Ten years
From four providers and a gateway that mostly worked, to here. The things that turned out to matter were not the things we expected.
We registered the company in early 2015 with four signed providers and a gateway that mostly worked. Ten years is an arbitrary milestone but a good excuse to look at what actually mattered.
What we thought would matter
Breadth of catalogue. We assumed the platform with the most APIs would win and spent 2016 chasing listing count. It was the wrong target — customers do not want four hundred APIs, they want the six they need to be good and to work the same way. We now decline more providers than we accept.
Enterprise features. We built SSO, audit logs and configurable retention in 2017 for customers who did not exist yet. They exist now, so the work was not wasted, but it was four years early and it came out of the budget for things our actual customers were asking for.
What actually mattered
Consistency of the envelope. One auth header, one error shape, one pagination scheme, one rate limit contract, across everything. It is unglamorous and it is the single most cited reason people give for staying.
Free tiers that do not expire. The median gap between someone's first free call and their first invoice is still around four months. No trial period would have survived that.
Publishing real numbers. Latency percentiles, uptime including bad months, our own support response time. Every one of these was argued against internally and every one has been a net commercial positive.
Answering support tickets like they are part of the product. We measure providers on this and hold ourselves to the same standard, and it has done more for retention than any feature.
What we got wrong and fixed slowly
The normalisation layer, the five-tier pricing page, a year in a single region worrying about the wrong failure, and seven years of insisting that per-request was the only honest billing unit — until multi-storefront customers patiently explained that for them it was not.
That last one is the pattern worth naming. Most of our mistakes were not bad decisions. They were good decisions we kept applying after the conditions that justified them had changed.
The next ten
The same, mostly. More providers, refused more often. Numbers published whether or not they flatter us. And a gateway that stays boring, because everything anyone builds on top of us depends on it being the least interesting part of their stack.
Thank you to everyone who has built something on this. Some of you have been here since the first ten providers.