How To Protect Your Incentive-Based Marketing Strategy From Fraud

With the increasing popularity of incentives-based acquisition strategies in competitive markets, fraud is rampant. How can you prevent fraud at scale using Machine Learning and Data? Let’s have a look!

Incentives-based acquisition strategies can be really useful for growing and scaling up your businesses. However, where there is free money, there is a risk of fraudulent activities. Scale-up and growth businesses can’t afford to burn their valuable marketing budgets on fraudsters. Spending hundreds of man-hours and thousands of dollars on risk operations can become unfeasible in the long run. Is there a more efficient way to prevent the marketing dollar from going to fraudulent and unhealthy acquisitions?

There actually is — enter the tech, data, and AI to save the day! 🙌

So sit back and relax while I paint the picture for you to understand how the fraud “industry” works in today’s e-comm world and how our artificial heroes help us slay them! 😉

Context: What does incentive fraud mean in the e-commerce industry?

When you hear fraud, what’s the first thing you think about? High-profile financial scandals, stolen credit cards, or… free food? I know right! Crazy as it may sound, the “free stuff” fraud is actually quite a big problem in the e-comm industry, besides the payments fraud.

There are even professional fraud rings, coordinated attacks, the deep dark web, and a lot of “tips and tricks” on TikTok and youtube to show people how to commit these seemingly harmless fraudulent activities. First things first, they aren’t completely harmless — this guy went to jail for it. So do not try this yourself at home!

And secondly, they cost the business a lot more than it brings value — either leading to “false” acquisitions or a lot of low-margin acquisitions that churn away. Here are some statistics to give you an idea of the size of the fraud industry for e-commerce.

Why is Fraud prevention challenging?

There are simply too many ways to defraud a system. Fraud prevention is an endless cat and mouse chase, it never ends and it keeps evolving. So the job is never done. It is a pretty interesting problem to detect and prevent fraud for multiple reasons:

1. By the nature of the problem, it is just difficult to know what is fraud and what isn’t? If a user simply churned after the first discounted purchase or if someone is exploiting the incentives by creating multiple accounts. So, we never know how big is the actual size of the whole fraud problem.
2. In some cases, it might be easier to identify fraud for example in cases of users complaining about stolen credit cards. In others, it simply isn’t possible. You wouldn’t trust the prime suspects in the crime thrillers — the same way we can’t go around and simply ask if the user was defrauding the system. So we have unlabelled data, to begin with.
3. Even if we do manage to generate some labels for training the machine learning models, which metric do you pick to optimize the system for? It is tricky because you don’t have reliable ground truth data. The metrics are subjectively defined by you and hence bring bias to the whole system. For example, having 0 false negatives might just mean we don’t have fraud the way we define it, not necessarily that there is no more fraud in the system.
4. Localization: Local behaviors in different markets lead to different methods of committing such fraud. Some markets allow cheap sim cards that enable users to create multiple accounts which is unlikely in other markets. Some markets are cash payment heavy and others have credit cards as a primary payment method — both would require a different approach to prevent fraud to ensure a smooth user experience. Hence one solution can’t be applied to all markets, there has to be a scope of context configuration.
5. Data Privacy regulation: Although data privacy regulations allow you to use user data for fraud prevention solutions, a lot is lost with deleted cookies, account resets, and incognito windows. It adversely affects account deduplication accuracy, making fraud detection even more difficult — often inflating the growth numbers.

Well, despite the challenges, there are a few things you can indeed do to keep the incentive fraud in control. Let’s look at some of those.

How to build your incentive fraud prevention strategy?

• Incentive Design: By designing incentives in a way that incentivizes the user to come back to the platform to get the reward can help filter out the fraudsters out there only to redeem vouchers. Restricting acquisition vouchers for the first order only and designing retention vouchers to be redeemed over the consecutive set of orders is one way to do it. Another way is to gamify the incentives to collect beans, coins, or points to redeem later in future orders.

• Account deduplication: Most of the acquisition voucher fraud can be redeemed by creating a new account. So fraudsters create multiple accounts on the same device or using the same phone numbers. This can easily be prevented, *if* you invest in a good solution for account deduplication in (near) real-time(eg. a graph database to link accounts using multiple criteria or a probabilistic matching of the accounts based on heuristics and machine learning). Not only does it help in preventing one of the biggest incentive frauds but it also helps get a better sense of your acquisition numbers and growth.
• Fraud Scoring: If you have invested in a good data and analytics infrastructure for your product, this is a natural next step towards making the system more secure. Using behavioral analytics, the fraudulent user activities can be identified from the genuine user behavior since it’s an anomaly from the norm. For example, the time from login to order might be significantly lesser for the fraud user since they know the user flow already and are not exploring the platform.
• Flow Friction: The fraud scores bring us to the deliberate user flow frictions that can be used to add an additional layer of security for medium-risk users. High-risk users can directly be rejected the incentives while low-risk users can be let pass so as to not affect the user experience adversely. Simple friction like verifying the phone number or email ID before placing an order can help reduce fraudulent activities significantly.
• Refund claims: Again, based on the user behavior identified as fraud and their refund claims activities, you can be proactive about identifying which new users are trying to take advantage of the claims option on the platform. This is on top of the reactive policy of not allowing more than x number of refunds on n number or orders or y$ of order value.
• Blacklisting: Even though the fraudsters can go around creating new accounts, they can be stopped by simply blacklisting incentives for high fraud probability email domains, country codes in the phone number, geo hash locations, or IP addresses. This of course has to be figured out based on data and past behaviors on the platform flagged as fraud.
• Shaping Product policies: Lastly, instead of fixing the gaps with short-term solutions, look at your analytics data to drive product and incentives policies to ensure the fraud prevention efforts have a long-term impact. For example, not allowing incentives for accounts with non-local numbers, disincentivizing redemption of vouchers on the web by increasing minimum order value or increase in flow friction can be product policies to prevent fraud on the platform.

Easier said than done, like everything. But with every single effort, you make the job more difficult for the fraudsters and hence move towards healthier business growth. And with every new loophole they find on your system, you can keep evolving your product policy and fraud prevention system as soon as you discover those loopholes.

Conclusion

Fraud estimation, detection, and prevention is hard problem. It also is worth solving since it affects business growth, user experience, and marketing efficiency significantly. We might not have all the answers to solve the challenges we face, but we sure can try to evolve our approaches and our prevention efforts to make sure we aren’t letting the fraudsters get too far away in the cat and mouse chase. And in this effort, make our product much more secure for the users and more efficient for the business. 🙌

I write about Tech Products, Data Science, Product Management, Productivity, and Leadership. If you would like to read along with my learning journey, you can follow me on medium or LinkedIn. If you would like to catch up and talk about things in detail, you can book 1/1 call with me here. 📆