Architect
The Last-Mile Launch Checklist: 10 checks before you trust the pipeline.
Green dashboards tell you the parts ran. Only the last mile tells you the work arrived. Run this before calling any automated marketing system live. By Debbie Moynihan.
The stretch between "the system says it worked" and "a human saw the result in the wild" is where automated pipelines fail silently, because tools treat empty configuration as valid. Ten checks close that gap: trace one real item end to end, hunt every empty state, and give silence its own alarm. Run the list at launch, after every change, and whenever a channel goes quiet.
- 1. Pick one real itemNot a test record. A real post, a real email, a real lead. Test data takes paths real data does not.
- 2. Find it in every tool in the chainOpen each system in order and locate your item in it. The moment you cannot find it, you have found the gap.
- 3. Finish in the wildEnd on the public feed, the real inbox, the live report. The trace is done when you see the result where your audience sees it, and nowhere earlier.
- 4. List every empty-state settingSchedules, destination lists, mappings, filters. Anything that can be validly empty can silently hold your work forever.
- 5. Confirm the schedule has slotsIf a scheduler publishes from recurring time slots, open it and count them. Zero slots is a valid configuration and a stalled pipeline.
- 6. Check every account connectionTokens expire and integrations disconnect without ceremony. Confirm each destination account is currently authorized, not just that it was once.
- 7. Watch the queue drainA healthy pipeline empties. If items enter a stage faster than they leave it, that stage is your bottleneck or your stall.
- 8. Alert on absenceErrors already page someone. Add a check for "nothing has shipped in X days" on every channel that has an expected cadence. Silence needs its own alarm.
- 9. Re-run after every changeNew tool, new account, new plan, new person: any change re-opens the last mile. The trace is ten minutes; the quiet month it prevents is not.
- 10. Name the ownerThe last-mile check is someone's explicit job on launch day and after every configuration change, or it is nobody's.
check five is the one that caught us
Check five is on this list because it caught us. Our posts were written, approved, and queued, every upstream stage provably working, and the feeds stayed quiet. Following one single post through every hop ended at the scheduler inside our posting tool: zero recurring time slots configured. No error, no warning, no failed job. We built out the weekly slot cadence for every brand feed and the queue started draining the same day.
- 11. Rate limits and quotasA pipeline that works on day one can die on day twelve when an API quota runs out. Same silence, later calendar. Know each tool's limits and what happens when you hit them.
- 12. No double-postingRetries can double-publish. Verify each item arrived exactly once, not at least once; a duplicate post is a public bug. Your developer will call this idempotency, which is the word to use when you ask them to check it.
- 13. Timezone and posting windowSlots can exist and still be wrong. Confirm the schedule's timezone matches the audience's, or the 9am slot quietly becomes a 3am post.
- 14. Token scope, not just validityConnected is not the same as allowed. An account can authenticate fine and still lack the permission to post. Test the action, not the connection.
- 15. For email: placement, not deliveryDelivered means a server accepted it, nothing more. Keep seed inboxes on the major providers and check the spam folder, not just the delivery report.
Check eight needs a number. These are suggested starting points, not laws; tune them to the channel and how fast you can react.
| Expected cadence | Alert after |
|---|---|
| Daily feed | 2 missed days |
| Several posts per week | 4 quiet days |
| Weekly | 9 quiet days |
| Monthly | 35 quiet days |
starting points, not laws
List the hops. Write every system your content passes through, from queue to public feed:
List the empty states. For each hop, what setting could be validly empty and silently block it?
Name the owner and the date. Who runs this checklist, and when is the next scheduled re-run?
- Trusting the tool's own success messages as proof of delivery.
- Verifying with test data only, then never watching a real item travel the pipe.
- Checking once at launch and never after tools, accounts, or plans change.
- Assuming a quiet channel means an empty queue instead of a stalled one.
- Monitoring for errors only, when the failure mode you will actually get is absence.
Silence is a failure mode. Give it an alarm of its own.
Tools mostly fail loudly when something breaks and silently when something is missing. Broken things throw errors. Missing things, an unconfigured schedule, an empty destination list, a filter that matches nothing, just produce absence. Absence does not page anyone unless you build the pager yourself. That is check eight.
This is Field Notes No. 002. Every check comes from my own stack. Companion piece: No. 001, The Approval Pipeline, covers the system upstream of this checklist.
Questions people ask about this
What does verifying the last mile mean in a marketing system?
Following one real piece of work through every stage of your pipeline until you see the final result with your own eyes: the live post, the delivered email, the row in the report. Component checks tell you the parts run. Only an end to end trace tells you the system delivers.
Why do automated posting systems fail silently?
Because most tools treat an empty configuration as a valid state, not an error. A scheduler with no time slots, a feed with no destination, or a filter that matches nothing will all report healthy while producing zero output. Nothing crashes, so no alert fires.
How often should you re-verify a marketing automation?
At launch, after any tool or account change, and on a recurring schedule you can actually keep. And any time a channel that should publish on a cadence goes quiet, re-run the checklist before assuming the queue is empty.
Download the PDF version
The printable edition of No. 002, write-in worksheet included. Free during launch. The email edition will carry these going forward.
Get Field Notes by email, or subscribe by RSS. Printable PDFs ride along.
No. 001
The Approval Pipeline: the status flow that lets AI draft at scale while a human ships every post.
Open No. 001Join the community
The free community where these notes get discussed.