DEEP DIVE
The Last 10% of an Automation Is 90% of the Work
The workflow is easy to demo. Production is everything around it.
There is a predictable moment in automation projects where the core flow works and everyone feels like the project is basically done.
Usually that is when the real work starts.
The exciting part happens first
You connect the systems. The trigger fires. The AI generates the right output. The CRM updates. The Slack message appears.
It works.
This is the moment that makes a great demo.
Then someone asks if it handles duplicate records.
Then permissions.
Then retries.
Then logging.
Then what happens if the API succeeds in one system and fails in another.
Then who owns the workflow.
THE DEMO
Trigger → logic → action
The core flow may take a day.
REALITY
Authentication, permissions, retries, observability, alerts, idempotency, fallbacks, exception states, documentation, training, ownership, testing, rollback, maintenance.
That is production.
Failure is part of the design
A workflow is not done when it works once.
It is done when you know what happens when it fails.
Does it retry? Can it create duplicates? Can the same event run twice? Can a human recover the process? Does someone get alerted? Is the customer left in an inconsistent state?
Permissions matter
What can the integration read? What can it change? Does it use one employee's personal credentials? What happens when that person leaves? Can the automation delete something it should only update?
These questions are boring until they become expensive.
Someone has to own it
Production systems require ownership.
Who receives alerts? Who approves changes? Who reviews failures? Who updates credentials? Who notices when a vendor changes an API?
If nobody owns it, the automation slowly becomes infrastructure nobody trusts.
Users need to understand the handoff
You can build a perfect workflow and still fail if employees do not know when it runs, what it changes, or what they are supposed to do when it escalates.
Training is part of implementation.
The workflow is not finished when the automation works. It is finished when the business can depend on it.
STEAL THIS
Production Readiness Checklist
01 · Authentication does not depend on one employee.
02 · Permissions are limited to what the workflow actually needs.
03 · Duplicate events cannot create duplicate outcomes.
04 · Failures retry safely or escalate cleanly.
05 · Critical actions and decisions are logged.
06 · Alerts go to a real owner.
07 · Manual recovery is documented.
08 · Users know what the system does and when to intervene.
09 · Someone owns maintenance after launch.
10 · The workflow has been tested with bad inputs, not just good ones.
This is why cheap automation gets expensive
The first 90% is visible. It is the boxes and arrows. It is what gets scoped in a quick sales call.
The last 10% is all the work required to make that workflow safe, understandable, recoverable, and maintainable.
That last 10% is usually where the business value lives.
BUILD SYSTEMS THAT SURVIVE REALITY
If this is the kind of operating problem you are dealing with inside your business, this is exactly the kind of work we do at SMB Ops.
Drew Reynolds
The SMB Operator