IBM Counter-Fraud Point of View #BDA13

June 6, 2013

in Analytics, Data Mining, Decision Management

Share

Another solution session, this time Bob and Rick from IBM focused on IBM’s Signature Solution for Counter-Fraud, Waste and Abuse.  This is a classic Decision Management use case so I am looking forward to this one. Most fraud, they point out, is committed by insiders and goes undetected for 18 months. Frauds are also becoming a major source of funding for more serious crimes, making this an increasingly serious security issue.

IBM has experience countering fraud in banking (AML and KYC as well as credit card or mortgage fraud), insurance (claims fraud), government (security and anti-terrorism) and healthcare (medicare and disability fraud). Any company can see identity theft, improper payments, procurement or expense fraud etc. Fraud detection has great ROI too, with the US Government for instance being able to get $7+ in savings for every $1 spent in addition to the potential for avoiding brand damage. Meanwhile technology is giving companies new tools to fight fraud.

Fraud is  intentional, its illegal, its about making a financial gain and is deceptive. It’s often associated with abuse or waste (though there may be different elements here), with threats (where there’s no monetary gain) and financial crimes (where the company is focused on compliance rather than worried about a direct loss).  Fraud can be committed by people who do fraud for a living and involving organizations or providers or it can be more opportunistic (individuals or employees).

Different approaches are required to catch different kinds of fraud – organized crime groups tend to understand that rules are being developed to catch them but can be detected by their networks while opportunistic fraud can be caught more easily using rules. And you have to keep working at it because countering fraud is a bit like whack-a-mole, with fraud popping up where you are not focused any given moment.

Four elements to a counter-fraud solution:

  • Detect
    Detect the fraud within a business process to secure and protect those processes.
  • Prevent
    Take action in real-time, before the transaction is acted on. A Decision Management problem to predict and anticipate the fraud.
  • Discover
    Find fraud within the data to manage fraud and report on it.
  • Investigate
    Confirm fraud for prosecution, determine rules and watch lists, learn and apply.

The first two tend to be in-line, decision management problems while the others are more about back-office analytics. All of this requires both underlying capabilities and domain/industry experience (specific rules or modeling approaches for instance). It also requires multiple  analytical techniques must be applied too, everything from content (comments about claims for instance) and entity analytics to predictive analytics to forensic analysis (social network visualization) and retrospective analytics (descriptive analytics).

IBM tries to get customers not to see fraud as a point solution or a single score at a particular step. Organizations need to see fraud as preventable, predictable, provable and countered across the process lifecycle. It requires an ecosystem, Big Data for new data sources, transparency through guidance and alerting, multi-layered and “smart”.

For example, a claims fraud scenario:

  1. There’s a first notice of loss, a claim, entered into a claims processing system
  2. Entity analytics figure out who is who, what relationships they have , if anyone is a known or suspected as a fraudster etc
  3. Decision Management applies business rules, predictive analytics, entity analytics, anomaly detection and more to optimize a fraud decision, flagging the ones that need to be investigated (potentially also fast tracking those that are clean). Rules for this step are continually refined in ongoing discovery and leverage IBM’s industry experience.
  4. Suspicious claims are then managed as cases, investigated using social network analysis and other tools that interact with the entity analytics and other data to build a rich picture for the investigation.
  5. Once fraud is clearly identified, additional rules and flags are pushed into the decision engine.
  6. Finally a set of fraud-centric reports and dashboards are provided to manage and report on the whole process.

Nice use case showing how new kinds of entity and social network analytics, decision management, case management and more combine to provide a powerful counter-fraud framework. If detecting and managing claims fraud is important to you, check out this white paper on next generation claims processing.

Share

Previous post:

Next post: