Many financial institutions want to use more advanced AI, but remain constrained by data sensitivity, latency, cloud dependency and regulatory expectations. For banks, insurers and trading firms, the question is not simply whether AI can make better decisions, but whether those decisions can be made quickly, locally and with a clear audit trail.
Viroshan Naicker, CEO and co-founder of Refiant, believes AI efficiency is becoming a commercial and compliance issue for financial services. The company uses nature-inspired optimisation methods to reduce the computing power and energy needed to run AI systems, while allowing models to operate closer to where sensitive data and real-time decisions sit.
In this week’s Five Minutes With…, Fintechly speaks to Naicker about the hidden cost of cloud-based AI, why latency matters in financial services, the misconceptions around AI efficiency and why auditability is likely to become a bigger regulatory priority.
Can you tell us about yourself and what brought you to this point in your career?
I’m CEO and co-founder of Refiant, an AI optimisation company. My background is in mathematics. I completed my doctorate at the University of Johannesburg in 2015, focusing on networks. I spent several years as a lecturer at Rhodes University, before working on insurance and economic design in the Web3 space.
When Sid, Mathew and I founded Refiant in 2025, we wanted to solve a fundamental problem in AI: modern computation has a large energy footprint and bringing this footprint down is paramount to realising the upside of AI while reducing the projected costs to broader society. This includes the privacy cost of data handed over, which is a high price to pay in the most sensitive industrial sectors; finance, insurance and sectors involving research and development, for example.
For financial services, this matters enormously. Banks and trading firms need their AI to run on their own infrastructure where their sensitive data lives, not in someone else’s data centre halfway around the world. They need instant responses, not delays. And they need to be able to show regulators exactly how their AI reached a decision. Our nature-inspired algorithms – methods that mimic how natural systems solve complex problems – can reduce the computing power and energy needed to run AI while keeping everything local, fast and auditable.
What problem or opportunity are you most focused on right now?
Right now, financial institutions face an impossible choice: either send sensitive data to big tech cloud providers and accept delays, costs and loss of control – or don’t use advanced AI at all. We’re focused on eliminating that trade-off.
In finance, milliseconds matter. If you’re running a trading algorithm or detecting fraud in real-time, you can’t afford to send data to Amazon’s or Microsoft’s data centre, wait for their AI to process it, and hope the internet connection holds up. Many financial institutions need AI running locally on their own servers, with complete visibility into how decisions are made – for regulatory compliance as much as performance.
We’ve demonstrated that you can run powerful AI systems with a fraction of the computing and power use. That means a fraud detection system or trading algorithm can run on your own infrastructure, respond in milliseconds, cost a fraction of what cloud AI costs, and produce a complete audit trail showing exactly how it reached each decision. For a bank or trading firm, that changes what’s possible.
What do you think deserves more attention than it is getting in your part of the industry?
Latency – the time it takes for AI to give you an answer. Everyone talks about how accurate AI models are or how much data they can process, but in finance, speed is often the difference between a system that works and one that’s useless.
Take fraud detection: if your AI takes half a second to decide whether a transaction is legitimate, you’ve already lost – the transaction has gone through. Or trading: if your algorithm has to wait for a response from a cloud provider in another country, you’re competing with traders whose systems respond instantly because they run locally.
Yet the AI industry is obsessed with building bigger models that require more cloud computing power, which means slower latency.
Many real-world finance applications – high-frequency trading, real-time fraud detection, instant credit decisions – are limited by speed, not by AI capability. Making AI run faster and closer to where decisions happen is vastly more important for these use cases than making models slightly more accurate.
What do you think people still misunderstand about your part of the industry?
That AI efficiency is purely an environmental story. Yes, reducing energy consumption has environmental benefits, but the business case is far more compelling.
If you can run the same AI workload with complete control over your data, and with full visibility into how every decision was made, that’s transformative for a business.
In financial services, where speed is money and regulators demand explanations for automated decisions, efficiency isn’t about being green; it’s about being competitive, compliant and cost-effective. The environmental benefit is a nice side effect, but it’s not why CFOs and CTOs care.
What do you expect to rise up the agenda over the next year?
AI auditability, especially for AI agents making decisions autonomously. As AI systems start approving loans, executing trades, or underwriting insurance policies without human oversight, regulators and compliance teams are going to demand more than a simple explanation of how a decision was reached.
They’ll need a complete reasoning trace: which customer data points were accessed, which patterns the model recognised, how it weighed different factors, how it arrived at its conclusion. It’s the difference between an AI system saying “I rejected this loan application because of credit risk” and showing you a visual map of exactly which data points it considered, which historical patterns it matched against, and how it calculated the risk score.
As regulators wake up to the fact that AI is making consequential decisions, the question won’t be “Can your AI do this?” It’ll be “Can you prove exactly how your AI did this?”
Bonus: What’s the one question about your company we should have asked, and what’s your answer?
“What happens when AI models need to make decisions in environments where internet connectivity isn’t guaranteed?”
Most discussion around AI assumes you’re always connected to the cloud, but real-world financial infrastructure – think payment systems in remote regions, trading floors with backup systems, or emerging markets with unreliable connectivity – can’t depend on that. If your fraud detection system loses connection to Microsoft’s or Google’s data centre, it simply stops working.
Our optimisation approach means you can run powerful AI models locally, in your own infrastructure or on device, which is great from a privacy and sovereignty perspective.
