3. www.olsps.com
OVERVIEW
• Segments all incoming short term insurance claims
• Ranks claims on likelihood of fraud (0-1)
• Low risk = Fast track
• High risk = Investigation
Low Risk Claims
Fast Track Channel
Medium Risk Claims
Normal Assessment Channel
High Risk Claims
SIU Channel
0 1Risk Score
4. www.olsps.com
• Direct Claims to the correct assessment channel at First Notification of
Loss (FNOL). Basic channels are:
• Fast Track (lowest cost processing channel)
• Normal Assessment Channel (medium cost processing channel)
• Forensic Assessment/SIU (high cost processing channel)
• The Goal is Two-Fold:
• Increase the proportion of claims that are fast tracked, thereby reducing
processing costs
• Improve profitability of forensic unit
• Spin-off is increased Client Satisfaction
OVERVIEW
5. www.olsps.com
SIMPLIFIED LOOK AT THE MECHANICS
OF THE SOLUTION
CLAIM VARIABLES SEGMENTATION OUTPUT
• Driver’s profile
• Type of car
• Time of accident
• …
• …
• …
Business Rules
Predictive Models
+
Low Risk
Medium Risk
High Risk
7. www.olsps.com
CONVERSION OF TWO DIMENSIONS
OF RISK TO ONE
Highly
Decreased
Decreased None Increased
Highly
Increased
None 0.1 Low Risk 0.2 Low Risk 0.4 Low Risk
0.5 Medium
Risk
0.6 Medium
Risk
Low 0.3 Low Risk 0.3 Low Risk 0.4 Low Risk
0.5 Medium
Risk
0.7 Medium
Risk
Medium
0.5 Medium
Risk
0.5 Medium
Risk
0.6 Medium
Risk
0.7 Medium
Risk
0.8 High
Risk
High
0.7 Medium
Risk
0.7 Medium
Risk
0.8 High
Risk
0.9 High
Risk
1.0 Send to
SIU
PREDICTIVE ANALYTICS SCORE
JUDGEMENTRULES
SCORE
0 1
Min
Max
13. www.olsps.com
PROBLEM DESCRIPTION
• Large insurers often use in excess of 1 000 MBRs to facilitate repair
• Costs and capabilities of MBRs vary greatly
• Detailed information available on claims & costs
We need to develop a way to leverage the information that we have in
order to unlock the cost-saving potential
POTENTIAL FOR COST SAVING
14. www.olsps.com
MARK UP BASED APPROACH
• Mark up % on each component of
repair is obtained from each MBR
• Score is calculated for each mark
up %
• Overall discount offered by MBR
is also considered
• All components are combined into
an overall score
EMPIRICAL APPROACH USING
PREDICTIVE ANALYTICS
• Each repair done by each MBR is
considered over a time period
• Repairs are ‘normalised’ so that
different repairs can be compared to
each other
• Repair cost per MBR is averaged
over all its repairs
• Average repair cost per MBR is
used to obtain relative cost
effectiveness
TWO METHODS TO TACKLE PROBLEM
15. www.olsps.com
MBR 1
• Labour cost = R500/hr
• Fairly allocates labour to jobs
• Repairs parts where possible
MBR 2
• Labour cost = R500/hr
• Over-allocates labour to jobs
• Replaces all parts with new
POSSIBLE SHORTCOMING OF MARK
UP APPROACH
MARK UP BASED SYSTEM SEE THESE AS EQUIVALENT
INTRODUCING ROOM FOR EXPLOITATION
PREDICTIVE APPROACH USES ACTUAL COSTS CHARGED
– IMPOSSIBLE TO EXPLOIT
VS
This slide is rather complex to understand, but it conveys the overall benefit of the solution as a whole. It is important to understand that the value proposition from this solution comes from two areas: 1. fraud is prevented, saving the insurer the cost of repairing claims; 2. low risk claims can be sent down a lost cost fast track route, saving a lot of money on processing costs. The larger saving will normally come from the second area. This slide indicates this graphically.
The x-axis on the graph represents all of the claims submitted to an insurer during a set time period. These claims are ranked in descending order of fraud score (fraud score = likelihood of claim to be fraudulent). The y-axis represents the ratio of actual fraud % to average fraud %. Therefore, we would expect claims with a higher fraud score to also score higher on the ratio of actual fraud to average fraud.
