See the forest and the trees – using analytics to detect warranty fraud
Is warranty fraud an issue for us or not?
In our earlier blogs (“The Dark Side of Warranty Management”, “What is needed to control warranty costs and prevent fraud?”) we have referred to multiple client cases and several research studies, estimating fraudulent warranty claims to occupy up to 15 percent of the average company’s warranty costs. However, averages are averages. Do you know the real situation in your company? In many cases we hear, that “warranty fraud is a major issue, but not for our company” or with slightly more awareness “we have a feeling there may be something wrong, but we don’t know the scale of the problem”. In any company with a large number (tens, hundreds or even thousands) of customers or service agents, a certain proportion of them will try fraud and some of them will do it in an excessive manner. Analytics, if done right, can be very impactful in getting the picture of the situation, either by uncovering anomalies, suspicious behavior and direct malpractice or giving confidence, that the situation is under control.
Just like looking at the attached pictures (taken from a very short distance), where you can admire the beauty of the tiniest detail, as well as the entire flower, in warranty fraud analytics you need to take a look at the individual claim level, as well as global and regional averages and everything in between. Analytics can also be used as a part of the transactional process and support the decision to accept or reject each individual claim, but in this writing we focus on analytics as a way to understand the current situation and the potential leakages, which should be dealt with.
Why is warranty fraud analytics so effective?
In a common example, if I ask you to go to the next room, toss a coin, come back and tell me the result – heads or tails, it is impossible for me to know, if you really tossed the coin or if you just tell me a random result. However, if I ask you to go to the next room and toss the coin 100 times, the situation is different if you want to cheat.Even with such a simple arrangement it is not easy to invent a statistically plausible series. How long should the longest chain of similar results be? How many times should there be 2 or 3 in a row before changing? How long should the longest alternating series be with every second heads and every second tails? In a similar way, it is quite easy to create a few fraudulent warranty claims, but it is very difficult to execute large scale warranty fraud and remain statistically consistent with the reality.
Typically, fraudulent warranty claims, whether invented or inflated, leave a mark in warranty data, which can be detected as an anomaly in analytics. With statistical analytics you can identify individual claims, which have a high likelihood of being incorrect and service agents or customers, who have a high number of these suspect claims and therefore are candidates for further investigation.
For instance, if a valid spare part is used in a repair, the related claim is accepted. If there are multiple claims with the same spare part, each of them is also accepted. If the same part has been used in most or all of the repairs, the rule-based claim validation still considers each of the individual claims to be acceptable, even if such a widespread failure in one part would be highly unlikely.
No differentiation is made between cases where the part has been used in 1 percent, 20 percent, or 100 percent of the repairs. What looks like a valid claim at individual level can appear fraudulent when looked at an aggregate level. Analytics is needed to evaluate claim validity across individual claims. This can be targeted or general. In targeted analytics you know a certain fraudulent pattern and want to verify its existence with a specific customer or service agent. In general analytics you slice and dice the data in different ways, try to identify anomalies and outliers and then understand whether the reason for the anomaly is fraud or something else.
What is needed?
For effective warranty analytics you need the right data and systems to capture data, store data and retrieve it for analytics and other usage.
- Consistent warranty claims data – including customer, product, service center and cost related data
- Installed base data – which products are sold and delivered to which customers, under what type of warranty or service contract
- Warranty master data – such as customer and service agent master, warranty offerings, service bill-of material, fault and symptom codes, claim entitlement and validation rules.
Additionally, the right tools and right people are needed, people with a combination of analytics expertise, contextual understanding of warranty fraud and the right attitude, willingness and persistence to spot anomalies and dig deeper to find the rationale for them or detect suspect behavior. In our book, we have created a framework for warranty fraud, looking at the actors, victims, motivations and methods of fraud.
- Actors committing fraud can include any party in the warranty chain either alone or in collusion with other parties, typically customers, service agents or warranty providers
- Victims of fraud can also include any party, typically warranty providers and customers
- The motivations fall broadly into two categories – service cost avoidance (I have a faulty product and want to get it repaired free-of-charge) and revenue increase (claiming and re-selling parts or products, excess charging of existing or non-existing warranty service)
- The methods are numerous – new methods are invented as old methods are uncovered and blocked.
Understanding these combinations helps out in defining the processes and procedures to detect and avoid warranty fraud. In many cases the actual analysis can be quite simple and straight forward, when you know what to look for. With more sophisticated analytics methods, you can then detect the more complex fraud schemes and even find cases you don’t necessarily know about. The simple way can be to benchmark service agents in terms of proportions of different types of repairs, spare parts used, cost-per-repair, product age at the time of failure etc. In one case we noticed one service agent having 80 % of the claims of a certain service type, whereas the regional average for the same product was only 35 %. In the end it was verified, that the true proportions for this particular service agent were the same as with the others, the whole difference was due to fabricated warranty claims. In another case we noticed a peculiar peak just a few weeks after the products were sold. That particular service agent was also a sales channel partner and wrote down product serial numbers at the customer delivery, just to make a bogus warranty claim for the same products very soon after.
Why is it so difficult?
Why is it so difficult for companies to recognize warranty fraud or do something about it? By its nature, fraud is hidden and its exact scale unknown, so many companies are simply not aware of the problem. Some fraud cases are clear – claim data is not correctly filled or a validation rule is violated. However, it is always possible to create an individual fraudulent claim, which can’t be caught by looking at the individual claim alone. With analytics you can identify suspect claims and also notice patterns and anomalies, suggesting fraud. For many companies there is quite a big hurdle from that initial suspicion and seeing something really strange in warranty data (something you only see with large scale fraud) to really believe something is wrong and take confident action.
Also, it is a very sensitive topic. “We don’t want to talk about warranty fraud, since the term fraud implies intent, which is an overly harsh statement,” said one client executive. “Yes, the numbers from this dealer look really peculiar, but I cannot go and accuse them of being dis-honest”, said another. People don’t want to upset their customers or business partners and are reluctant to asking questions implying they suspect something. Understandably, companies do not want to go public if they have been victims of fraud. Therefore, the issue remains under the radar for other companies.
The third reason we have seen in many clients is simply the lack of focus, skills and discipline in warranty transactions and analytics. On the high level all processes and policies — validation rules, claim process, statistical data analysis, customer entitlement — you name it, seem to be in a good shape, but the devil is in the details. When you dig a bit deeper, you notice big holes here and there. Warranty control is often a part time role and the last activity between the person responsible and the weekend. High turnover of and limited induction to the validators often leads to a situation, where the processes and controls are defined, but not followed properly. Issues with timeliness and quality of warranty data. Issues with warranty data analysis and identifying fraud related anomalies.
The combination of the sensitive nature and the difficulty to produce water-tight evidence keeps companies hesitant to take action. This keeps them vulnerable and some individuals or companies are ruthlessly taking advantage of the situation.
Wish to get clarity on your own situation?
How about you? What is the level of awareness, the maturity of warranty control process and the amount of warranty fraud leakages and other over-charges in your company? If you think this is not an issue at all, you might think again. If you feel there might be something, but you really don’t know, we can help you. Provide us with sample data and we will run a set of analytics and present the findings to you – providing either confidence and peace of mind (you are in a good shape) or a call for action with clear savings potential.
Author : Matti Kurvinen