Running a marketplace while running an IT function

My day job has correct answers. The mailbox migrated or it did not. The profile connects or it does not. When something is wrong there is a state of the world that is right, and the work is getting there. Being right is the job.

A marketplace has almost none of that, and the adjustment was harder than the workload.

Decisions that cannot be verified

Should ratings go both ways, or only from the person who asked for help to the person who provided it? There is no test that settles this. There are arguments on both sides, and the only way to find out is to ship one, watch how people behave, and accept that you will not fully know even then because you did not run the other version.

I spent an embarrassing amount of time trying to reason my way to certainty on questions like that, because reasoning my way to certainty is what my job rewards.

Where infrastructure thinking transferred well

Two habits carried over cleanly.

Assuming the failure case. Infrastructure teaches you that the interesting path is the broken one. Applied to a product, that means designing what happens when a helper does not turn up, when a listing is stale, when someone leaves a bad rating out of a misunderstanding. Those paths get built early rather than deferred, and that has held up.

Density over reach. This one is straight from capacity planning. A marketplace serving twelve districts properly is worth more than one serving a country thinly, in the same way that a system sized for its actual load beats one sized for a launch announcement.

Where it transferred badly

Wanting to finish. Infrastructure projects complete. A migration has an end, and there is a real moment where the old thing is switched off and you are done.

A product does not do this. There is no state where the marketplace is finished, and treating each release as an ending produced a long stretch where I was disappointed to still have work after every launch. Reframing it as an ongoing operating position rather than a project with a finish line took about a year.

What it gave back

I run the IT function differently now. Fewer decisions escalated to certainty, more of them made deliberately as bets with a date to review. Infrastructure has more genuinely uncertain calls in it than the culture admits, and having spent time somewhere they cannot be hidden made me better at recognising them here.