Initials

2024 was a year of great success for us at Pandabase. We successfully proved our MVP and delivered various features that helped our merchants. Over the past year, we launched features like coupons, categories, and teams. Many of these features were shipped in February 2024. But a feature called “Fraud Signals” has been with us since our early days, which was created in mid-2023, around July.

In 2024, we’ve worked to improve Fraud Signals for our merchants. We deployed systems to detect anomalies and block them before they turned into disputes, which were always a hassle. We also took on the responsibility of managing disputes on behalf of our merchants for a small fee of just $0.10 per successful payment on their store.

We renamed it to Radar, a name that we felt like it fit the purpose. The success of our initial versions was remarkable, but they required manual review, involving pattern checking and other tedious tasks that we wanted to automate. We achieved this automation with our new version of fraud monitoring tool built from scratch just to provide proper evidence.

A deep dive

Radar utilizes a range of pattern fraud monitoring techniques. We designed it to allow for the addition of new features without breaking it. While it doesn’t rely heavily on AI, it primarily uses pattern matching and evaluates results from a specially trained LLM.

We have a specialized tracker on our checkout pages to track and analyze user behavior. We record this data in an encrypted and secure manner to ensure compliance with local and international privacy regulations.

This event tracker securely transmits data to our internal data ingestion pipeline, which is built to handle up to one million events per minute.

The event samples look like this:

{
  "event": {
    "id": "r_evt_7lSehM3vs2EDKySk",
    "meta": {
      "type": "checkout.confirm_clicked",
      "timescaped": 1.66,
      "ray_hmac": "2cd8a970b02a4dd.....18d5c2ac6671b453"
    },
    "data": {} // contains various information
  }
}

By tracking what the buyer is doing during checkout, we can detect anomalous behavior. If the payment is disputed, we send this information to the card bank to prove that they did indeed agree to our terms and were present during the checkout; it is additionally verified by 3Ds.

For our buyers

We realized that we were being less buyer-friendly with our new techniques, as there are valid disputes made by buyers. For example, they got the key but it wasn’t working, or the merchant didn’t keep their word. To fight this abuse, we’re requiring customers to contact us directly via our dispute resolution system so that we can get involved and prepare a fair, positive resolution.

Conclusion

Over the past year, we’ve successfully blocked over $10,659.66 (USD) in fraudulent transactions using with the help of manual reviews and automated processes, including 3D Secure. Our ingestion systems, which collect enough information to verify legitimate purchases, have saved our merchants over $3,162.00 (USD) in successful dispute cases. 12.91% of payments on our platform were manually reviewed, and 63.91% of successful payments went through 3D Secure.

We’re shipping manual reviews and more anomaly detection features for our merchants in the upcoming days.