Ravelin was joined by BlackLane, Hello Fresh, Get Your Guide, Delivery Hero and Sixt mydriver in the beautiful SO/Berlin Das Stue Hotel in Berlin. Ravelin has run these roundtables pretty much since our inception. We have finally taken it on the road where we were privileged to host some of the leading lights from Berlin’s bustling e-commerce industry.
The conversation and discussions were driven by the participants and it was certainly a lively discussion. Our broad theme was trying to understand what a good relationship looked like between a fraud detection vendor and a merchant in 2018 and beyond. How much of the investigations and analysis of fraud does a merchant want to retain in-house, and how much are they expecting to consume as decisions by the vendor?
It’s a difficult topic to know where to draw the line. Some in the room were strongly of the opinion that whatever happened, the process had to be automated. There was no room or time for human intervention in a decision. Even where there was that possibility, ‘reviewing’ a transaction is not a desirable process.
However that is not to say that the businesses did not want to understand why decisions were reached and of course they want some mechanism or mechanisms to be able to influence future decisions in order to get them right.
This perhaps excited the most conversation; some felt very strongly that a vendor which continually asks a merchant to make a call on uncertain transactions was not doing their job. Others were of the opinion that understanding this uncertainty was something that they wanted to investigate on their side as it gave them a deeper understanding of how the fraud impacted their business.
There was an interesting point made that perhaps the answer was not that it belonged on one side or the other but that while a merchant has a deep view of their own business, customers and fraud, a vendor has the benefit of a much broader set of data to investigate.
At the core of the discussion though is that for any of these scenarios to work there needs to be trust between the vendor and the merchant. This trust is based of course on a proven ability to keep chargebacks and acceptance rates at agreed levels. But even deeper than that, it is a proven ability to explain how a decision was reached. And deeper again an understanding to show an ability to discover and react to an emerging fraud threat either through a model update or some other way.
This then led to a brief discussion of the place of rules. Everyone at the table used rules in some way but fewer at the core of fraud detection. However, it was strongly felt that rules were a great way to express policy, and interestingly good backstops while trust is being created.
For instance, while a merchant learns to trust the decisions of a new fraud prediction model it needs to protect the business from some worst-case scenarios - so creating a rule that allows new accounts to have a maximum order value makes sense. In a few months, based on real data, it might be the case that the rule is not needed but the willingness to trust that decision can only be gained over time.
Over the two hours we touched on a range of other topics including the viability of data being shared across merchants, frustrations at the inability of banks to either enrich data or to provide chargeback information efficiently, and the merits of a 64 degree egg (excellent).
If you’d like to find out more about fraud in the on-demand economy and how Ravelin can help, download our free online guidebook.