A Conversation on Fighting Policy Abuse With An “Identity-First” Approach

Meet Maya 

Protecting online retailers from policy abuse is not as clear cut as tackling traditional card-not-present fraud. To meet this challenge head on – along with its nuances – we’ve enhanced our Policy Protect solution with a capability we’re calling Identity Explore. This engine works by forming identity clusters from billions of accounts, behaviors, and transactions across a global merchant network, thus broadening the view of customer identity beyond their singular profile with a merchant.

To get an insider’s look at this new capability, we sat down with Maya Kerem 🎗️ the Product Manager behind Identity Explore. Maya has been with Riskified nearly three years and has spent that time focused on our identity resolution capabilities. Prior to Riskified she was in the B2C space, which gave her insight into merchant pain points which she’s able to hone in on to ensure that our products always offer our merchant partners the best solutions to their needs. 

Here is our conversation with Maya: 

Q: Can you tell us about Identity Explore and why this capability was needed in the market? 

Identity Explore allows merchants more (and much needed) visibility into their customers while also managing and preparing for future interactions. It allows merchants to see the cluster of accounts that are linked to a specific customer, how many claims, chargebacks or orders they have, as well their buying patterns. 

This robust information allows merchants to decide on things like blocking customers from receiving additional refunds or using promo codes. In some cases, merchants may even use this information to block certain customers at checkout. Having the ability to make these kinds of decisions, at the customer level, is important for merchants, and something they would not be able to do with basic linking capabilities. 

Q: Do the differences between cluster technology and linking matter? Why or why not?  

The simple answer is yes, the differences absolutely matter. In fact, there are two significant distinctions between cluster technology and linking. The latter answers the question: “Does THIS ORDER belong to the true cardholder?” while cluster technology takes it a step further and answers the question, “Who is the true IDENTITY behind this order?” 

Machine Learning cluster technology is the most advanced approach in tackling policy abuse because it considers what additional accounts are connected to any given identity, it’s capable of dynamically tracking aggregated identity usage, as well as understanding which connections are relevant and which to ignore.

Q: Why doesn’t linking work for policy abuse? 

When tackling policy abuse, it’s critical to consider how accurately the individual customer is represented. While linking shows a limited segment of a customer and isn’t comprehensive enough for policy abuse – because it doesn’t provide merchants with an “identity-first” approach – clustering does provide a robust and holistic view of your customer.

And to be clear, linking is ideal for traditional fraud analysis as it gives merchants insights from the perspective of a specific order while evaluating the anomalies in the related customer behavior patterns. 

But when it comes to policy abuse, we find that looking at one order is not good enough. Instead, to efficiently take on this challenge, we need to consistently look at all interactions, at the same time, and evaluate commonalities in behavior, not just anomalies.

Q: What are the use cases for Identity Explore and its clustering technology? 

Refunds:

Identity clustering helps merchants detect refund abuse/abusers and flag/deny abusive refund claims (INR, missing, damaged, etc.), flag checkouts as high risk for abuse, and block or apply friction to serial abusers at checkout.

Excessive Returns:

Pinpoint excessive (potentially unprofitable or suspicious) customers and apply friction at claim submission, and/or at checkout (add fees/warn/block).

Return Scams:

Uncover return scams (e.g., wardrobing or rocks in a box) and prevent pre-inspection refunds, deny future claims, and/or decline future orders.

Unprofitable Returns:

Once we know a customer is not abusive, we can also determine the profitability of each return request and offer good customers to keep specific items (low margins/unstackable items) without inviting or encouraging abuse.

Bulk Resellers:

Flag or block entities or groups of entities with high volume orders and/or frequency to enforce item-limits, as well as combat sophisticated drop-shipping, reselling, and triangulation schemes.

Launch Resellers:

Uncover individuals (using bots and reseller rings) hiding behind multiple accounts to automatically block/flag orders exceeding specific limits.

Sign-up Promos:

Block individual entities from redeeming multiple promotions associated with first-time or unique sign-ups.  These could be related to sign up coupons, referrals, free trials and more.

Promo Reuse:

Detect when a single entity is using coupons from multiple linked accounts and/or stacking multiple different coupons per order.

Investigating Identities To Validate/Determine Friction

Offer enhanced visibility into our identity mapping and decisions/recommendations while seeing all orders, claims, promos in one holistic view. While this alone may not be a use case -this is a significant part  of our offering and can open many doors.

Q: Lastly, can you tell us about the team you worked with to build out Identity Explore? 

This is such an exciting product to be working on, and it was a true collaboration between our innovation, development, data science, business insights, marketing, and design teams. And of course, we worked hand in hand with our merchant partners who were kind enough to give us their time, feedback and insights to ensure that Identity Explore met all of their needs. 

For more information on policy abuse and Identity Explore visit our website.