A bank’s contact center is a monument to problems that reached the customer. Every seat, script, and carefully tuned handle-time target exists because something went wrong upstream and nobody caught it. David Fapohunda has spent more than 25 years leading operations, AI enablement, and digital transformation across global banking platforms carrying millions of interactions, and he says the industry has spent decades perfecting the machinery of response while leaving the upstream failure largely untouched.
“For decades, banks measured customer service by one number, how fast they answer the phone,” he says. “This is no longer the number that matters.” The institutions pulling ahead ask something harder, how quickly can they tell a customer is in trouble, find the fix, and resolve it before that customer ever has to follow up? Their platforms watch the journey in real time and cure friction in the moment it appears, which is a different discipline entirely from answering the call quickly.
Design First, Data Second
Most teams begin with the data, since that is what a model consumes. Fapohunda starts a step earlier, with the customer problem, and treats everything downstream as an implementation detail. Platforms built the other way get engineered from the inside out, organized around the org chart rather than the person on the other end, and a customer feels that structure as friction. His method reverses it. Begin with the experience the customer should have – a dispute resolved before it escalates, a credit decision returned in seconds – then work backwards to the model that delivers it. “Design first, data second,” he says.
One Pod, One Outcome
The failure Fapohunda encounters most often is organizational. The operator who runs the process, the engineers who build the system, and the AI engineers who train the models sit in separate groups, connected by handoffs. He puts them in a single pod, accountable for the same outcome and owning the redesign of the flow from day one, with customer experience, operational efficiency, and control present from the start rather than bolted on at the end. Separate those functions, he warns, and you get a model that performs in a lab and fails at the front line. The handoff is the defect. Each group optimizes its own stretch of the process while nobody owns what the customer actually experiences.
A Single View Is What Makes Anticipation Possible
An AI engine is only as good as its view of the customer, which Fapohunda considers the point that matters most. Banks tend to treat paper, mobile, web, interactive voice response, text, and live agent as separate conversations. The customer has only ever had one. Bringing that data into a single view, what he calls a customer graph, lets a system meet people in the channel they prefer and resolve issues before they reach an agent. It is also the precondition for anticipating anything, since no system can get ahead of a problem it sees only one fragment of. Bank of America’s Erica has handled more than 3 billion client interactions and delivered over 1.5 billion proactive, personalized insights, reaching customers before they had to ask. “That is anticipation at scale,” Fapohunda says.
Governance Is What Makes the Speed Safe
The value at stake is considerable. Fapohunda cites McKinsey & Company’s 2023 research, which estimates that generative AI could add up to $340 billion a year to banking, as much as 15% of operating profit, nearly all of it turning on how the customer is served. The technology, in his assessment, is not the hard part.
Governance holds the rest together. Early in his career, he led controls that cut fraud losses by millions, and the lesson has only sharpened. Models that scale are the ones that are explainable, fair, and defensible to a regulator. “Governance is not the brake,” he says. “It’s what makes the speed safe.”
Built this way, an AI servicing platform does more than cut costs. It earns trust, and trust is the return that compounds. To learn more about building AI-powered servicing platforms, connect with David Fapohunda on LinkedIn.
