Machine Learning – Failing Fast

July 8, 2021 - Reading time: 3 minutes

In the last instalment of our article series on machine learning we touched on the human elements and the AI interventions of the system we utilise to make our business efficient. The logical discussion point from that perspective is the risk-based approach we employ in our product functions.

The benefits of machine learning (ML) using a risk-based approach mean we can process many transactions quickly and on a wide-scale in a way that is largely automated, and this allows for a form of real-time document processing to take place, which in many ways speeds up the verification and approval process beyond human capabilities as it allows users our system to receive near-instant feedback as their documents are processed.

While our system parameters which define what an acceptable document is are sufficiently broad enough to deal with enterprise-scale volumes, they are also stringent enough to catch the things that should not go through. In the context of our risk-based analysis using ML “doubt” really just means “don’t.” And we adhere to that principal closely to maintain a standard of quality within our system.

In the validation phase, if the user provides us with an unacceptable document we can then reject the content quickly in a manner we call “fail-fast” which keeps the processing pipeline fluid, allowing a user to be informed that they must send us a new document while they are engaged in the process. We can keep users more satisfied in this manner as the document turn-around time is quick and the feedback instantaneous. And so to fail-fast when there is doubt is fundamental to our approach when using ML in our business.

While machine learning powers the volume and scale of the risk-based approach, it is the interplay between both the AI elements and our human users that determine the efficacy of the process. Since the ML side of the operation can match the engagement of the users in real-time it keeps the document queue numbers as low as possible as the majority of users complete the verification steps on their first session. If we were to rely only on a human back-office for these functions we would not be able to match the speed and accuracy we presently enjoy, and that is the power of this amazing technology.

#intergreatme #remotekyc #kyc # digitaltransformation #onboarding #knowyourcustomer #regtech #fica #rica