Design the Handoffs

A handoff is any point where the work changes hands. Between two roles, between a system and a person, or between this workflow and a different one that can hold the same human. The third kind is the one that gets missed, and it is the one that produces the strangest bugs.

Between workflows

Two workflows that can both contain the same person need a written rule about what happens at the seam. Without one you get the situation everyone has seen: somebody who asked to be called back next week is still receiving the cold outreach they were in, because nothing ever took them out of it. Two systems are both correct according to their own rules and the person on the other end is getting a company that appears not to talk to itself.

The rule that solves this is worth stating as a principle, because it is not obvious: priority is enforced by moving someone, not by writing a number.

The instinct is to add a priority field. Warm is higher priority than cold, so set priority to high and let the systems sort it out. They will not sort it out, because the cold workflow does not read that field and has no reason to. A priority nothing enforces is a note.

What works is exclusivity through movement. The person leaves the cold workflow and enters the warm one. There is exactly one active workflow holding them, and the seam is specified: already in the cold sequence means quietly finished there, and started fresh here.

Publishing the latency

The last piece of a handoff specification is how long it takes, said out loud.

Most cross-boundary handoffs run on a clock rather than on the event. A reconciliation runs on a fixed cadence, a sync runs on an interval, a queue drains when someone opens it. So the change is not instant, and the gap between the human action and the visible result is real.

Write it into the specification and tell the people affected. The good documentation for this kind of system says something like: the switch is not instant, it lands on the next cycle, and if you need something to go out the same day, plan around it. An undocumented delay gets reported as a bug, and the support cost of that, repeated across a team, comfortably exceeds the cost of writing one sentence.