If AI Says 87%, Tell Me Why: Explainable AI and the Payments Clock

Quick Summary

Natasha Lapierre, who heads innovation and product strategy for financial messaging at Bottomline, explains why AI is becoming essential to payment ecosystems. Payments are getting faster, domestically and cross-border, and every decision made once a payment starts has to be accelerated. The answer today is not handing decisions to AI but using it to speed up people, built on two principles: human in the loop and embedded explainability.

Why is AI becoming essential rather than a nice-to-have?

Because the pressure is mounting, Lapierre says. Things are happening faster and payments need to be instant, not just domestically. There is an increasing expectation that cross-border payments will speed up too. Simple task automation is not enough to keep up with that shift; the bottleneck has moved to the decisions that have to be made while a payment is under way.

What happens once the clock starts ticking?

Lapierre describes the moment a payment is initiated as the point where the clock starts ticking. From then on, friction builds and decisions sometimes need to be made quickly:

• A sanctions alert needs resolving
• A fraud alert needs a decision
• A payment investigation needs to be carried out

All of those count against the clock, and all of them need to be accelerated. That is where AI becomes critical.

Should AI make those decisions itself?

Not yet. Lapierre is candid that automating decisions to AI is going to take a while, given banks' appetite. The battle that can be won today is about the ways AI can accelerate human decisions by surfacing the right information. If a fraud alert is raised, the person reviewing it needs to know exactly why, so they can make an effective decision.

That rests on two core principles. The first is human in the loop, which is where the industry is today. The second is embedded explainability. As Lapierre puts it: if AI tells me 87%, what do I do with that? AI needs to say why it is 87% and where the data comes from, so that what it provides is directly actionable.

FF NEWS TAKE:

The 87% line is the whole argument in one sentence. A risk score without a reason does not save an analyst any time; it just moves the investigation. In instant and faster cross-border payments, where every second after initiation counts, explainability is not a compliance nicety but the thing that makes AI useful at all. Vendors who cannot show their working will struggle to get past the pilot.