Map a workflow's handoffs before adding AI
A faster draft hasn't made the process faster. The work may be waiting in somebody else's inbox.
What you'll make
a handoff table that shows where work happens, where it waits and who decides.
What you'll need
ChatGPT in your work account, a short process description and someone who knows the handoffs. Allow 20 minutes. Describe the process without including private records.
Example result — illustrative map:
| Handoff | Work and waiting |
|---|---|
| Curator to editor | Draft prepared; waits for catalogue reference |
| Editor to curator | Wording checked; curator confirms meaning |
| Curator to print desk | Confirmed copy enters the next print batch |
1. Describe one completed journey
Select Chat › New chat in ChatGPT. Start with one completed item, rather than the process everybody wishes they followed.
Map this process in order. For each handoff give the work done, input required, waiting described and decision owner. Use only the description. Keep missing durations as unknown.
An exhibition label starts with a curator's note. An editor drafts the label, then waits for a catalogue reference from the curator. The curator checks the meaning and confirms the text. The print desk includes confirmed labels in its next batch. No times or batch dates were recorded.Why this works: a real sequence reveals dependencies that a generic workflow diagram can hide.
What you should see: work and waiting separated, with no invented durations.
2. Walk the map with the people doing the work
Check whether the required input is available at each handoff. Ask who makes the decision and what lets the next person start. Add any correction to the source description, then ask for the same table again.
Why this works: the map needs to match the process before it can support a change.
What you should see: a sequence that the people involved recognise, including the delays they can confirm.
3. Ask where an AI draft would change the journey
Using the checked map, describe one possible place for AI drafting. State the input it would need, what a person must check and which waits or decisions it would leave unchanged. Do not claim a time saving without measured durations.Why this works: the instruction ties the proposed assistance to a handoff, rather than treating faster writing as faster completion.
What you should see: a possible first label draft that still needs the catalogue reference, curator confirmation and print batch.
Check the result
- Does each row say what allows the next handoff?
- Are unknown durations left unknown?
- Does the proposal preserve the meaning check and print dependency?
The usual failure: the answer removes waiting because AI can draft immediately.
Restore the wait for the catalogue reference and the next print batch. AI drafting does not remove either dependency. Keep the curator's confirmation, and remove any predicted time saving because the source contains no durations.Worked example
Fictional example.
Exhibition label: draft from curator note; catalogue reference required.
Editor checks the source; curator confirms meaning before print.
Printing waits for its batch. End-to-end timings have not been measured.The exhibition label process requires a catalogue reference and a curator’s confirmation before printing. A weak proposal says: “AI generates the label instantly, so it can go straight to print.”
The repaired proposal says: “AI could prepare a first draft from the curator’s note. The editor still needs the catalogue reference; the curator confirms the meaning; printing waits for its batch.”
The drafting opportunity is plausible. Its effect on total completion time needs observation of the whole journey.
Go deeper
Measure elapsed time separately
For a small trial, record when an item starts, changes hands and finishes. Keep hands-on work time separate from elapsed time. A shorter draft may help the editor while making little difference to the print date.