One of the most common myths about AI is that it’s a black box that runs itself — all you need to do is plug it in, let the machine do its learning, and allow the algorithms to take over.
Wrong. Just like any new partnership or team member, AI requires an onboarding process to establish familiarity with operations, time to figure out how to do its job well, and regular check-ins with its manager.
In a recent Forbes article, Joe DeCosmo lays out steps for how fintechs can implement AI into middle and back office operations first, before they go “all in” and extend it to front office (consumer-facing) ops. Pairity’s Machine Learning as a Service is aligned with this incremental approach, and our tiers of service were designed with similar tenets in mind.
For instance, “embrace redundancy and remediation.” Every automated process should be tested against a manual process to make sure it’s doing what it’s supposed to do (only quicker and better than the manual model.) AI is no different, and when we’re onboarding clients to Pairity for the first time, we encourage them to start with one portfolio, worked with both machine learning and a manual process concurrently. Not only does this enable you to see immediate results of your AI implementation, it also allows a firm to understand what decisions are being made and the kind of algorithms that are being built.
“Monitor everything” is another recommendation, which is of course a requirement for staying compliant in more ways than one. For instance, an algorithm is developed based on a set of factors that seem to work, but if that algorithm goes unmonitored for an extended period of time, it could also develop a pattern of unwanted or suboptimal practices that may be tougher to explain to regulators than the technology itself. (Although the CFPB has championed AI for cutting down on discriminatory practices, the possibility is still an example of what could happen when a firm adopts a laissez-faire approach to an investment in AI.)
This is why we take explainable, regular reporting at Pairity seriously. We make sure that our clients can access and generate reports as often as they want, and we make sure those reports are transparent and explainable to different stakeholders within an organization, and therefore its regulators.
While AI is probably the most exciting new tool for fintechs and can easily live up to its hype, it’s not a set-it-and-forget-it kitchen appliance (even a crockpot ultimately needs a human at the helm). We’ll expand on this in a future blog post, but if your firm is considering implementing AI into any part of its operations, you’ll find that half-measures rarely produce big results. Even if you’re onboarding AI incrementally, a dedicated CTO or data scientist within your organization should be charged with managing it: keeping an eye on its activity, communicating tweaks and changes, and maintaining the right controls.