We could have rented all of this. There are content systems for publishers, email services with newsletter templates, analytics dashboards, media libraries, and a company of one is exactly who they are sold to. We did not, and this last piece is the reason, stated as plainly as I can.
A publication is becoming a source
For a generation the local news website was a destination. You went to it. The business was built on the visit: the ad on the page, the subscription at the door. That model is ending, and not slowly. Search sends less every year. Social sends almost nothing. And the readers who matter next are not going to visit at all. They are going to ask an assistant what happened at the county commission, and the assistant is going to answer from whatever sources it has learned to trust.
In that world a publication is not a place. It is a source. And a source is judged on things a destination never had to think about: whether every subject has one permanent address, whether every claim can be traced, whether the structure of the thing can be read by a machine, whether the archive is complete and the dates are right and the person named in 2026 is the same record in 2030.
None of that is a feature you can buy from a vendor whose product was designed for the visit. It is the shape of the data. You either own the shape or you do not.
What owning it made possible
Every piece in this series is an instance of the same thing. One database, so a person is one record across sixteen publications. One codebase, so a new publication is a row and the sixteenth costs less than the sixth. Images, email and analytics on our own terms, with nothing in them that reports to an advertiser. A desk shaped like reporting. A ledger behind every article. A graph of every subject we have followed. Rules that fire whether or not anyone remembers. Checks that run every night. Backups we have restored from. Half a second found because the whole path was ours to look at.
Rented, each of those would have been a ticket, a plugin, a workaround, or a no.
What it cost
Attention, mostly, and I have tried to be honest about the days it cost: the outage our own analytics caused, the ten nights a backup reported green while it skipped a bucket, the afternoon we optimized the wrong thing. Owning the stack means owning the failures. I would rather own them than rent them from someone who will not tell me they happened.
It did not cost what people assume. The platform runs on open, standard components that we could pick up and move, and the bill for it does not grow with the number of publications the way rent does.
The position
Mercury Local is infrastructure. Not a publication; the thing publications stand on. Charlotte and Farmington are where it was built and proven. They are not the ceiling. The model is a newsroom's stack that a small town can stand up in a day, that gets better for every publication when it gets better for one, that never asks a reader for money and never sells a reader to anyone, and that will still answer correctly when the reader is a machine.
We built our own because that is not something you can rent. That's not optimism. That's arithmetic.
This is No. 16, the last in We Built Our Own, a Mercury Local series on building our own publishing stack. Previous: What AI actually did.