The potential impact and role of Artificial Intelligence (AI) is undeniable. We increasingly see highly repetitive tasks moving from human manual labour, to machine-based automation. Fraud prevention is often labelled as one of the ‘quick wins’ for AI, and there is no dispute that the technology’s use is driving better detection.
However, AI is also enabling improved accuracy in determining what is fraud and what is genuine behaviour. An often-overlooked aspect of fraud prevention is the human aspect of fraud operations, and the role that team plays in fraud prevention, in even the most advanced, machine-driven environments.
Fraud operations is referred to in many ways; alert management, fraud escalations, case investigations, or the fraud desk. The single key activity that unifies all these names, is the activity of manual reviewing cases that are highlighted for investigation. The outcome of those investigations is an assessment of whether a case is acceptable or not from a risk perspective.
In organisations that process millions of customer transactions or interactions, daily, it is often the fraud operations team that is the closest team to these events, via their manual investigations. The modern equivalent of the ‘coalface’.
Sometimes an activity that has been highlighted to fraud operations may not be fraudulent. It may be that a customer is using the product or service in a way that is against the conditions of usage. Manual review currently continues to remain on the front-line of business decision-making, despite advances in AI technology.
As a result, I do not believe the fraud operations team will be disappearing anytime soon. Yet, the role they play in many organisations is changing. The future role of the team will be to act as a key human validation component in the machine led approach.
We are not yet in a perfected world of AI. There are still numerous instances, where the recommendation from AI systems are either not conclusive, or simply not trusted. Manual review or validation is therefore required, before a final decision is made that could impact business and ultimately an organisation’s bottom-line.
Even as dependence on AI solutions increases, there will always be the need for validation or calibration of the classification models that AI relies on. Not least to ensure that customer inconvenience or friction in the payment process is minimised. To assure the future role of fraud operations, however, it is critical that these teams embrace several core principles to ensure they evolve with the move to a machine-led environment.
Knowledge and specialism
Fraud operations is not a low-skilled job. You cannot pull people ‘off the street’, give them a decision matrix sheet, and set them to work. Simple decision-tree based assessment is low-skilled and it can, and should be, automated by machine-driven solutions.
True fraud operations teams need to be skilled and specialised in their particular area of expertise. They are required to build a picture based on evidence, step back and view that evidence in its entirety, and make a reasoned judgement: “is this explainable or unexplainable activity?”.
Organisations need to support their fraud operations by providing training, to ensure that team members understand the holistic environment and can therefore make more rounded decisions. Improved skill and specialisation should not be mistaken as a drag on productivity. As well as higher quality decisions, a properly trained fraud operations team may be able to work quicker based on more certainty in the thought process.
Process and tools
Fraud operations is a human-driven function, therefore maximising the capability of the human-element is critical to success. The study of behavioural science helps us to understand the benefits and fallibilities of humans as decision makers and enables the creation of process to maximise performance. For example, recognition of decision fatigue as a potential issue can result in regular rotation of team members’ focus. This pro-actively avoids a deterioration in quality of decisions. Providing the fraud operations with sophisticated workflow and investigation tools can also benefit productivity.
Benchmarking and review
Just as machines require validation and calibration, so to do humans, when they are making decisions. Benchmarking performance is a core requirement for Fraud Operations, to highlight areas of excellent and areas requiring further focus.
Having a strong and open framework for ‘wash-up’ review, to examine both good and bad decisions, is critical to performance evolution. Review of decisions can help defeat biases in decision making and drive consistency in decision-making.
Communication and domain knowledge transfer
Interaction with stakeholders is a critical success factor for fraud operations; whether that stakeholder is a data scientist, fraud strategy manager or business product owner. The real value of fraud operations is to supply qualitative data into the data-driven decision environment. There is a trove of valuable insight and information that can be gathered at the coalface, that can be pushed back out to the organisation to add value.
Equally though, it is the transfer of domain knowledge into the team from other areas that can help to improve both decision quality and holistic understanding of the wider business. This enables fraud operations to contribute effectively in the next-generation fraud management process.
The inference is clear for all fraud prevention leaders and organisations that are currently evaluating their next steps in machine-led fraud detection: don’t forget to both review and invest in your team at the coalface. The evolved fraud operation could be the human validation component that helps to realise the immediate benefit of your wider investment in new technology.
By Matthew Attwell
Risk and Analytics Services