Case studies
Two projects, from the inside
They're here because they show how architecture decisions get made, which is exactly what you're hiring when you hire development.
Trackfinite
Trackfinite monitors the errors that happen in an online store and connects them with the purchases that never got completed. When a cart is abandoned halfway, it tells you what happened: the website failed, the payment failed, or the customer simply left.
What it saves, above all, is hours of work between departments. The argument over whether the problem is the website, the payment or the customer disappears, because there's a fact where there used to be opinions.
The architecture decision that defines it
Under the hood there are four independent processors: each one does a single job, and none needs the others to be doing well to do its own. They were designed to work in batches, but compatible with real time from the start: the day the data needs to be live, nothing will have to be rebuilt.
And the WooCommerce plugin that collects the data has a strict per-event overhead limit. The reason is common sense: measuring cannot slow down the store you're measuring.
Status · Our own product, in development, with public funding granted at the validation stage.
Property-listing valuation
A Chrome extension that switched on while browsing property portals and scored every listing by relating the home's characteristics, its area and its price. Two decisions made it different.
The weight of each characteristic wasn't fixed
A characteristic counted for more in the score if the best-valued homes in that area had it. In other words: the system learned what matters in each neighbourhood (a terrace doesn't weigh the same in the city centre as on the outskirts) instead of imposing one general rule.
The score was relative, not absolute
A listing wasn't compared against a theoretical "right" price, but against the offer that existed at that moment. That solves two problems at once: areas with little data (there's always a best and a worst among what's there) and spotting the real opportunity, because whoever sells below the rest to get out fast raises their own score and lowers everyone else's.
The algorithm was the flashy part, but around it there was a whole operation that also had to be built: an extended area report to complete the context, with a conversational assistant to ask questions; reports and emails going out daily; and the agencies the tool worked with, managed in a custom ERP built for the tool itself, not an off-the-shelf one.
Status · It never launched.
What they have in common
In both projects, the architecture decisions were made before writing code: which pieces exist, what each one does and what the system must withstand the day it grows. That's what avoids rebuilding it later, and it's the same thing we do in every client development.
If your project needs that kind of decision-making, tell us about it: we reply within one business day.