Intelygenz advises Forbes on how to make AI finance projects work in the real world

Financial institutions are no strangers to AI investment. They launch innovation programs, run dozens of pilots, and build promising proofs of concept. Yet time and again, many of these initiatives never see the light of production, or fail to generate meaningful ROI once they do.

The problem isn’t enthusiasm or even technical talent. It’s the gap between AI exploration and AI execution. Getting AI into production and delivering value at scale is where most organizations stumble. The skills, experience, and muscle memory required to cross that chasm are rare.

For many, AI is still new. For us, it’s not. We’ve been developing deep technology solutions for over 20 years, and working with AI in real-world environments for more than a decade, well before it was fashionable. We’ve partnered with major banks and financial institutions to deploy AI in highly regulated, high-stakes environments. Along the way, we’ve seen what works, and what derails even the most well-funded programs.

In a recent article with Forbes, Chris Brown, President of VASS Intelygenz USA, laid out our lessons from the trenches, offering an approach to raise the success rate of AI deployments across industries. 

Here, we’re sharing some of the key insights to help turn ambitious AI goals into operational wins.

1. Choose the right business objective

The best AI projects start not with models or technology experiments, but with a tightly defined business objective. When you pick a small, winnable challenge to start, you see prompt returns, collect valuable data, and gain the momentum to expand further.

Instead, focus your first AI project on a narrow, well-scoped use case that: 

Fraud detection, for example, is an excellent candidate because it’s data-rich, high-impact, and outcomes are easy to measure in real business terms (fraud dollars saved, false positives reduced, etc.).

2. Build a cross-functional delivery team

AI in banking is not a science experiment. Real-world AI delivery requires tight collaboration between platform engineers, product owners and frontline operators. When these roles are brought in from day one, they can work through the trade-offs early.

3. Build compliance and governance in early

AI in financial services is subject to intense scrutiny, from internal risk teams, external auditors, and regulators alike. Building in compliance as an afterthought creates costly delays and mistrust.

Instead, design your AI systems with:

This is especially important in fraud, where false positives affect customer experience and false negatives expose the bank to loss.

4. Think beyond the model

Getting a model working in lab conditions is one thing. Keeping it stable in production, across thousands of decisions per second, is another challenge entirely.

Real-world AI delivery means:

Good MLOps practices and rigorous shadow-testing, is the unglamorous, but essential, side of AI success.

Financial institutions can’t afford to treat AI as a lab experiment. They need outcomes. The technology is there. The data is often there. What’s missing is a disciplined, production-minded approach to implementation—and that’s where experience matters. 

Our advice: Start narrow. Build cross-functional. Align to real business impact. And bring governance in from the start. That’s how banks cross the AI “chasm of doom” and come out stronger, faster, and more secure on the other side.