Why So Many Fintech AI Projects Are Failing (And How to Fix Them)

When tech founder and investor Draven McConville implemented GitHub Copilot at his work management platform Klipboard, the productivity gains were immediate and dramatic. “Tasks that previously took days were completed in hours,” he says. But it’s becoming clear that that’s not always the case.

Despite the hype around AI’s disruption, recent statistics tell a very different and sobering story. 95% of AI pilots across all industries fail to achieve rapid revenue acceleration, and fintechs face even steeper odds, 75% of which fail. What adding a layer AI of complexity onto an already challenging regulatory environment is yet to be proven.

McConville believes there are predictable patterns in these statistics and that they are avoidable if you understand why they happen.

The Technology-First Trap

“The biggest mistake I see fintech companies make is starting with AI capabilities rather than customer problems,” McConville explains. “This approach is dangerous in any industry, but it’s particularly damaging in financial services where regulatory requirements add layers of complexity that can’t be retrofitted.”

McConville says customers don’t want sophisticated AI algorithms, but they do need software that works reliably. “Financial services customers aren’t asking for machine learning models. They’re asking for faster loan approvals, better fraud detection, and more personalized investment advice.”

It’s a crucial difference in mindset for builders. “When you start with customer problems, you naturally build solutions that can be explained, audited, and regulated.”

Data Infrastructure Challenges

Data quality and readiness account for 43% of AI project failures, and this challenge is amplified in financial services. Banks and fintech companies often have data scattered across legacy systems, with inconsistent formats and varying quality standards.

“The temptation is to use AI to clean and organize this data, but that approach creates a circular problem,” says McConville. “You need clean data to train effective AI models, but you’re trying to use AI to clean the data. The result is often AI systems that perpetuate existing data problems while adding new layers of complexity.”

Garbage in, garbage out, as the old adage goes.

A Practical Framework

McConville offers a practical framework for implementing AI successfully in fintech:

• Start with Regulatory Requirements: “Before writing any code, understand the regulatory framework your AI system must operate within. Design your AI architecture to meet these requirements from day one.”
• Define Success in Business Terms: “Whether it’s reducing loan processing time or improving fraud detection while maintaining customer satisfaction, ensure every stakeholder understands what success looks like.”
• Invest in Data Infrastructure First: “Don’t implement AI if your data is dirty or your infrastructure is unreliable. Clean your data and implement robust standardization, governance frameworks, and audit trails.”
• Build Explainability from the Ground Up: “AI shouldn’t be seen as an add-on feature. Design your models to be interpretable from the beginning. This may mean accepting slightly lower technical performance in exchange for regulatory compliance and business sustainability.”