A brief history of automated fraud detection
Manually does it
The mission to detect fraud started off with excel spreadsheets, macros and formulas - yuk! Sometimes the process was even as basic as searching a spreadsheet manually for matches - known as data matching.
What does data matching mean? It is where the claimant is shown as being involved in a claim before. For a long time that was the the focus of fraud detection.
The amount of manual effort involved made this slow, inefficient and prone to error (we called these false positives).
Then came the era of databases and running batch jobs, still based on spreadsheets, but slightly more automated and less labour intensive.
The thinking was very much, whomever had the biggest database of claims was providing the best fraud detection.
What are flaws in this approach?
- The assumption that your fraudster has made a claim or perpetrated a fraud before - giving the fraudster one free 'go' before they start to be picked up by data matching fraud detection methods.
- Data errors (accidental or deliberate) would result in the fraudster, address, postcode, vehicle registration, etc. being missed by inflexible rules and with them the opportunity to detect and prevent a fraudulent claim.
With insurer digital claims strategies, integrated data via web-enabled APIs and the Claims Portal we are more data rich than ever before with quality, quantity and depth to allow greater sophistication in fraud detection.
From a claims data perspective the Claims Portal has been the best thing since sliced bread.
We still do data matching of course - because why wouldn’t you want to know if this claim links to another claim? However, it now only takes seconds to identify these matches and new breed technologies can link systems and databases together in real time.
In Ki we embed the subject matter expertise of our fraud specialists. With our technology, we have been able to evolve our automated fraud detection service. In Ki we embed the subject matter expertise of our fraud specialists and flex the rules to meet unique fraud threats faced by our clients in order to add: fraud indicators and a cross-industry watch list analysis and detection methods to data matching.
For example, Ki is assessing multiple factors in a claim, including but not limited to:
- Time of accident
- Medical attention/treatment
- Vehicle occupancy
- Claim chronology
- Professional enablers
and is doing this across different types of claims and business lines and sectors.
We understand that these rules can have expert knowledge biases - we are looking for the things we know (from our experience) point towards a claim being of a higher fraud risk. Our processes, and those of our clients, then ensure that anything system detected is subject to a qualified, human, assessment.
We are now entering the era of data science. Through the application of statistical analysis, natural language processing, machine learning techniques and coding our R&D data science team, we are, in conjunction with the University of Manchester, developing new and advanced methods to further enrich our data and detect fraud in claims.
Ki learns from feedback on the fraud detection and claim outcomes and moves away from expert bias towards, all the time reducing false positives and increasing the prospects of successfully challenging fraudulent claims.
If you would like to know more about how we are helping our clients increase fraud detection, please get in touch.