Why The Love Affair Between Fintech and AI Needs to Be Checked Out

In the fintech world, artificial intelligence (AI) has become the darling of innovation. Investments into AI soared, with 30% of all dollars invested in Q2 funneled into this domain, according to AI Business. Yet, as fintech companies rush to label their products as AI-driven, John Downie, CEO of SteadyPay, urges caution, emphasizing that AI should augment human intelligence, not replace it.

Reflecting on the evolution of fintech, Downie recalls the earlier days when linear regression was cutting-edge. Today, machine learning and large language models (LLMs) are the norm, but their application often lacks strategic nuance. At SteadyPay, AI is a critical tool in analyzing open banking data, but its limitations are clear. «When a customer’s financial future hangs in the balance,» Downie states, «they deserve better than a probability score from a neural network.»

The key to success, he argues, lies in leveraging AI for augmentation, not domination. Companies that recognize this distinction will lead the future of fintech.

Downie identifies three major issues with the way AI is currently implemented in lending:

1. Overengineering in Risk Assessment

Fintechs often deploy overly complex AI systems where simpler tools could suffice. Neural networks might seem impressive but aren’t always necessary. «We’ve tried fancy ways to predict affordability, but basic affordability corridors remain the most reliable,» Downie says. Transparency and explainability are critical, particularly when customers deserve clarity about decisions affecting their finances.

2. The Myth of Personalization

AI-driven personalization is often more about appearance than effectiveness. Many solutions, such as chatty chatbots or subscription spend optimizers, fail to provide meaningful value. Downie shares an example: «A Yahoo or Hotmail email is statistically a lower fraud risk than Gmail. It’s simple insights like this that work.» Customers want empathy and real solutions, not generic advice or flashy graphs.

3. The Speed vs. Accuracy Trap

In lending, faster isn’t always better. SteadyPay takes time to ensure affordability and fairness, sometimes intentionally delaying evaluations to gather more comprehensive data. While algorithmic decisions drive efficiency, human oversight remains critical for monitoring defaults, affordability, and vulnerable customers.

Downie also addresses common misconceptions about AI in financial services:

• Complexity Equals Accuracy: Sophisticated AI models don’t necessarily yield better results. SteadyPay’s experiments revealed that neural networks performed similarly to simpler models without trillions of transactions to refine them.
• AI Eliminates Bias: While AI can uncover biases, it doesn’t inherently remove them. For instance, statistical significance in postcodes has historically been used for discriminatory practices. Downie advocates for explainable AI to ensure fairness and transparency.

Despite its challenges, AI proves invaluable in specific areas like classification, offering cost and efficiency advantages over traditional models. However, Downie underscores the importance of grounding AI implementations in clear business cases. «AI isn’t special—it’s just another tool to help you get to an outcome,» he concludes.

As fintech continues to navigate the AI revolution, Downie advises a pragmatic approach. While embracing AI’s strengths, companies must also recognize its limitations. The ultimate goal should always be to enhance customer outcomes, not complicate or compromise them.

Downie’s candid reflections offer a timely reminder: success in fintech will belong to those who use AI to augment human intelligence, fostering trust and delivering tangible value.