Kognitos Founder & CEO Binny Gill previously served as CTO at Nutanix and holds nearly 100 patents in computer science.
Before calculators existed, the word “computer” described a person. In fact, as early as the 1640s, according to the Online Etymology Dictionary, the word was used to describe someone who calculates.
As industry modernized, large organizations employed rooms full of human computers whose entire job was calculating numbers. They were skilled and accurate, but they remained human. They grew tired, lost concentration and made mistakes.
The original version of what we call computers today solved these inefficiencies by delivering the right answer every time.
That understanding matters more today than most AI discussions recognize. Traditional computing and modern AI were created to solve fundamentally different problems. Computers were deterministic machines built to eliminate human inconsistency. AI is probabilistic, and it was developed to emulate human reasoning and behavior. Both are powerful, but they are not interchangeable.
Based on my experience building an AI platform for finance leaders, the mistake many organizations make is assuming that technology designed to mimic human thinking can automatically replace systems that were designed to eliminate human error.
When electronic computers replaced human computers, they earned trust not because they were intelligent but because they were consistent. The same input produced the same output every time. Finance accepted the computer’s arithmetic because it consistently produced the same result. There was no bias to suspect, fatigue to account for or relationship to question.
That is the standard against which AI in finance must be measured, and one that most current AI systems cannot reliably meet.
The Problem With Smarter
The assumption is that as AI models become more capable, they become safer to trust. This is fine in theory, but it is also likely wrong, at least in the context of financial operations.
Consider a compound interest calculation where an AI model encounters the number 7.2 in the inputs and, drawing on its knowledge, associates it with the Rule of 72, a mental shortcut for estimating how long it takes an investment to double. The model may take that shortcut and produce an approximation rather than an accurate computation, without disclosing it. A less capable model, presented with the same inputs, would have simply performed the calculation with brute force.
I am not arguing against AI’s ability, but rather looking at what finance requires.
A brilliant, creative thinker in your accounts payable department does not make that department safer. A diligent, rule-following professional who does exactly what the process requires, escalates when something falls outside the process and does not improvise, does.
If greater intelligence automatically created safer outcomes, finance departments would hire people like Einstein, but finance requires precision and accountability, not creativity.
A Two-Layer Architecture For Financial AI
The answer is not to choose between AI and determinism. But we also can’t treat them as the same thing. Instead, financial operations need two distinct kinds of AI working together.
The first is the creative layer that reads processes, understands context and produces the rules that govern how work should be done. It writes those rules in plain English for human review and approval before execution. Its job is to think, and to make that thinking understandable.
The second is the execution layer, which is a deterministic system that follows approved rules precisely, does not improvise and raises its hand when it encounters something outside its authorized scope.
When a question escalates, the creative layer advises, and the human approves. Once approved, that decision becomes part of the future operating logic. The execution layer remembers it. The human is not interrupted again for the same situation.
This is how humans have always managed intelligence at scale. We write rules down, like tax codes and accounting standards, and create structures that enforce them. Humans do not rely on intelligence alone. They rely on structures that enforce behavior.
The challenge with AI is that it has no inherent concept of accountability or consequences, which makes governance even more important.
Finance has always understood this. General ledgers are immutable. Audit trails exist for a reason. SOX controls, external auditors and SEC oversight are not there because the finance team is untrustworthy, but because accountability requires structure, and structure requires documentation.
AI in finance cannot be exempt from these regulations.
What This Looks Like In Practice
Invoice processing is a useful example because every business does it. When an invoice arrives, it needs to be matched against what was ordered and what was received. Matching invoices against purchase orders and delivery records is one of the most common controls in finance.
Most companies process high volumes of invoices that do not match because names, part numbers and currencies vary across systems. Historically, this required multiple people to manually review and resolve exceptions. AI can now read across those documents and identify whether a match is reasonable, in ways that traditional rule-based systems could not.
But the question whether to pay is not a call you want AI making autonomously.
Consider the case of a duplicate invoice: the same invoice submitted twice in the same period. That could be an honest administrative error by the vendor, or a fraud attempt. How you handle it has implications either way. Accuse a legitimate vendor of attempted fraud, and you damage a business relationship. This is not purely a computational decision.
The right deployment is one where the rules for what constitutes a match, what tolerance is acceptable, what triggers escalation and what requires human sign-off are prescribed explicitly and approved in advance.
In this model, the deterministic layer executes those rules without deviation. When something falls outside them, it escalates. The creative AI helps refine the rules over time. And humans have meaningful oversight over the judgment calls that carry real consequences.
The Trust Problem
The original human computers were replaced not because machines were more intelligent, but because they were more reliable.
Today, AI will earn a place at the heart of financial operations when it can deliver not only capability, but accountability. Until then, trust will remain the most important calculation.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?