Now, without detailed knowledge of the fraud score of a claim, an insurer would be represented by the blue line. i.e. there’s not a lot of difference in the ratio of actual fraud: average fraud between claims that they think are highly likely to be fraudulent vs claims that they think are unlikely to be fraudulent. However, if the insurer had a reliable system to estimate the fraud score of a claim, then they would be represented by the red line which shows a marked difference in the ratio of actual fraud: average fraud between claims that they think are highly likely to be fraudulent vs claims that they think are unlikely to be fraudulent.
If this is the case, then the value of this fraud detection solution lies in the light blue arrows on the graph. The vertical arrow on the LHS of the graph shows the saving on the Forensic or SIU segment of the solution. If the blue and red insurer both chose to send 5000 claims to their forensic unit, then the actual fraud % in the red insurer’s claims will be much higher than the actual fraud % in the blue insurer’s claims. The horizontal arrow on the RHS of the graph indicates the saving on the Fast Track segment of the solution. If both insurers made a decision that all claims with an actual fraud % of <60 can be sent down a low cost, fast track segment, then the red insurer would be able to send all the claims after the 23 000 mark on the x-axis of the graph to SIU. On the other hand, the blue insurer would only be able to send claims from the 35 000 mark on the x-axis of the graph to SIU. Therefore, more of the blue insurer’s claims would need to go to a higher cost assessment channel.
This slide demonstrates the intended outcomes after implementing the claims segmentation solution. The number of Fast Track/Express Claims (depicted in green on the diagram) should increase as the insurer can now safely send many more claims down this low cost assessment route. This will also have the effect of many more claims being processed quickly which increases overall customer satisfaction.
At the other end of the scale, there may also be an increase in the number of claims sent to the SIU channel. One of the many benefits of the claims segmentation solution is that the number of claims being sent to any channel (including SIU) can be calibrated. Therefore, we will calibrate the solution so that the correct number of claims will get sent to SIU depending on the size of the SIU department within the insurer. Either way, the claims sent to SIU will all be of a higher quality (more likely to be fraudulent) due to the nature of the solution which will result in increase repudiations, thereby saving the insurer money.
Finally, because of the increase in fast track claims (and the possible increase of SIU claims) this will mean that fewer claims will get sent to the ordinary assessment route, which will result in reduced handling costs in this department as depicted in the blue section of the diagram.
Segmentation can be applied to supplier allocation as well. OLRAC SPS has developed a model that can do this.
This is very simplistic. There are obviously processes to deal with this type of exploitation, but it is difficult to monitor. The statistical system completely eliminates the need for this.
This is an example of how the solution works. We analyse all of the claims over a recent claim period and look at the normalised cost which each MBR charges to repair a typical claim. This normalised cost takes into account all factors such as severity of repair, type of car, age of car etc..
Each dot on the graph represents a MBR with the actual cost of an average repair on the y axis and the predicted cost of the average repair from the model on the x axis. Therefore, the 45 degree line through the graph represents any MBRs that have an actual cost = predicted cost. All of the MBRs have been divided into ten categories based on their vicinity to the line. Any MBR above the line has an actual cost > predicted cost. Therefore, they are charging too much. Any MBR below the line has an actual cost < predicted cost which means that they are charging below average.
The goal of the solution is to identify the MBRs above the line and either move them below the line, or else stop utilising their services so that all of the MBRs being used are either close to or below the line. This solution enables the procurement department within the insurer to have a objective reason to drive down the price of any MBR above the line. The effects are show in the next slide.
In this slide, the yellow highlighted bit indicates the MBRs which need to move below the line in order to reduce costs. In this example which was done for a large South African insurer, the goal was to move category 1 MBRs down to category 3, category 2 MBRs to category 4, category 4 to category 6, and category 5 to category 6. By moving these MBRs in this manner, the estimated saving to the insurer amounted to $3.3m per year, based on them servicing 55 000 motor claims per year.
This use case is fairly self explanatory. If you read the more detailed document on this solution there is quite a detailed description of the process.