Understanding Chargeback Decisions and Training Period
Learn how comprehensive and accurate data sharing is essential for fine-tuning fraud detection models
Understanding How Fraud Decisions Are Made
When merchants opt-in for chargeback protection services, there are a few key pieces of information that Coinflow needs to pass along to our provider. The accuracy of the fraud decision-making models heavily depends on the quality and completeness of the data merchants share with us.
For example, if a merchant opts-in for chargeback protection but has limited historical transaction data or small purchase volumes, it will take more time for the data models to adjust to the merchant's customers' behaviors. Or, if a merchant provides too large a range of transaction sizes, but has a low risk tolerance, the models are likely to make a more stringent decision in its fraud analysis.
During onboarding, merchants are expected to fill out a comprehensive questionnaire covering various aspects of their business operations. This includes:
✅ Estimated Gross Monthly Volume
The anticipated total transaction volume per month.
✅ Average Cost of Goods
The typical price range of products or services sold.
✅ Expected Transaction Sizes
The average and range of transaction amounts.
✅ Customer Geolocations
The primary geographic locations customers will make purchases from.
✅ Historical Transaction Data
Detailed records of past transactions, including any previous chargebacks.
✅ Previous Chargeback Approval Rates
Historical data on chargeback occurrences and approval rates.
✅ Ideal Training Period
Preferred duration for the initial training period of the fraud detection system
(e.g., How much time are you willing to allow the data models to learn your customer purchase profile).
✅ Risk Tolerance
The level of risk you are willing to accept
(e.g., Are you willing to accept more false positives vs. more false negatives).
✅ Known and Trusted End-Users
A list of customers who are verified and trusted.
Examples of what customer data is helpful to optimize these fraud models include:
- Location
The specific country, city, state where the trusted customer is located. - IP Addresses
The IP addresses used by the trusted customer for past transactions. - Personal Identifiable Information (PII)
The emails, wallet addresses, or any PII for the trusted customers. - Device Info
Details about the devices used (eg., smartphone, tablet, desktop). - Additional Data
Any other relevant data that can help in identifying a safe and reliable customer.
Training Period
Implementing effective fraud prevention measures requires an initial training period where the system learns and adjusts to your customers' spending patterns.
During this training phase, there may be some false positive decisions, which may cause approval rates to be initially lower, especially if there is limited transactional data available. This period is crucial for calibrating the system to distinguish between legitimate and fraudulent transactions accurately. Our team will monitor all transactions, but your input is essential in identifying a fraudulent user that makes a purchase.
In order to bring approval rates in line with merchant expectations quicker, we have multiple options available to you to override chargeback protection decisions during this training period.
Updated 2 months ago