Skip to main content

Prevent Metrics and Explainability

Yaniv Cohen avatar
Written by Yaniv Cohen
Updated over 2 weeks ago

Prevent Goal

  1. help merchants to understand the decision and act fast

  2. gain trust in the system

  3. avoid analysis-paralysis

  4. not overwhelming merchants with data

Value realization metrics

The goal of the main metrics in the Dashboard page is to show the merchant the product coverage and value for them - the bottom line.

  1. Total analyzed transactions value + count

  2. Loss prevented - the sum of direct cancellations and verification-driven cancellations.

  • Direct cancellations - orders where the product recommends "cancel," and the merchant follows through.

  • Verification-driven cancellations - orders where the product recommends "verify," and the buyer fails verification, leading to cancellation.

Assumption - orders approved or successfully verified and fulfilled are assumed safe and don’t contribute to loss prevented

3. Missed savings - sum of the overridden Verify or Cancel recommendations, based on the the transaction outcome (CHBs, dispute, nothing)

4. Potential Risk Exposure - the total revenue tied to active flagged orders (e.g., orders marked for verification or cancellation)

Actor card

Actor risk score

Risk score over time in the last 12 month and shows the trustworthy trend of an actor and Increase the trust in the product, specifically for bad actor with low score over time.

Actor tags

Tags are calculated daily, based on historical data of the last 6 months

  1. Account juggler - A buyer who uses multiple accounts, possibly to hide their identity or engage in fraudulent activity.

  2. Burst buyer - A buyer who makes multiple high-value purchases within a short period, which may indicate suspicious behavior.

  3. Risk area - A buyer associated in a geographic area with a high dispute rate, potentially signaling abuse of return or refund policies.

  4. Frequent disputer - A buyer who uses payment methods that are often linked to higher dispute rates.

  5. Chargeback champion - A buyer with a high success rate in winning chargebacks.

Metrics

Metric

Definition

Why it’s interesting

Thresholds

Dispute rate (across the network)

The percentage of transactions a buyer has disputed across the network, compared to the network average dispute rate.

If a buyer has a dispute rate of 10% when the network average is 0.5%, it’s a red flag they might habitually claim non-delivery or dissatisfaction. Merchants can see this isn’t just their problem, it’s a pattern.

Low risk: 0.-0.002 Medium risk: 0.002-0.01 High risk: 0.01+

Refund rate

The percentage of transactions refunded out of the total transactions, compared to the industry average.

A high refund rate can signal potential risk, such as friendly fraud, where buyers exploit refund policies by claiming issues like non-delivery or dissatisfaction after receiving goods.It directly relates to a merchant’s concern about revenue loss, making it a practical and relatable metric.

Low risk: 0.-0.01 Medium risk: 0.01-0.05 High risk: 0.05+

Dispute timing (time-to-dispute average)

The average time between the buyer’s purchase and when they file a dispute, benchmarked against network and industry norms.

A buyer who disputes transactions unusually quickly (e.g. within 48 hours) or after a consistent delay (e.g. just before the chargeback window closes) might be exploiting policies.

Network normal: 14d≤time≤30d Network anomalies: time<14d, time>30d

Velocity of transactions

The number of transactions the buyer has made across the network in a short time frame (e.g. 5 purchases in 24h). threshold is p.9.

High velocity can signal intent to exploit return policies or dispute charges before merchants catch on (could be more interesting if we can pair it with disputes)

Low risk: 0-2 Medium risk: 2-5 High risk: 5+

Linked entities

The number of entities linked to the actor compared to the network average

If a high number of entities are linked to one buyer this could hint on someone who wants to hide themselves.

Low risk: 0-4 Medium risk: 4-10 High risk: 10+

Return gap (Return-to-purchase gap)

The time difference between when the buyer receives an item (based on shipping data) and when they initiate a dispute or return, compared to typical behavior in the industry.

If a buyer disputes a charge long after delivery, or immediately after, while keeping the item, it hints at friendly fraud.

Actor age

The amount of days since the first time the actor was seen in the network.

Fraudsters often create accounts quickly to exploit systems before detection. A young account with suspicious activity (e.g., multiple purchases or refunds) can be a red flag. Older accounts might suggest lower risk if paired with a clean history, offering merchants reassurance.

Low risk: 90+ days Medium risk: below 90 days

Did this answer your question?