Vamsee Jasti

Most of the AI-in-finance conversation is about the obvious stuff — replacing traders, automating customer service, chatbots that help you check your balance. Fine. But that’s surface. The thing that actually matters is more structural: the cost of making a financial decision is collapsing, and that changes what’s possible to build.

The core of most financial products is a decision. Should we lend to this person. How do we price this risk. What portfolio allocation makes sense given these constraints. For decades those decisions were made by humans using heuristics, or by rules-based systems that encoded those heuristics into code. The models existed, but they were slow to build, expensive to maintain, and brittle when the data shifted underneath them.

AI changes the unit economics of that. You can build an underwriting model that ingests non-traditional data — cash flow patterns, transaction velocity, even how someone interacts with an application — and get a risk score that outperforms the old FICO-based approach. Not because the math is magic. Because the cost of building and iterating on that model has collapsed. What used to take a team of quants six months now takes a product engineer a few weeks.

When the cost of making a financial decision drops, you can make more of them. You can underwrite smaller loans that weren’t worth the overhead before. You can offer personalized insurance pricing instead of bucketing people into crude demographic tiers. You can rebalance a portfolio continuously instead of quarterly. The product surface area expands because the underlying computation got cheaper.

I keep watching this play out across fintech and the teams that are winning aren’t the ones with the fanciest models. They figured out the right product wrapper around a good-enough model. Same pattern I see everywhere in AI product work: the model is commodity infrastructure, the product decision is the moat.

The obvious risk: cheaper decisions become careless decisions. Automate underwriting at scale and you can automate bias at scale. The speed that makes AI useful in finance is the same speed that makes it dangerous. Most compliance frameworks still assume a human is in the loop somewhere. They’re not wrong to assume that. They just haven’t caught up to a world where the loop is closing.

Finance is moving from a world where the bottleneck was information processing to one where the bottleneck is product imagination. The models can handle the math. Whether we can design financial products that are genuinely better for the people using them — not just cheaper to operate — that’s a product problem, not a model problem. And it’s the most interesting one in fintech right now.