Companies today face a rapidly evolving landscape of fraud attack types such as account takeover and credential stuffing. Bad actors are constantly changing their tactics to find new ways to commit fraud, leaving businesses challenged to stay proactive and vigilant against malicious attacks. As stolen personally identifiable information (PII) becomes increasingly prevalent, fraudsters can often impersonate good customers and bypass authorization.
But this tends to be a one-sided approach to identity verification. Knowing what your customers are doing is just as important as knowing who they are, which is why Ekata built the Identity Network – a proprietary database of 200M monthly cross-border and cross-industry transactions provided by customers, to analyze and identify patterns of how identity elements are used. Verifying the validity of identity elements as well as understanding how they are being used online provides a more holistic picture of the physical identity behind a digital interaction.
The Network Score
Network Score is a machine learning prediction that provides insight into how risky a digital interaction is, based on activity patterns of the identity elements that are being used. The top activity patterns Network Score evaluates are:
- Transaction velocity – the number of transactions an identity element has been used within a given timeframe (ex. An email address in 1000 transactions in the last 90 days)
- Merchant popularity – the number of businesses an identity element is used at (ex. An email address used across 1000 businesses)
- Attribute volatility – the number of times an identity element and transaction identity attributes change in transactions (ex. An email address used with 1000 shipping addresses)
The Network Score provides a predictive indicator based on these activity patterns to help you identify good from bad customers and reduce fraud as well as false declines.
The score ranges from 0-1 with a higher number indicating riskier activity. It has precision up to 3 decimal points; for instance, a 0.901 is riskier than 0.899. The Network Score can be used in both machine learning or rules-based fraud models.
The score is trained using 147 Network single dimension features such as email or IP address as well as pair features such as the combination of email and IP address seen together in the last 90 days. These features were all carefully selected based on their predictive power.
Benefits of the Network Score
- Volume: Network Score is trained using billions of transactions
- Intelligence: Since the Network Score focuses on the top behavioral patterns of the Network (velocity, popularity, and volatility), it is a powerful signal that can be used in conjunction with other features to identify legitimate vs. fraudulent activity
- Breadth: Network Score provides insight into cross-border and cross-industry fraud patterns outside a business’s internal data
- Adaptive: Our machine learning models grow stronger and adapt over time to learn new patterns as more transactions are added to the Network
Early Testing Results
In early testing with select customers, we have found Network Score increases the precision in detecting fraud as well as confirming good customers (reducing false declines). Some of the results of our testing below:
[fusion_table fusion_table_type=”1″ fusion_table_rows=”” fusion_table_columns=”” hide_on_mobile=”small-visibility,medium-visibility,large-visibility” class=”” id=”” animation_type=”” animation_direction=”left” animation_speed=”0.3″ animation_offset=””]
|Identifying false positives||Multinational beverage retailer||Increased precision in identifying false positives (worth up to $500k revenue/month)|
|Auto-acceptance of “safe” orders/ bookings||APAC travel marketplace||Increased total transaction value by 20%|
|Transaction fraud||North America marketplace delivery service||Network Score alone accounted for 22% of the increase in recall Ekata data provided to the customer|