Manual First, Automate Second
Prove a high-touch process manually before encoding it in software
- Difficulty
- Moderate
- Time to result
- ~months to results
- Steps
- 5
- Confidence
- 98%
This framework treats manual work as process discovery rather than permanent operating design. A team first performs a task by hand to learn the customer's needs, the exceptions, and the decisions required for a good outcome. Once the team has established that the activity is useful, it extracts the repeated parts and automates them. The automation can execute the task itself or help an employee move faster, as Send My Bag did with customs processing, tracking, and canned-response tools for support agents. The company keeps people involved where human service matters, while software absorbs predictable work. Repeating this cycle across operations compounds capacity and lets the business grow with a smaller increase in headcount.
Origin
Extracted from The Foundr Podcast
Core principles
- 01Use manual work to discover what good service requires
- 02Automate only after the task proves valuable
- 03Support people with tools before replacing them
- 04Fund efficiency from operating cash flow
How to run it
- 1
Select a recurring task
Choose a manual activity that occurs often enough to constrain service or growth.
Pro tip Start where employees repeat the same decisions or data entry.
- 2
Deliver it manually
Perform the task at a high standard and observe what customers and employees actually need.
Pro tip Record exceptions instead of hiding them.
Watch out Do not automate an imagined workflow before proving the task is useful.
- 3
Extract the pattern
Separate repeatable inputs, rules, and outputs from judgement-heavy exceptions.
Pro tip Turn frequent answers and decisions into structured templates.
- 4
Build the smallest aid
Create a tool that removes the repetitive portion while preserving human handling where it adds value.
Pro tip An agent-assist tool may be safer than full automation.
Watch out Do not sacrifice service quality merely to remove headcount.
- 5
Measure and repeat
Compare capacity, response quality, and exception rates, then apply the cycle to the next constraint.
Pro tip Keep improving proven internal tools as volume grows.
In the wild
Send My Bag wanted genuine people to answer live-chat questions rather than relying on the ineffective bots available at the time. The company built tools and a database of canned responses that helped agents answer faster while preserving a human conversation.
→ Support agents could handle more work without handing the customer experience entirely to a bot.
After manually learning the operational requirements, Send My Bag invested in internal customs-processing and tracking tools. The systems simplified documentation and passed cleaner information into partner networks.
→ The company reduced repetitive processing while supporting international scale.
Common mistakes
Automating before proving value
A fast version of the wrong process still wastes resources and may lock in bad service.
Ignoring exceptions
Automating only the happy path without a human fallback can turn unusual customer cases into failures.
Replacing judgement unnecessarily
Some tools should assist employees rather than remove the person customers need.
Is it for you?
Best for
It is best for repeatable operational work whose quality can first be demonstrated by a human team.
Not ideal for
It is not ideal for rare tasks, unstable processes, or decisions where the cost of an automated error is unacceptable.
From the transcript
“once we have established that that is a good idea whatever the particular task 22 30 that we're testing is and we've just hammered it…”
“we have literally dozens of tools that we've built over the years”
“it can still be 23 30 a genuine person there to help you”
From the episode
619: Airline Charged Me $65 - So I built a $250M Competitor