Below you will fine three snippets of thought from our in-house Data Scientists and our Fraud Experts.
The graph below is a visual representation of the state of play as far as fraud detection is concerned.
The two axes showing Complexity and Business Value show visually how as the complexity of the processes increases, so does the per unit value it brings to the organisation. It is also worth noting that as the Complexity increases in those processes, so do the associated costs to the business.
Below is a breakdown of those individual stages:
Reporting: what happened?
Tools employed: Word, PowerPoint etc.
Analyse: why did it happen?
Tools employed: Excel, SQL etc.
Monitor: what’s happening now?
Tools employed: Dashboards, Alerts etc.
Predict: what might happen?
Tools employed: Machine Learning, AI
Simulate: what will likely happen? – This is where we want to be
The X on the graph shows where the industry finds itself today in terms of fraud prevention – we are entering the Prediction phase, and we are likely to be in that phase for a while yet. The Simulation phase is where we are striving to be one day – essentially, we want to be able to employ Simulation to know what will happen, before it happens.
So, let’s clarify the difference between Prediction and Simulation with a simple analogy:
You can predict (with a fair degree of accuracy) that one of your colleagues, who regularly drinks at least 1 cup of coffee per day, will drink a coffee today. This is an example of a prediction, and functions solely as a simple extrapolation of historical data.
Simulation allows you to predict the likelihood that your colleague will drink a coffee today, not only by taking into account the historical data directly regarding their coffee drinking habits but also by considering all possible contributing factors – what time did they get into work this morning; How much demand is there for coffee in the office; etc.
Additionally, simulation combines all these contributing factors and predicts alternative outcomes, calculating their likelihood: If my colleague doesn’t drink coffee today, there is a 35% chance that he will leave work early.
Prediction is essentially a predictive list (score) which one can react to. Simulation on the other hand is where a multitude of individual predictions (of varying aspects) are used to simulate all possible outcomes in order to more accurately predict those outcomes. When Simulation reaches this stage our fight against fraud will stop being reactive in nature – instead it will become pre-emptive.
Dr. Samiul Ansari, Head of Data Science
Fraud analysis is identifying people involved in fraudulent activities, and discovering criminal trends in order to prevent them from occurring in the future. Even though typical rule-based fraud detection systems are performing well, they only work with previously seen fraud activities and they do struggle with unknown fraud.
Also, only transactions are labelled as fraud, or the probability for a transaction to end up as fraudulent activity. What it doesn’t tell us is why, and how, a particular transaction will end up as fraud, which inherently makes the decision-making tasks more difficult.
Ultimately, it is all about revealing the real picture behind the data. This is where modern data visualisation fits in, also known as graph visualisation or link analysis. Graph based visualisation methods help in uncovering criminal links that would otherwise be difficult to detect. Visualising the data as a graph allows insight into the associated fraudulent network and can pinpoint where the fraud lies by exploiting the relationships between various entities such as people and events.
At TruNarrative we are working towards both graph-based link analysis and graph-based visualisation in order to reduce referral rates and improve fraud detection.
Dr. Sulaiman Khan, Senior Data Scientist
With the invention of new technologies, such as contactless payments, you will always find articles trying to either expose the fraud weaknesses or reports that fraud losses are trending up. With all new technology there will be inherent weaknesses; there have been recent reports that fraudsters can now contactlessly read card details employing a “contactless skimmer”.
While yes this is true, and in theory they can, that does not paint the whole picture. Firstly, the fraudsters would need to know how to de-encrypt the message from the skimmer in order to see the details of the card number (before being able to use that number to commit fraud); which is no easy task. Secondly, there are also other factors to consider, like the fact that the contactless limit is currently set at £30 – for the effort required to commit a fraudulent contactless payment, the reward for the fraudster is actually quite small. Whilst CNP card fraud certainly does exist, fraudsters will more often than not opt for the path of least resistance.
Statements like “recent figures from UK Finance showed that contactless card fraud overtook cheque fraud in the first half of 2017, hitting GBP £5.6 million” are there for impact and to gain clicks. When you dig a little deeper this figure equates to 0.02% of total contactless transactions, not to mention that the usage of cheques as a payment method is at an all-time low.
When it comes to fraud strategy, as long as you are constantly reviewing your controls and updating them in line with fraud trends and risks, you will stop the vast majority of attempted fraud. The teams that struggle to combat fraud are the ones which are not updating their controls, and not reviewing their fraud trends. You will never stop every single instance of fraud, but fraudsters tend to take the path of least resistance – so the harder you make if for them, the less likely they are to target you.
Nick White, Fraud Manager
Find out more about how Graydon and TruNarrative work together here.