A fintech founder on what trust actually means in money software
An interview with someone who's built three fintech products on what really moves the needle — and what's mostly theatre.
Editor's note: this piece is a composite based on common themes from conversations and public writing in the consumer-fintech space. The speaker, Priya Subramanian, is illustrative rather than a real individual; the views attributed to her reflect patterns we have seen across founders, engineers and operators in the category.
We sat down with Priya Subramanian, who has spent fifteen years building consumer fintech products — first as an early engineer at a major neobank, then as the founder of a payroll-savings startup, and most recently as the CTO of a tax-software company. She has, in her words, "watched trust get built and watched trust get incinerated, sometimes in the same quarter."
This interview has been edited for length and clarity.
MoneyPatrol: When people in fintech say "trust," they usually mean security. You've said that's only a small part of what actually matters. What do you mean?
Priya Subramanian: Security is table stakes. If you don't have basic security, you don't get to play. But security is a necessary condition, not a sufficient one. I've worked at companies with absolutely impeccable security postures that users still didn't trust, because the behaviour of the product undermined the security message.
The example I use a lot is dark patterns. You can have SOC 2, you can have penetration tests, you can have all the right encryption — and if your cancel button is buried under five screens of "are you sure?" prompts, the user knows, in their gut, that this company doesn't have their interests at heart. That's a trust failure that no security audit will catch.
MP: What's the trust failure you see most often in this category?
PS: Surprise charges. Either literal — a free trial that converted to paid without a clear notice — or figurative, like a feature that was free getting moved behind a paywall with no warning. Money software has a uniquely low tolerance for surprises, and a uniquely high tolerance for dullness. Users would much rather have a boring product they can predict than an exciting product that occasionally does things they didn't expect.
The companies that win in this category over time are the ones that internalise: "we are the boring trustworthy thing in a part of our user's life that already has too much excitement."
MP: You've said you actively distrust a lot of the AI features being added to fintech right now. Walk us through that.
PS: It's not the AI per se. It's the way it's being deployed. Most of these features are built by teams who think of AI as a marketing surface — "we have AI now" — rather than as a system that has to operate under the same constraints as the rest of the product.
The constraint that gets violated most is uncertainty calibration. A traditional finance product, when it doesn't know something, shows you a blank or an error. A poorly built AI feature, when it doesn't know something, generates a confident-sounding answer and serves it to you with the same visual weight as the real numbers. That's catastrophic in a context where the user is trying to make decisions about their actual money.
The fix isn't to not have AI. The fix is to have AI that says "I'm not sure" out loud, that distinguishes between things it's calculating from real data and things it's estimating, and that never, ever generates a number it can't trace back to a transaction. If a fintech AI feature can't do those three things, it's a liability dressed up as a feature.
MP: What does well-built fintech AI look like to you?
PS: Three properties. One: it's grounded in your actual data, so its answers are auditable. If it tells you you spent $X on groceries this month, you can click through to see the underlying transactions. Two: it's calibrated about uncertainty, so when it's projecting forward — "you'll probably hit your goal by August" — it tells you what assumptions it's making. Three: it has memory of context you've given it, so the conversation doesn't reset every time.
What I'd add is that the interface matters as much as the model. A great AI feature is one that surfaces information you'd have wanted anyway, faster — not one that performs intelligence at you. The best version of this is invisible. The user doesn't think "the AI did a good job." They just think "this app is unusually clear."
MP: You mentioned the cancel-button thing earlier. What other product behaviours signal trustworthiness or its absence?
PS: A few I look at, in order of how predictive they've been in my experience:
The cancel flow. How many clicks, how many "are you sure" prompts, do they make you call someone? A trustworthy product has cancel-button parity with the signup flow.
The export flow. Can you get all your data out, in a usable format, in a single click? The companies that make this hard are telling you something about how they think of you.
The pricing page. Is the actual price visible without signing up? Are upcharges clearly disclosed before you commit? Is there a free tier with a hidden upsell, or a clean tiered structure?
The error states. When something goes wrong — a bank connection breaks, a sync fails — does the product tell you plainly what happened and what to do, or does it serve you a generic "something went wrong" and quietly hope you forget?
These are not exotic standards. They're just basic product hygiene. The companies that get them right are signalling that they treat users as adults. The companies that get them wrong are signalling the opposite.
MP: What about the regulatory side? Does compliance equal trust?
PS: Compliance is necessary, not sufficient — same as security. The companies in our category that have been most embarrassed in the last few years were almost all in compliance with the letter of the relevant regulations. They just behaved badly within those bounds.
The thing I tell every founder I advise: regulation tells you the floor. It doesn't tell you the ceiling. A company that's only doing what it's legally required to do is a company that hasn't yet decided what it actually believes.
MP: Last question. If you were starting a personal-finance product from scratch today, what's the one principle you'd put above everything else?
PS: Make it easy to leave.
Sounds counterintuitive. It isn't. The companies that grow the longest in this category are the ones whose users stay because they want to, not because they're stuck. Easy export, easy cancel, transparent pricing, no dark patterns at the exit. Every dollar you spend making it harder to leave is a dollar you're spending on short-term retention metrics that quietly eat your long-term reputation.
The version of this I find myself saying out loud is: the door out should be the same size as the door in. Get that right and the rest of "trust" is mostly downstream of it.
This interview is part of an ongoing MoneyPatrol series on what trustworthy financial software actually looks like. We talked to Priya specifically because she has worked at all three sizes of company — early-stage startup, mid-stage scale-up, and large incumbent — and her view of trust has been forged in seeing what works and what doesn't across all three. The ideas here are hers, not ours, but they line up uncomfortably well with the principles we've tried to bake into MoneyPatrol from day one — particularly the part about the door out.
MoneyPatrol is not a financial, tax, investment, legal or accounting advisor. This article is for general educational purposes only and is not a substitute for personalised advice from a qualified professional. See our full disclaimer.
