Map the manual work before you automate: a one-week time study
Before building any AI workflow, I time the manual loop for a week. The method, the four loops I look for, and the sheet I use.
The most common mistake I see is starting with a tool. Someone has seen a demo, picks a platform, and wires up whatever looked impressive. A week later the team is doing the same manual work plus maintaining an automation.
I start by sitting next to the team and timing the work.
The week
01Shadowsit with two or three agents for a full shift
02Logevery task, start and end, in one sheet
03Groupcluster tasks into loops that repeat
04Priceminutes per loop x times per day
05Pickautomate the most expensive repeatable loop first
The four loops that usually eat the day
Sortingreading a message to decide what it is about
Tagginglabelling it so reports and routing work
Copyingmoving the same facts between inbox, CRM and chat
Chasingasking someone for an update that should have arrived
In my current role these four loops added up to about two hours per agent per day. That number became the target, and it is why the classification and routing workflow took it down to about twenty minutes.
The sheet
| Column | Why it is there |
|---|---|
| Task | named in the team's own words |
| Loop | sorting, tagging, copying or chasing |
| Minutes | start to finish, including switching apps |
| Times a day | counted, not estimated |
| Needs judgement? | yes means a person stays in the loop |
The last column matters most. It tells you which steps can run alone and which need a quality gate and a human review.