We see this pattern constantly: a team automates 90% of a workflow beautifully, then leaves one manual step in the middle because it seemed too small to bother with. That one step usually erases most of the value of everything around it.
Why the smallest step does the most damage
A fully automated pipeline runs continuously, at any hour, with no queue. The moment you insert one manual checkpoint, the whole pipeline now runs at human speed and human availability — nights, weekends, and vacations included. The automation upstream and downstream just produces a longer queue in front of that one person.
The tell-tale signs
- Work finishes fast until 5pm, then sits until the next morning.
- One person becomes an unofficial bottleneck for a process nobody thinks of as "theirs" to own.
- The manual step is framed as "just a quick approval" — but the queue behind it says otherwise.
A pipeline is only as fast as its slowest manual checkpoint, no matter how fast everything else in it runs.
How to fix it without losing the judgment call
Most of these steps exist because someone wanted a human sanity check, not because a human is genuinely required every time. The fix is rarely "remove the check" — it's narrowing when the check fires. Let the automation handle the routine 90% of cases automatically, and route only the genuinely ambiguous cases to a human, with full context already attached so the review takes seconds, not minutes.
Before you call a workflow automated, trace it end to end and time-stamp every handoff. If there's a gap where nothing moves until a specific person opens their inbox, that's not a finished automation — that's a manual step wearing a badge.