“In Two Weeks, False Positives Disappeared Completely”: How a Ticketing Platform Fights Fraud
Online event ticket sales is a burgeoning industry, but it also comes with a bevy of challenges. We talked to José Cabanes at Zentense, which offers web application development and event technical production, about how his team beats fraud on the platform.
Tell us about how fraud affects your business…
Our customers produce events, and they sell tickets for those events through our platforms.
The most common type of fraud we face is criminals buying tickets with stolen credit cards. What at first seems like a valid transaction, becomes a chargeback a few weeks or months later when the legitimate credit card owner raises a claim against that transaction. Sometimes, the transaction is cancelled and the merchant loses that money.
There’s also a second layer to the damage: tickets that are bought fraudulently are usually resold at low prices – and that raises concerns for the event organizer because, from a marketing perspective, that lowers the event value.
What was your fraud management process before implementing Sift Science?
Before Sift Science, we were fighting fraud on our own – and it was tough to stay ahead without the proper tools. We realized that fraudsters had more resources than us!
We would manually review transactions, looking for something suspicious. Fraudsters tend to follow similar patterns until something requires them to change strategies, so patterns can usually be spotted by looking at the raw data – that is, if you know where to look.
But with a large volume of transactions, manual review is very time consuming. The type of events we sell tickets for usually show an exponential selling pattern – in other words, ticket sales go up massively during the final days before the event. Fraud management was really hard during those spikes. Thousands of transactions had to be processed each day.
Tell us about the process of integrating and using Sift Science
It was pretty easy. The Sift Science API is clear and well documented. One engineer can do it in a few days, depending on the level of integration required (there’s more than one way of using Sift Science).
Once we integrated, we were impressed. The web interface is very intuitive and clean, the API is bug-free, and it’s very well documented. It’s a very good user experience.
How long before you saw meaningful results from Sift Science?
After integration, once real data was flowing, the system showed meaningful results. I guess it depends on the volume of data that is generated each day, but in our case, in one or two weeks fraud false positives disappeared completely and the system was scoring orders and users accurately.
For us, the greatest benefit of Sift Science is being able to detect fraudulent orders early and automatically. Whenever Sift Science detects a fraudulent transaction, I get an email – so I don’t need to proactively check orders.
Can you describe your experience with machine learning?
Machine learning can seem very intimidating at first, but it works reliably and your model keeps training through regular feedback. When I was at university 20 years ago, I had some contact with AI research. At that time, it seemed like it was very far away from achieving interesting results, and I moved on to other fields.
Now I see that AI is a thing to be taken very seriously. I also fear the day that fraudsters begin to use AI against us, but I hope that groups like Sift Science will help us always stay one or two steps ahead in the game.
