60

Fraud detection analytics

What is it?

Fraud detection analytics is the process of uncovering fraudulent actions or behaviour so that you can then predict fraud and reduce or stop it.

This type of analytics looks at vast amounts of data to identify patterns or certain behaviours that flag fraudulent activity so that processes or systems can be changed to prevent fraudulent activity.

Why does it matter?

Fraud detection analytics matters because it can help you to identify patterns of behaviour or actions that are the precursors to customer or employee fraud, and therefore stop it before it happens.

Fraud costs many businesses a great deal of money every year – money that could be bolstering profits and allowing the business to grow. And it can happen in every company from the fraudulent use of an expense account to customers making fraudulent insurance claims to credit card fraud by organised criminal groups. Online fraud is also a growing area and every business needs to be vigilant – small or large. In fact, often small companies are particularly vulnerable because they assume that the criminals will be targeting larger businesses. This is a mistake because often smaller businesses are easier targets for criminals who can, for example, remotely access your network or manufacturing systems and demand a ransom before giving you back control. Clearly analysing your vulnerabilities is incredibly important so you can stay one step ahead.

When do I use it?

How often you conduct fraud detection analytics will depend on the nature of your business. Credit card companies and insurance firms are assessing for fraud on a constant basis. If an individual makes a purchase on their credit card that is unusual, for example, then they will usually get a phone call from their credit card company almost immediately. This is because algorithms have assessed your normal credit card activity and geographic location and anything outside those and many other parameters raise a red flag. For example, if your credit card is used in London at 11 a.m. and then used in Glasgow at 12 noon this is clearly fraudulent activity because it’s not possible to get to Glasgow from London in one hour.

Even if you don’t operate in high-risk fraud areas such as finance or insurance you should conduct fraud detection analytics at least every six months.

What business questions is it helping me to answer?

Fraud detection analytics helps you answer business questions such as:

  • Are any of our customers engaged in fraudulent activity against your business?
  • Are any employees engaged in fraudulent activity against your business?
  • Are there any warning signs or patterns that indicate fraud is imminent so that we can intervene and stop it before it happens?

How do I use it?

Activity, written and spoken data offer a rich vein of information from which to conduct fraud detection analytics. You can, for example, use CCTV footage to monitor warehouses and picking and packing areas. Video analytics (Chapter 11) can be applied to this data to extract insights, while voice recordings can also be used for analysis (Chapter 12).

For example, customer service calls are usually recorded; voice analytics can identify stress levels in a customer’s voice which may sometimes indicate fraud. This type of analysis could be especially helpful on insurance claim helplines. When someone is scared, stressed or lying, they exhibit vocal clues that can be picked up by voice analytics. Of course the person may be stressed because they’ve just had their home burgled, but this type of technique can at least flag those cases that need a closer look to root out those that are lying and minimise insurance fraud.

Text on forms, emails or social media can also be used for text analytics (Chapter 8). And data mining (Chapter 6) and correlation analysis (Chapter 3) can be used to identify patterns of behaviour or connections between seemingly random activities that could indicate fraud.

Practical example

Insurance companies use data mining and factor analysis on their online application forms and were able to find a correlation between the time a customer took to fill in their online claim form and fraud. This can be indicated by either filling the form in too slowly or too quickly. Often when a customer takes too long to complete the form, or hovers over a field for too long, they are thinking too hard about what happened or what they should write. This can indicate they are not being entirely truthful about the event.

Of course this is not the only assessment – the insurance companies allow for the fact that sometimes the person may just be slow. Perhaps they were interrupted by someone at the door, or they need to take their children to ballet class or they are older and not very good with computers. But this data will raise a red flag that is collated with other data points such as how many times a person changed the data in a particular field. If too many red flags are raised then the insurance assessor knows to look more closely at the case.

Conversely if the form is completed too quickly this can also raise alarm bells. Criminals often use bots (i.e. an automated web robot) to complete forms, or they will cut and paste from previous claim forms making the completion process very quick.

The same insurance company also monitors how we fill in application forms online. This can show them how often we retype data into certain boxes to attempt to get a better quote when we say our car is parked in a garage rather than on the road. Big data analytics tools are now watching out for this type of behaviour and will flag any potential fraud.

Tips and traps

Fraud detection is a constantly evolving area because the ways people perpetuate fraud is constantly evolving. Don’t decide on patterns or red flags and simply then test for them – you need to run full-scale analysis fairly frequently to identify how fraud is changing. Also, ask your employees and customer services people to be on the lookout for new and novel ways that customers are finding to defraud the business.

Once you find fraudulent activity, also run some data mining on those cases to see if you can identify patterns that you can then use to prevent future fraud. For example, if there any demographics data that would indicate a ‘type’ of customer to steer clear of?

Further reading and references

For more on fraud detection analytics see for example:

..................Content has been hidden....................

You can't read the all page of ebook, please click here login for view all page.
Reset
18.191.172.180