In the last five years we have seen the emergence of online marketplaces, often referred to as the on-demand economy. Companies like Uber and Deliveroo have transformed the way we access good and services in a manner that was unthinkable just a short time ago. Now it’s possible to take a taxi, shop for groceries, order a take-away and pay back a colleague with a click on a screen.
Beneath this surface of convenience, however, there is a super-complex set of managed relationships between the marketplace and its suppliers. Cab drivers, restaurants, stores, masseuses, and cleaners that make up the supplier network are part of a fragmented group that must be guaranteed payment in order for them to supply their service. This raises new challenges for businesses; the need to be truly instant.
Older ecommerce businesses also claim to be instant but this just refers to an order placed on their website or app being instantly accepted. The essential difference between on-demand and traditional ecommerce is that not only is the buyer’s transaction instantly accepted, but the order is also placed instantly with the supplier. That payment has to be guaranteed and if something is wrong with that transaction - e.g. it is fraudulent - the liability sits with the marketplace. Managing this reality requires a new set of tools along with a new approach to controlling risk.
New business models need new payment and fraud tools
What has become increasingly clear is that many of the tools in place, most of which have been inherited from the ecommerce market, are not fit for the purpose of fighting fraud in online marketplaces.
In fraud detection the big shift is away from manual review and towards a much more data-driven approach to fraud detection. In this new world where the buyer and supplier are connected and the transaction is guaranteed in seconds, referring risky transactions to a human reviewer is a non-starter. An automated approach is the only way to provide correct fraud determinations in the required timeframe. This is the core reason we have seen this shift towards machine learning in fraud detection and the move away from human insight at the point of payment.
A data-first approach: moving beyond manual review
That’s not to say of course that human insight does not play a key role. Machines and the models they run are made much more effective when they are based on intelligence about the specific fraud a business experiences. And essential to really optimised machine learning fraud detection is, perhaps surprisingly, human oversight. In a data-first environment, an analyst role switches from determining whether a transaction is good or bad to confirming whether a decision was right or wrong. This not only keeps the machines ‘honest’ but every definite fraud determination (either confirming it as correct or confirming it was wrong) is highly valuable in helping the models adapt and become more efficient over time.
We are now starting to see these techniques move into more traditional ecommerce businesses. The efficiency, speed and scale arguments that a data-first approach makes hold true even when there is still a window of time for manual review, though it always makes sense to reduce the number of manual reviews to as few as possible. We believe that over time these reviews will increasingly be done post-facto – which means the machines will make the initial determinations and the experts will confirm their accuracy, improving the models to make better decision in the future.
Summary: refocusing human insight for greater efficiency
The world of instant fulfillment needs new techniques if is it not to be overwhelmed by fraudsters taking advantage of the need for instant order fulfillment. These new techniques are coming to market through data-first machine learning fraud detection products like Ravelin. This doesn’t mean the end of human insight however. It simply changes the focus of that insight from approving transactions to confirming decisions taken automatically by machines. This allows fraud teams to scale, to operate with equal effectiveness 24 hours a day, 7 days a week and at peak times. For this reason, the data-first approach is finding adherents in traditional ecommerce and not just in on-demand.
About Gerry Carr
Gerry is CMO of Ravelin, which provides fraud protection for online businesses. He joined Ravelin from its inception to help define and articulate a product vision for the changing face of fraud in ecommerce. Prior to Ravelin, Gerry has led the product marketing functions for products as diverse as Ubuntu and Sage CRM. Gerry loves to snowboard and compete in iron man contests when time allows.
About Ravelin
Ravelin prevents fraud and protects margins for online businesses. Companies all over the world are accepting more transactions with fewer chargebacks thanks to our machine learning-based approach to fraud prevention.
Every day we send out a free e-mail with the most important headlines of the last 24 hours.
Subscribe now