Our client is one of the largest short term home rental marketplace in the world. Fraudsters were using fake IDs and blurry selfies to bypass algorithms. After getting onto the platform they would defraud our clients’ customers by trying to sell inventory that wasn’t there, or block inventory from being sold without the intention to rent.
We worked with our client to add a human judgement element to this process to reduce the amount of bad actors and will create a safer environment for hosts and renters. We would double tag each selfie and escalate if there was no consensus.
Our work led to approximately 50k double labels each month with an escalation of about 10%. Over 12 months, there is a sub 1% escalation as TTEC has built a library for proper IDs around the world and best ways for annotators to recognize fraud faster. Each review takes under a minute, whereas the processes initially took seven. Our desire is to continue to annotate in order for escalations to go to single digit percentages and for the model to have a 99% accuracy score.