When we map processes, the handover between two people is almost always where time disappears. Not the work itself — the gap between someone finishing and someone else starting.
It is also, usually, the easiest place to begin, because improving it does not require changing how anybody does their actual job.
Why handovers cost so much
A handover involves several failure modes at once.
Work waits in a queue nobody is watching. The receiver has to reconstruct context the sender already had. Information gets summarised on the way, so detail is lost. If something is missing, a round trip is needed — and the sender has moved on. And nobody owns the item while it is in transit, so delays are invisible.
Six minutes of work either side of a two-day wait is a two-day process. Optimising the six minutes is the wrong target.
The pattern
The reusable shape looks like this.
Capture the trigger. Something arrives or completes — an email, a form submission, a status change. Detect it automatically rather than relying on someone noticing.
Assemble the context. This is where AI contributes most. Rather than passing along a document, gather what the receiver will need: the relevant customer history, the related order, the applicable policy section, a summary of the preceding thread.
Prepare a proposed next step. A draft reply, a completed form, a recommended classification. The receiver's job becomes checking and adjusting rather than starting.
Route it with ownership. Assign it to a person or queue with a visible state and a time. If it sits too long, that becomes apparent.
Record the outcome. What was decided and what was corrected. This is your quality signal and your improvement material.
A concrete example
An enquiry arrives in a shared inbox. Today, someone reads it during a triage round, decides whose it is, and forwards it. That person opens it some hours later, looks up the customer, reads the history, and writes a reply.
With the pattern applied: on arrival, the system identifies the customer, retrieves their recent history, classifies the enquiry, drafts a reply grounded in the relevant documentation, and places the whole package in the correct person's queue within a minute.
The person still decides and still sends. But they open something with the context already assembled and a draft to work from, and it arrived immediately rather than after a triage round.
Why this is a good first project
Nobody loses their job or their judgement — the same person makes the same decision. Errors are caught naturally, because the receiver is reading it as part of their normal work. The benefit is visible quickly in turnaround time. And adoption is easy, because receiving a prepared item is plainly better than receiving a bare one.
What to watch
Do not let the assembled context become so long that people stop reading it. The point is to save the receiver effort, not to relocate the effort.
And check that you have not simply moved the queue. If items now arrive instantly at a person who was already the constraint, you have made the waiting more visible without making it shorter — which is useful information, but it is not the improvement you were expecting.
