Why the quality of referrals matters more than the volume

Insurance fraud detection has changed significantly in recent years. Insurers now have access to more data, better analytical tools, and increasingly sophisticated technology for identifying unusual claims activity. While this creates more opportunities to spot potential fraud, it can also create another problem for already-busy fraud teams: too many alerts to review.

Insurance fraud teams only have so much time available for investigation, so increasing the number of referrals does not necessarily improve results. If a large proportion of those referrals involve low-risk or legitimate claims, investigators can end up spending valuable time on cases that ultimately go nowhere. The number of alerts generated is therefore only one measure of performance; the quality of those alerts and whether they lead to useful investigations matters far more.

The cost of false positives

Fraud detection often starts with something that falls outside an expected pattern. It could be a claimant with a previous claims history, an unusual sequence of events, a shared address or telephone number, or a supplier that appears across several apparently unrelated claims. These can all be useful indicators, particularly when several are present, but on their own they do not necessarily mean fraud has taken place.

Where rules are too broad, or models have not been sufficiently refined, they can produce large numbers of false positives. This increases the workload for investigation teams and can result in genuine claims receiving unnecessary scrutiny. It can also make prioritisation more difficult, particularly when a potentially high-value or organised fraud case is sitting in the same queue as dozens of much weaker referrals.

For this reason, a system generating 1,000 referrals is not necessarily performing better than one generating 500. If the smaller number contains a higher proportion of cases that genuinely warrant investigation, it is likely to be delivering considerably more value to the fraud team and the wider business.

Giving investigators the right information

The usefulness of an alert also depends on the information that sits behind it. Knowing that a claim has been flagged is not enough if the investigator then has to start from scratch to understand why.

There may be links to previous claims, or the same vehicle, repairer, address or contact details may appear elsewhere in the insurer’s data. In other cases, several fairly minor indicators may become much more relevant when they are viewed together. Bringing those connections into one view gives the investigator a clearer starting point and helps them decide whether a case is worth pursuing.

This is where connected data and analytics can add real value. Rather than simply identifying something unusual, they can provide more of the context surrounding the claim and help investigators understand the relationships and behaviours that led to it being referred.

Deciding which cases to look at first

Not every referral carries the same level of risk, and it does not make sense for every case to receive the same priority. Investigation teams need to be able to focus their resources where they are most likely to have an impact.

Risk scoring can help with this by taking into account factors such as financial exposure, the strength and number of fraud indicators, previous activity and possible links to wider organised fraud. A relatively low-value claim with one weak indicator may require a very different response from a high-value claim with several links to previous suspicious activity.

This becomes increasingly important as insurers bring more data into their fraud detection processes. Having more information available is useful, but only if it helps teams reach decisions more quickly and identify the cases that need their attention.

Learning from investigation outcomes

What happens after a referral has been investigated is also important. If particular alerts regularly result in no further action, that should help inform the way future referrals are generated. In the same way, when certain combinations of indicators repeatedly lead to confirmed fraud, that information can be used to improve rules and models.

Investigator feedback is particularly valuable because fraud patterns do not stay the same. New methods emerge, organised networks change their behaviour, and rules that once worked well can become less effective. Regularly reviewing investigation outcomes allows fraud detection to adapt rather than relying on a fixed set of assumptions.

It also means the experience of investigation teams becomes part of the process. Analytics can identify patterns across large volumes of data, while investigators bring the knowledge and context needed to understand which of those patterns are genuinely significant. Used together, the two can improve the quality of referrals over time.

Looking beyond alert volumes

Alert volume is straightforward to measure, which is one reason it is often used as an indication of fraud detection performance. On its own, however, it does not show whether a fraud strategy is working effectively.

It is worth looking at measures such as the proportion of referrals that progress to investigation, how many result in confirmed fraud, the value of losses prevented, the amount of investigator time spent assessing referrals, and how quickly higher-risk cases are identified. For some insurers, reducing the number of unnecessary referrals could be just as valuable as increasing the amount of suspicious activity detected.

The most useful measures will vary depending on the insurer, the types of fraud it is seeing and the resources available to its investigation team.What is important is understanding whether the information being produced is helping investigators make better decisions, rather than simply measuring how much activity the system generates.

Improving the quality of fraud detection

Insurers have more data available to them than ever before, but having access to that data is only useful if it helps fraud teams make better decisions. Effective fraud detection should make it easier to identify the cases worth investigating, understand why they have been flagged, and see connections that may not be apparent when a claim is looked at in isolation.

At kbs Intelligence, this is what we help insurers do: use their data more effectively so fraud teams can focus on the cases that are most worth investigating. The aim is not to create more alerts, but to make the alerts investigators receive more useful.