and false claims are taking place across international borders, according to
At each step, fraudsters are exploiting gaps in information and knowledge — taking advantage of data that's out of date or cannot be accessed in time by officials.
The good news is that, as insurance goes digital, transactions and interactions create a digital footprint for analysis by officials — which opens the door to AI. AI can work on huge datasets to uncover patterns and activities that escape conventional detection.
But AI at this scale demands a ready supply of trusted, quality data — a significant sticking point. Accessing, preparing and managing data for AI is a problem for 40% of firms
Two factors are blocking the path:
● A labor-intensive approach to finding and working with data: engineers are coding data pipelines between systems that must be managed and updated manually as requirements change. Engineers are also expected to clean and classify data to achieve the consistency AI needs for reliable results. The whole process is inefficient, imperfect and slows anti-fraud teams' response to fast-moving criminals.
● The struggle to operationalize data: to make it freely and consistently available to anti-fraud stakeholders. This means teams cannot easily share or reuse data for analysis and it's difficult both to set and enforce rules on data use and to build compliance reports for regulators.
AI can help, but only with the right data
Firms are right to enlist AI in the fight against fraud. Speaking from our experience with clients, AI delivers; one big international insurance firm found a fraudulent billing scheme hidden within batch audits which was inflating premiums and threatening re-insurance agreements. Feeding trusted data into AI models, the company found suspicious patterns in seconds and recovered millions of dollars.
The journey towards machine-driven fraud detection and prevention begins with modernizing inherited data processing environments. Realizing that in a large and data-rich sector like insurance comes in three steps.
● First, speed up delivery of pipelines connecting disparate data sources by squeezing out as much of the manual process as possible from building and managing data connections. Automate the process of creating pipelines and migrating data regardless of data type and location, whether on different cloud provider platforms or on-prem systems. As requirements shift, ensure pipelines can adapt to keep the volume and variety of data flowing into AI models at the necessary pace and scale.
● Next, build a framework capable of establishing trust in data. Much of the data flowing into corporate machine models is incomplete and lacking consistency, which leads AI to produce inaccurate results and outcomes that can't be trusted. Building trust means making sure data achieves standards of consistency regardless of source or type. Data accuracy and reliability can be achieved using tools capable of identifying gaps and resolving inconsistencies from the point it's onboarded right through the claim lifecycle.
● Finally, turn data into a reusable corporate asset. Create an environment where teams can easily find data and understand it to quickly and securely build datasets of the scale and quality demanded. A centralized catalog founded on a rock-solid system of data classification is key: classification should use standardized taxonomy and metadata applied with automated tools and procedures to discover, publish and contextualize the data.
Digital is a double-edged sword in modern insurance: a platform to quickly deliver highly tailored products to customers but a significant opportunity for fraudsters. AI offers detection and monitoring capabilities on the required scale, but it can only do it with the right data. It's time to modernize conventional data processing and get a match fit for AI.
