Kaushik Subramanian: The Sectors Where Promising Fintech Startup Ideas May Be Hiding

In the first part of this series for fintech entrepreneurs, I argued that most promising startup ideas come from following the money through a business to find bottlenecks. Now, it’s time to talk about sectors.

Once you know what a good bottleneck looks like, it helps to know where in a business they tend to cluster. So, let me walk through a few of the places money flows and point out what I think are potential friction points today.

Opportunity 1: The office of the CFO

Let’s start with the function that touches every flow of money in the business: the CFO’s office.

Think about what actually sits inside the office of the CFO: accounts payable, accounts receivable, treasury, the financial close, financial planning and analysis. Each of these usually represents a separate piece of software that costs real money (and that rarely talk to each other). They all sit on top of an Enterprise Resource Planning (ERP) tool.

This structure reflects how the function of this office has evolved over time. Accounts payable, accounts receivable and treasury were split apart decades ago for convenience, and the tooling has cemented the split.

Now let's put each of these functions through the two tests for startup ideas I outlined in part one. (Spoiler: they pass comfortably.)

Test one: Is it possible? Yes. The workflows are repeatable to the point of tedium; the data is rich because this is where the company'’s money is recorded, and the context is deep because to do any of this work you have to know how the whole business fits together. According to one widely cited estimate compiled by McKinsey in 2018, 42 percent of finance activities can be fully automated and another 19 percent mostly so. The technical ceiling to unlock these bottlenecks is high, but what has been missing is software built to reach it.

Test two: Is it transformative? Yes. This is the cost center of the company’s finance operation, and it usually involves a stack of expensive tools and the people to run them. Collapse that and you have not only shaved off a major cost, you have potentially changed the shape and throughput of the function.

AI can demolish the walls that have historically stood between those silos. The reason accounts payable (AP) software also couldn’t do accounts receivable (AR) was never a law of nature. It was simply that each tool was built for one use case.

Once all the data sits in one unified layer, you can unleash agentic workflows on all that data. And at that point there is no good reason a single agent can’t run both AP and AR, inside one piece of software.

The legacy silos were a constraint of single-use software, but remove the constraint and the lines between these jobs start to blur. And there are a lot of fintech ideas sitting in that blur.

Opportunity 2: On top of the ERP

There is an obvious question lurking here: if the ERP is the thing everything sits on, why not replace the ERP itself with something built for this era?

There are people who are trying, and some new companies are growing quickly. Rillet raised over $100m in under a year – a $25m Series A from Sequoia followed by a $70m Series B co-led by Andreessen Horowitz and ICONIQ just 10 weeks later – and signed more than 200 customers along the way. Campfire also raised a comparable sum across two rounds last year.

Yet the market for a new ERP is perhaps narrower than the excitement suggests. When I was looking at AI-native ERPs, a CFO at a large company described replacing his ERP as a “heart transplant performed while running a marathon”. The odds of surviving it are not good, and the reward for attempting it is mostly the absence of a problem he is already living with. He won’t do it, and most CFOs in his position won’t either.

But there are companies that will switch: fast-growing enterprises where revenue is still relatively simple, and where the incumbent ERP is painful to use. These are mostly SaaS businesses, whose subscription revenue the older systems were never really built to handle.

I find it useful to think about this along a single axis: revenue complexity. Where complexity is low, an AI-first ERP can win, and that is the market the new entrants are taking. Where complexity is high, the legacy systems stay, because the cost and risk of moving are enormous and the benefits are unclear.

So my expectation is not that old ERPs die, but rather that they become a commodity that settles into being the system of record. The real value moves up to the workflow layer that sits on top of these ERPs and treats the record as a pool of data to act on. The most interesting fintech companies – and most lucrative fintech ideas – are mostly up there, not down in the foundations.

Opportunity 3: Capital markets

The obvious place to build in capital markets is the front office, where there is enormous willingness to pay, budgets are large and the work is complex. Workflows here pass the twin test easily.

One sign this space is attractive is that plenty of people in the front office have skipped buying anything and are using general-purpose AI tools, such as Claude and ChatGPT, straight out of the box to do their work. Good companies are being built here too: Rogo is doing well, in a market it shares with the likes of Hebbia, whose Matrix product set an early bar for what AI document analysis could do in finance.

But I think the front office is getting too much attention; I might be going against the grain here, but I think the space is now far too crowded. The opportunities are visible to every founder and every investor looking at the space, and they have mostly converged on the same one or two companies.

The opportunities I find more interesting are in the middle and back office, where there can be little automation and where far fewer people are looking right now. One example is fund operations.

Many investment firms don’t run their own fund operations. Instead, they pay a company to do it: fund administration, LP reporting, capital calls, and the other machinery of running a fund. Some of the firms that handle it are enormous.

Again, let’s put it through the two tests. Possible is the harder and more interesting question here: Fund operations is not a single clean workflow, and you are not going to automate all of it tomorrow, but the core of it is exactly the repetitive, rules-bound, data-heavy work that AI-driven technology is good at. And the data and context already live inside the fund-admin function.

A smart new company should be able to do most of it well, but not all of it – at least not at the get go. But then your claim is not that you could replace the whole function, it is that the bulk of it is automatable and the rest, for now, is not.

Fund operations is also quite obviously transformative. The function is large, it is already outsourced, and the buyer is a CFO who knows the number to the dollar – and would love it to be smaller.

The pitch writes itself: You walk into the CFO of a large manager and say, you are spending this much on fund operations, and I can take out a large piece of that – maybe as much as 80 percent – using AI. In an industry that is ruthless about its own margins, taking a number like that off the cost line is big.

So fund operations is a place with ideas that are both possible and transformative, but everyone is up at the front of the house where the demos are prettier. Meanwhile, the middle and back offices are exactly the kinds of places a great fintech startup idea tends to hide.

Opportunity 4: Compliance

Let’s talk about compliance and fraud. The cost of a false negative – letting through something you should have stopped – is so high that the willingness to pay is structurally large. Workflows such as anti-money laundering (AML) and know your customer (KYC), license monitoring and flagging breaches are all specific examples of money flow blockages that easily clear both our tests.

These are just a handful of examples of places to look for great fintech ideas – and they are all places where I fully expect exciting new companies to sprout in the months to come.

But there are many, many more.

In conclusion: to find great fintech ideas, don't start from the sector. Follow the money through a business until it slows, and at each bottleneck, ask whether you can solve it – and whether solving it would change the customer’s business enough to matter. Demand a yes on both questions. And when you have a choice, lean towards the unglamorous, expensive, outsourced jobs that companies are already paying someone else to do badly.

That is where durable businesses will get built, and these places are usually a long way from wherever the crowd is pointing right now.

This is the second part of a two-part series on fintech startup ideas. Read the first article here.