The most common mistake we see when a business tries to automate its own processes is starting with the software. Someone picks a tool, tries to fit their work into it, and ends up with a system that models the tool's assumptions rather than how the job actually gets done. The better starting point is far less technical: write down, in order, exactly what one employee does on a normal day.
Step 1: Shadow the actual routine, not the job description
A job description says a coordinator “manages customer orders”. What they actually do is check WhatsApp for new messages, copy order details into an Excel sheet, email the warehouse, then call the customer to confirm delivery. The gap between the description and the routine is where the real automation opportunities live. Spend half a day watching, or ask the employee to note every task and every handover for one full day.
Step 2: Break the routine into discrete steps
List every step as a single action — not “process the order” but each individual thing that happens: message received, details copied into sheet, warehouse emailed, customer called, sheet updated again. Each of these is a candidate to examine on its own.
Step 3: Mark where information is retyped or re-checked
- Any point where the same information is typed into a second system is a retyping step — a strong automation candidate.
- Any point where someone waits for another person before continuing is a handover — a candidate for routing and reminders.
- Any point where the outcome depends on one specific person's memory or judgement is a knowledge dependency — a candidate for documentation or an AI knowledge system.
Step 4: Draw the routine as a chain
Once the steps are listed, arrange them as a simple left-to-right chain. This makes retyping and waiting visible in a way a bullet list does not — you can see the same piece of information travelling from a WhatsApp message through a spreadsheet, an email and a phone call before it reaches its destination.
Example: an order coordinator's daily routine, mapped
WhatsApp order received
Details copied to Excel
Manual retyping
Email sent to warehouse
Manual retyping
Call to confirm delivery
Sheet updated again
Third re-entry
Redrawn after automation, the same routine becomes a single flow: the order is captured once, routed to the warehouse automatically, and the coordinator is left with the one task that genuinely needs a person — speaking to the customer.
Step 5: Decide what still needs a person
Not every step should be automated. Judgement calls, relationship-building conversations and exceptions usually still need a person. The aim of mapping the routine is to remove the mechanical steps around those moments, not to remove the person from the process — see how this distinction plays out in Workflow & SOP Automation.
Step 6: Document the workflow before you automate it
Write the finished, agreed workflow down as a simple procedure — who does what, in what order, and what happens on exceptions — before building anything. This document is also what makes the process resilient to staff turnover, and it becomes the basis for an AI Employee & Knowledge System if you want new staff to be able to ask questions about it later, rather than shadowing someone for weeks.
This mapping exercise takes a day or two per role. It is worth doing properly, because a workflow built around a poorly understood routine simply automates the confusion faster.

