By Javier Borkenztain, Co-founder and CEO of FITER
Why the core banking decision is no longer the interesting one
A few years ago, every core banking conversation I had ended in the same place. Buy a proprietary license or build your own. Both roads led to the same wall: a contract with too many zeros, a vendor who owned your roadmap, and a system you could not change without asking permission first.
That conversation is over. Not because one side won the argument, but because the argument stopped being interesting. Serious banks and fintechs now start from open source. Apache Fineract is the proof. And once the core is settled, the question that actually decides who wins is a different one.
It is not what core you run. It is what you build on top of it.
The core is settled
My co-founder Robert made this case in detail a few months ago, using an airport as the metaphor: You own the building, and the airlines are just tenants. I will not repeat the whole argument here, but it comes down to three things.
You own the code. Not a license, not a seat, not a promise in a support contract. If a vendor disappears, pivots, or gets acquired by someone with different priorities, your core keeps running.
The cost curve is different. Running a Fineract-based core costs a fraction of what the proprietary alternatives cost over the same period. In our implementations, we have seen institutions save between half a million and two million dollars against incumbent systems. That is money that goes back into the product instead of into a renewal.
And the stack became modular. Nobody buys a monolith anymore. You assemble the best option per domain: a real-time ledger here, a payment switch there, and an intelligence layer above it all.
Ownership, cost, composability. Those are strategic advantages, not technical preferences.
But everybody gets them now
Here is the part most of these conversations miss. Open source lowered the barrier for you, and it lowered it for every one of your competitors at exactly the same time.
A modern core is a commodity. Commodities do not win markets. If your only differentiator is that you run open source, you do not have a differentiator. You have a starting line, and so does the fintech that launched last Tuesday.
The advantage moved up the stack, to the layer where records turn into decisions.
AI is where the game is won
A core banking system is a system of record. That is its job, and a good one does it without drama. But the value was never in the record. The value is in what you do with it, how fast, and at what cost per decision. That is the part AI changes, and it changes it in places every institution already knows are expensive.
AML and KYC. Static rulebooks generate alerts by the thousand and catch the patterns somebody thought of three years ago. Risk-based models score behavior in real time, learn from what turned out to be nothing, and let your compliance team spend its hours on the cases that deserve them.
Credit underwriting. Manual analysis is the difference between onboarding a borrower in four days and onboarding them in four minutes. Models trained on your own repayment history do not just move faster. They see the borrowers a scorecard was never built to see, which in most of our markets is most of the market.
Collections. This is still a spreadsheet exercise at institutions that would be embarrassed to admit it. It should be a prediction: who is about to fall behind, when to reach out, which channel they actually answer, and which accounts to leave alone because contact makes things worse.
Reconciliation. The eternal back office tax. Matching, flagging, posting, all of it done without a person comparing two screens at the end of the month.
Customer experience. Not a chatbot answering FAQs. Agents that can look up a balance, explain a charge, restructure a payment, and execute the transaction, inside the same conversation.
None of this requires replacing your core. All of it requires a core that is open enough to let data out and let instructions back in. That is the architectural point, and it is why the two halves of this argument are really one argument. AI does not sit inside a legacy core. It sits on top of an open one.
How we build it
We have spent years putting Apache Fineract into production across Europe, the Middle East, Latin America, and Africa, for everything from high-volume challengers to enterprise banks. What we learned there shaped the reference architecture we now start from.
Fineract is the product engine and the regulatory record. Loans, savings, products, accounting, the things a regulator will ask you about.
TigerBeetle handles the real-time ledger. Balances, debits, credits, with the atomicity a financial system cannot negotiate away.
Kafka decouples everything else. Events stream between the ledger, the core, and any service that needs to consume them or act on them, including every model you decide to run.
What comes out is a headless, event-driven core. APIs in, events out. That happens to be exactly what an intelligence layer needs to be useful: clean streaming data to learn from, and a clean streaming channel to act through. Build it the other way around, with the models bolted onto a system that was never designed to talk, and you will spend your budget on plumbing instead of on product.
This is not a whiteboard architecture. We are running it with clients processing hundreds of thousands of loan applications a day. The core holds. The product lives in the layer above it.
The strategy in one sentence
Run the most boring, most reliable, most open core you can find, and spend your creativity, your engineers, and your budget on everything above it.
Open source removed the toll booth. It did that for everyone, including your competition. The race starts after it, and the institutions that understand the difference are the ones that will still be here to explain it.
—
About the Author
Javier Borkenztain is Co-founder and CEO of FITER, a Member of the Apache Software Foundation, and a member of the Project Management Committee of Apache Fineract. He has spent decades working with financial institutions across five continents on how open technology changes what they can build.
FITER is the number one Apache Fineract implementer in the world, with over 100 implementations across more than 25 countries on five continents. Talk to us.
