Ai Editorial: Taking calculated risks via machine learning to boost sales

First Published on 7th October, 2016

Ai Editorial: If airlines adopt a risk-averse approach to managing fraud, then sales can suffer tremendously. Ai’s Ritesh Gupta explores this issue

When one talks about the role of machine learning in managing fraud, there is one question that immediately crops up.

How does machine learning take the onus and deliver – in terms of liability shift as well as handling fraud and boosting sales? The industry is already talking about 100% chargeback protection i. e. getting entirely refunded for any unauthorized transaction not getting detected.

As it turns out, a huge problem with the traditional rule-based fraud solutions and reliance on manual reviews lies in a risk-averse approach to managing fraud. Methods like these are overly focused on bringing down the fraud rate as close to zero as possible, and tries to prevent the first chargebacks from happening. When this happens, sales suffer tremendously.

Firstly, fraud managers would rather not take the risk of accepting a borderline transaction (which could be genuine), resulting in much greater false positives. At the same time, rules deployed (location based, amount based, time based, etc) limit genuine users from making transactions.

In addition to the effectiveness in detecting fraudsters, with machine learning, the system understands when to skip rules when positive behaviour is detected. Furthermore, an optimized algorithm (another form of machine learning) allows the system to optimize and make the most of all the transactions that are seen as part of a portfolio. Based on calculated risks, the system passes the optimized number of transactions while ensuring that chargeback rates are still under control. As a result, borderline genuine transactions can be passed and unnecessary rules and bans are lifted, improving sales greatly.

What hampers sales?

According to Justin Lie, Group CEO, CashShield, a SaaS based self-learning fraud prevention solution for ecommerce, around 3-4 years ago, airlines were reluctant to speak about big data and machine learning as they were still very reliant on payment gateways to handle fraud. However, in recent times, fraud has evolved to become much more complex, and airlines have increasingly come to understand the importance of fraud management to gain competitive advantage and optimize sales.

“We are definitely seeing a positive trend of airline companies gaining back control of their payment options, flow and procedures in the industry, and they are more and more knowledgeable about putting together the various pieces of puzzle to enhance performance,” Lie says.

The use of rules and manual reviews hamper sales and are not the most effective form of managing fraud, added Lie.

“When airlines move away from traditional methods, they must be comfortable with automating most or all of the fraud systems, which means that they can redirect resources to more important areas and focus on their core business, and also allows them to scale up operations much easily while keeping the cost managing fraud under control,” said Lie.

From detecting fraud to predicting it

When using traditional methods of detecting fraud (deploying hard rules and manual reviews), it is often based on analysing the standard fields (name, address, email, IP location, fingerprint and what can be found on the order form) and what transactions have passed through the hard rules. The problem here is that those standard fields and hard rules are extremely easy for fraudsters to manipulate and get passed once they have figured the rules in place. For example, it is now easy for fraudsters to generate hundreds or thousands of new fake emails, and once they realise that a time based rule (no more than 3 transactions in an hour) is in place, they will try to write their program to attack the system with 3 transactions per hour each time. Not only so, genuine customers are likely to be blocked. For instance, a geo-location rule would block customers booking transactions from ‘riskier’ locations.

Moving towards machine learning allows airlines to remove all these unnecessary rules that would have otherwise blocked genuine customers. The combination of big data and machine learning allows more effective fraud prevention. To simplify what has been said about big data and machine learning, big data is first used to collect information about the user’s behaviour on the website (how the mouse moves, what he likes or puts into his wishlist, etc), and this information is combined with machine learning, which uses pattern recognition to map the pattern of his behaviour to match it either with positive (genuine) or negative (fraudulent) behaviour, as well as predictive analytics that records the positive/ negative behaviour and uses that on future transactions for potential signs of fraud. Lastly, an optimized fraud risk algorithm should be used to make decisions on whether or not to accept a transaction based on calculated risks to best optimize sales while controlling fraud and chargeback rates.

Automating fraud analysis

Since the information and data that each airline collects are different (including their web structure and payment options), airlines should refrain from using a one size fits all solution. Instead, they should consider using fraud solutions that cater and adapt to their industry and business model.

Rather than collecting as much data as possible, the quality of the data and how the airlines use the data for better decisions in fraud prevention and increasing sales is much more important.

As for machine learning, it often encompasses different types, and simply using one type (predictive analytics) is insufficient. Merchants should learn to discern and understand the different types of machine learning, and be sure to know if the fraud solution uses only predictive analytics or covers more bases with more than one kind of machine learning. To “improve” machine learning, or rather just to get the best out of machine learning, businesses should deploy solutions that use more than just predictive analytics, or upgrade to a solution that uses predictive analytics, pattern recognition and optimization if they are still using traditional methods of preventing fraud.

Follow us on Twitter: @Ai_Connects_Us