Every founder building in AI right now faces the same question from a regulated buyer, usually within the first ten minutes of the conversation: are you going to let the machine make the decision?
I've had this conversation more times than I can count, and I understand exactly why it comes up. I spent 37 years in banking before I spent a single day building Finnulate, and for most of those years, the decisions that mattered, whether to extend a loan, whether an account carried risk, whether an institution was operating within its regulatory boundary, were made by people. People with judgment, with accountability, and crucially, with someone above them who could be held responsible if the judgment was wrong.
AI changes the speed of that decision-making. It should never change who is accountable for it.
Speed Without Ownership Is Not Progress
I've watched lending transform completely over my career. Twenty or thirty years ago, a loan decision meant documents, verification, and a wait. Today, platforms disburse in real time. That shift happened because AI can process information and assess risk far faster than any human or group of humans ever could.
But speed was never the actual constraint in lending, or in compliance, or in risk management. Accountability was. When a decision goes wrong, someone has to be able to explain why it was made, and someone has to own the consequence. A model can generate an output in milliseconds. It cannot stand in front of a board and explain itself, and it certainly cannot be held liable when a regulator asks hard questions.
This is the distinction I built into Finnulate from day one, and it's non-negotiable for me as a founder: AI suggests, humans decide. At every stage, the system tells you what needs to be done and why. It never gets the power to act on that suggestion unilaterally. The compliance owner, the checker, the human in the loop, retains full override authority. That isn't a limitation I've placed on the technology reluctantly. It's the design principle the entire platform is built around.
Why This Matters More in Regulated Industries Than Anywhere Else
I hear founders in other sectors talk about full automation as the end goal, and for plenty of use cases, that's the right ambition. Regulated finance is different, and I say this as someone who has sat as a group CEO answerable to a board, a regulator, and foreign institutional investors simultaneously.
When a regulator issues you a license, they're watching two things closely: whether you maintain adequate regulatory capital, and whether you operate within the boundary they've set for you. Both of those require a chain of accountability that a fully autonomous system simply cannot provide. If an AI model misjudges risk on a large commercial loan, or misreads a compliance obligation buried in a new circular, someone in the organization has to be able to say, here's what happened, here's why, and here's who signed off on it.
That's not a technology limitation. It's a governance requirement, and any founder building for this space who treats it as an inconvenience to be engineered around, rather than a design constraint to be embraced, is building the wrong product.
AI Removes Bias. It Doesn't Remove Responsibility.
There's a version of this argument that I actually find compelling, and I make it myself: AI, unlike a human decision-maker, doesn't carry the accumulated bias of forty years of gut instinct and pattern-matching. It works from the information it's trained on, and that can genuinely improve the quality and consistency of a decision, whether that decision concerns a loan, a KYC flag, or a compliance risk score.
But removing bias is not the same as removing the need for judgment. Intelligence, whether artificial or human, is fundamentally about giving perspective to a situation, and perspective still requires someone to weigh context that a model may not fully capture. A two-year-old today grows up alongside AI as naturally as I grew up alongside manual ledgers. That doesn't mean the two-year-old, or the institution, should ever hand over final authority to the system doing the calculating.
What I Tell Regulated Buyers Directly
When a bank or an NBFC asks me whether Finnulate lets AI take decisions, my answer is always the same, and it's the answer I'd want if I were still sitting on the other side of that table as a CEO. AI is a true co-pilot. It tells you what's required, why it's required, and how to get there. It never takes the wheel.
I think this is also where a lot of the "AI-enabled" compliance and GRC platforms in the market are getting it wrong. Many of the established players are patching generative AI capabilities onto platforms that were never built AI-native, and in doing so, some of them are quietly blurring the line between suggestion and authority. I built Finnulate the other way around, AI-native from the first line of code, with human override as a permanent, structural feature rather than a checkbox added later to satisfy a compliance team's objection.
That difference sounds small in a sales conversation. It is, in my experience, the single biggest reason a regulated enterprise decides to trust a platform enough to actually run it.
Dinesh Arora is the Founder of Finnulate AI and has spent nearly four decades in banking, including as Group CEO of a leading East African banking group and as a Board Advisor and Risk Committee Chair for banking subsidiaries in the region.