The first thing I deploy for a new integration client does nothing.
It runs in production. It has a health check. It logs. It's deployed the same way everything after it will be deployed. And it moves zero real records, because there's nothing real to move yet — we haven't finished the field mapping.
That's on purpose.
The usual way integration projects go: weeks of discovery and mapping, then a big deploy at the end where the new pipeline goes live for the first time, all at once, against production data, in front of everyone.
That's the worst possible time to discover that your deploy script has a typo, or that the client's firewall blocks outbound webhooks, or that nobody actually has the SFTP credentials the vendor promised. You've stacked every unknown into one high-stakes moment.
So I don't. I get something — anything, even a no-op — running in the real environment in week one, while the stakes are still zero. Deploy pipeline, credentials, network access, logging, alerting: all the boring infrastructure gets shaken out before there's a single real field mapped.
None of that is impressive. It's also exactly the stuff that blows up big-bang go-lives, and it's cheap to find out about when the blast radius is zero.
Once the do-nothing pipeline is live, week two is field mapping — the part that actually takes the time (more on that in Field Mapping Is the Real Work). Every week after that, a little more of the mapping ships through the same pipeline we already proved works. By the time real money is moving through it, the pipeline itself is old news. The only thing left to get right is the mapping — which is where the risk should live, because it's the only part that's actually hard.
Field Mappings