Map a human + AI workflow end to end
The opportunity brief says which flow deserves attention. A delta map shows exactly what changes between current and proposed operation. The goal is not a polished process picture but a working artifact whose design decisions can be inspected.
Applied output
A current-to-future delta map compares current states and waiting with proposed human-agent roles, decision evidence, handoff contracts, and exception paths.
Seven layers of a delta map
At every layer, mark the current state, proposed change, and assumption behind that change separately.
- Current state
- Show real steps, waiting, rework, and ownership gaps with evidence.
- State transition
- Name the event that changes work state and the acceptance condition for that move.
- Role lane
- Separate the concrete outputs produced by humans, agents, and deterministic systems.
- Decision and evidence
- Connect every flow-changing choice to its input, authority owner, and trace.
- Handoff contract
- State the context transferred, the new owner, and the acceptance condition.
- Exception path
- Draw a safe route for missing data, low confidence, timeout, and out-of-policy work.
- Delta marker
- Expose every removed, added, or re-authorised element and its assumption.
Workflow case
Fast classification, a lost maintenance request
Facilities classifies maintenance requests with AI in seconds. Work needing spare parts moves to procurement by email, while safety risks live in another list. Request state disappears across systems, and the building manager cannot see who accepted ownership.
What does the map reveal?
Classification is faster, but state transitions, context packages, and handoff acceptance remain undesigned. The future map must expose every request owner, transferred evidence, and the route for a safety exception.
Separate current and future flow
Current
Map what actually happens, not the ideal process in a policy document.
Change
Mark every removed, added, or re-authorised step explicitly.
Ownership
Name the human accountable for outcome and policy quality even when an agent acts.
Evidence
Show the data used and trace left by every consequential decision.
Common trap
“Mapping only the happy path.”
Operations are shaped by exceptions. A design is incomplete if missing data, low confidence, tool failure, out-of-policy requests, and timeouts have no explicit route.
Check before the PracticeLab
In your chosen flow, which state will AI read, which state may it change, and where will it hand over to a human with context intact?
Use one trigger, one decision point, one handoff, and one exception in your answer.
Keep this
A useful workflow map shows how state, decisions, context, and ownership move end to end—not just a sequence of tasks.