by Serguey Shinder
We pulled a week of traffic in September for a capacity conversation, and the figure that stopped the meeting had nothing to do with capacity. Seventy six per cent of the requests reaching our systems came through the API, from software written by our customers or by the integrators they hire. Twenty four per cent came from a person looking at one of our screens.
The roadmap for this year, which I had written and defended, was almost entirely the screens. A redesign. A new dashboard. An onboarding flow we had spent two months on.
Then I went through where our feedback comes from, and it is worse than the split suggests. The satisfaction prompt appears inside the application. Session recording watches people use the interface. Support tickets are written by somebody who is looking at something and is confused by it. Every instrument we own is pointed at the twenty four per cent, and we had been iterating happily against a sample that excludes three quarters of our actual use.
The seventy six per cent does not complain, which I had mistaken for contentment. It breaks. Last spring we made a field required that had been optional, announced properly in release notes, and two integrators' overnight jobs began failing at three in the morning. One rewrote around it within a day and never mentioned it to anybody. We learned about it six weeks later, by accident, in a call about something else. There is no mechanism anywhere in our company by which that would have reached us, and I do not know how many others there have been.
None of this is going to reverse. The share of our traffic that is software has climbed every year I have measured it, and an increasing part of that software is now being written by people who will never read our documentation because something else read it for them. The person at a screen is becoming the minority user and remains the only one we can hear.
We have started with the obvious things. A changelog for the API with a subscription and an actual deprecation policy with dates on it. Error rates alerted per integration, with a named human contact for each one, because an integration is a relationship and we had been treating it as traffic. A quarterly call with our five largest integrators, which has already been more useful than the last two rounds of user research.
The distinction I carry out of it is that a product's real interface is whichever one most of its use passes through, and almost every company I know still instruments only the interface it can watch a person touch. The question for the next decade is not whether your product is usable. It is whether it is legible to something that is never going to look at it.
– Serguey Asael Shinder
Leave a Reply