Skip to content
Moberg

Beyond fintech: what we learned while building across complex industries

The bar was set by money. Then it turned out to be the bar everywhere.

Written by Ihor Protsiv October 2023

Every industry we work in now — aquaculture, retail, telco and media, hospitality — we arrived at from payments. That order matters more than it sounds. Fintech is an unusually unforgiving place to learn to build software, and almost everything that made it hard turned out to be worth having somewhere else.

The hardest school

Four things are true of payment software and of almost nothing else. Money must reconcile: a rounding error in an ordinary product is a ticket, and in a payment platform it is a discrepancy in somebody's payout. There is no minimum viable settlement — you cannot ship a third of it and iterate. Somebody will audit it, which means a figure computed two years ago has to be reproducible from the exact code that produced it. And it cannot be down, because people are buying lunch with it.

We spent a decade inside that. Netgíró has been in production since 2012 and still is. Straumur went from an empty repository to live money in about seven months. The credit decision engine has carried five model generations without a rewrite, every one of them defended to a board. None of that bar was set by us. It was set by the domain, and we either cleared it or we did not get paid.

What actually transferred

Aquaculture

Reconcile before you cut over

Arnarlax replaced its source systems and its data platform at the same time, which is the riskiest migration there is: every number changes both where it comes from and where it lives. What made it survivable is a payments habit. Run the old and the new in parallel and reconcile until they agree. Map every migrated figure back to the table it used to live in. Nobody is asked to trust a new number — they can see for themselves that it is the old one.

Telco & media

A platform a regulated business would recognise

A bank's data platform and a media group's want the same things: one definition of each number, lineage from source to report, and an access model somebody can audit rather than infer. Sýn's lakehouse is governed like a financial system because that is where we learned to build one — and it is why putting AI on top of it was a feature rather than a risk assessment.

Retail

Sitting behind systems you do not control

A payment facilitator lives behind an acquiring bank and a card scheme it cannot change. That teaches you to put one layer of your own between your product and everything it depends on — which is the same reason Hópkaup's storefront, checkout and reporting sit on one backend it owns, with the payment provider and the analytics behind an interface of its own rather than wired straight into the shop.

Hospitality

One system of record, however many front doors

Straumur runs merchant, partner and admin portals, public APIs, a hosted checkout and a shop plugin on one core, because each additional surface is another way for the same figure to disagree with itself. Gaveia runs four booking platforms on one backend for exactly that reason — and when demand tripled in 2021 and doubled again in 2022, growth was a scaling exercise rather than a re-platforming.

And what does not transfer

Plenty. A balance is exact and a biomass is an estimate, so aquaculture reporting has to carry an uncertainty a ledger never does. In a shop the customer is standing up, one-handed, with a trolley — speed means something different there than it does at a desk. And in consumer products the compliance surface is accessibility rather than financial regulation, which is a different standard, differently tested.

Those are the parts you have to actually learn, which is why we start every engagement in the client's business rather than in ours. But they sit on top of something that does not change from one industry to the next.

One hard industry, learned properly

We never set out to learn six industries. We learned one difficult one properly and then found that most of what makes software trustworthy is not domain knowledge at all: reconcile before you switch anything off, keep one definition of every number, put a layer of your own between yourself and what you cannot control, and assume somebody will ask you to prove it later.

That is also why the list keeps growing. Healthcare came next, with Sidekick; real estate is current work with Reitir, and that story is not written yet. The industry is the part we learn from you. The standard we bring with us.

How we handle rapid changes in development caused by AI

Blog

How we handle rapid changes in development caused by AI

The pace of change in AI is genuinely fast. Here's how we keep our engineering grounded — and quietly improve faster than the noise around us.

Microsoft Fabric vs Databricks: a high-level overview

Blog

Microsoft Fabric vs Databricks: a high-level overview

Two strong answers to the same modernization question — different architecture philosophy, operating model and audience. A neutral comparison across the dimensions that actually decide it.

MTK: AI-driven development, held accountable

Blog

MTK: AI-driven development, held accountable

AI now writes a serious share of our code. MTK is how we make sure that never lowers the bar — spec first, tests second, AI-generated code third, every change reviewed and traceable.

Ready to write your story?

Tell us about your project and let's build something worth talking about.

Start a conversation →