The fintech space is full of inspiring stories about entrepreneurship and CX disruption. And the story of PowerPay is no exception.
Founded by Mike Petrakis, it started life as a two-person start-up on a mission to improve the customer's borrowing experience in three ways: deliver loan decisions responsibly and at speed, be there at the point of purchase, and do it all without processing fees. As a business model, such a USP would always have to lean heavily on automation, but PowerPay's real differentiator would be where, why and when humans stepped in to add the je ne sais quoi.
The idea quickly proved its worth and PowerPay scaled into a US-wide, fully owned, proprietary fintech platform. Today it counts more than 200 employees, has more than 12,000 provider partners, and has processed more than US$8 billion in loans to more than 75,000 borrowers.
AI has played a key role in PowerPay's growth, but it isn't used to deflect the 1,000 daily inbound servicing calls PowerPay receives. Instead, it's powering real-time decisioning while live, human support is used to build the relationships that drive customer retention. For Petrakis, that's just sensible operations, but he says most other organizations "don't actually know how to design workflows that capture AI's benefits".
In this interview with CX Network, Petrakis talks about AI guardrails, the big mistake most companies are making with AI, and the challenges fintech still needs to overcome.
CX Network: In which scenarios is AI a great fit for customer service and where does it fall short?
Mike Petrakis: AI is strongest on the front end of volume: routing, FAQs, and the repetitive 60-70 percent of inbound questions that don't need a human's response. PowerPay is building an AI system to handle a share of its roughly 1,000 daily inbound servicing calls, because most of those calls are the same handful of questions asked a thousand different ways.
AI shines in pattern recognition, instant lookup, and being available at 2 a.m., when customers have time after finishing their day jobs and taking care of loved ones to finalize a loan.
Where it falls short is judgment. AI cannot read the nuance in a borrower's voice when they explain why they missed a payment.
It cannot weigh unusual personal circumstances or negotiate a resolution that balances a lender's need for recovery with a borrower's ability to pay. In lending specifically, that's not just a service gap; it's a compliance requirement.
The moment a conversation turns emotional or ambiguous, it needs to land with a person.
CX Network: If an organization is deploying AI agents in service, what skills do the human agents need to thrive in a hybrid workforce?
Mike Petrakis: The agents who thrive are the ones who get comfortable being the escalation point, not the first line. That means stronger judgment and de-escalation skills, since the easy repetitive calls are gone and what's left are the harder, more emotional cases. It also means being familiar with what AI has already done before the call reached them, so they're not starting from zero.
Think of it as digital when convenient, human when critical. It's more important to know when to override the technology rather than operate it.
Give agents the training and authority to step in when they feel AI's suggested path is wrong. Employees who see AI as a collaborative partner rather than a threat consistently outperform and become the strategic layer of the team.
CX Network: What guardrails should practitioners give AI agents in service to ensure actions align with expectations?
Mike Petrakis: Start with data quality. Most AI failures trace back to messy or ungoverned data, not the model itself. Feed the system real, representative data and stress-test it intentionally with adversarial testing before launch.
Build in compliance from day one rather than retrofitting it. In lending, AI should never make the final call on anything involving hardship, dispute, or ability to pay; a human has to be in that loop as the check against fair-lending and UDAAP risk. Start with inbound-only automation before ever considering outbound, since that raises separate legal exposure. And keep a clear override path so a human can step in the moment a conversation moves outside the AI's lane.
The last guardrail is philosophical. The goal for AI is to amplify the person on the other end of the interaction, not replace them. Every design decision should be tested against that standard.
CX Network: How would you asses fintech's success embracing AI for the customer journey?
Mike Petrakis: Unevenly. There's a lot of noise and not much discipline. MIT research found 95 percent of generative AI pilots fail to produce measurable business value, and companies are abandoning AI initiatives at twice last year's rate.
PQ
The gap is in the strategy. Most companies are trying to use AI to cut headcount rather than to make their people better at their jobs, and that's backward.
The fintech players getting this right share three habits: they define a measurable ROI upfront rather than deploying AI for its own sake, they buy proven vendor solutions instead of building in-house (vendor implementations succeed roughly twice as often as internal builds), and they position AI as a real-time assistant to the human.
CX Network: Which customer journeys are benefiting the most from AI right now?
Mike Petrakis: Point-of-sale and loan origination is the clearest fintech win. It's instant and has embedded credit decisions at the moment someone needs financing, whether that's a US$20,000 home remodel or a hearing aid, with approvals landing in milliseconds.
That's where AI's speed advantage directly translates into a better customer experience with almost no tradeoff.
The second is front-line servicing and support. AI handling the routine share of inbound volume, FAQs, status checks, and basic account questions frees the human team to focus on one-call resolution for the harder cases. It also extends effective service hours well beyond the traditional nine-to-five. That's the difference between a customer waiting until morning and getting an answer after midnight, when they're actually thinking about it.
CX Network: What challenges does fintech still need to overcome?
Mike Petrakis: The learning gap is the biggest one. Most organizations don't actually know how to design workflows that capture AI's benefits; they deploy the technology and hope. McKinsey found only one percent of companies consider themselves AI-mature, and leadership alignment is the largest barrier to scale.
Regulatory tension is close behind. Lending can't move fast and break things. Every automation decision has to be built with compliance as a first-class requirement, not an afterthought, and that slows adoption in ways consumer tech companies don't have to deal with.
And the human factor is constantly ignored. Companies roll out AI as a purely technical project and never address the fear of job displacement, which undermines adoption on their own teams. The organizations that get ahead of that conversation are seeing real productivity gains. The ones that don't are in that 95 percent failure bucket.