Never automate a process you can't explain on a whiteboard
When we started building Papervee inside Pixel Solutions, the trigger wasn't a grand product vision. It was our own HR paperwork: leave requests, time tracking, employment documents - a swamp of PDFs, emails and "ask Ramona, she knows".
The first thing we tried to do was automate it. The first thing we actually had to do was understand it.
The whiteboard test
Before writing any code, we tried to draw our own leave-approval flow on a whiteboard. Three co-founders, one company, maybe forty employees at the time - and we couldn't agree on the diagram. Who approves when the manager is on leave? What happens to a request submitted for last week? Which document does labor law actually require?
If a company of our size couldn't whiteboard its own process, what were the chances our future customers could?
That test became a rule I now apply to every automation and AI project:
If the process owner can't draw the flow - every branch, every exception, every "it depends" - you're not ready to automate. You're ready to discover.
Discovery is the actual product work
The unsexy truth is that most of the value in an automation project is created in the discovery phase:
- Exceptions get named. "It depends" becomes a rule, an escalation, or an explicit human decision.
- Dead steps get deleted. Every process accumulates steps that exist because of a problem from 2019. Automating them preserves the fossil.
- Ownership gets assigned. A flow with no owner becomes an automated flow with no owner - which is worse, because now nobody even sees it happening.
What this means for AI
This rule matters more with AI agents, not less. An agent making decisions inside a fuzzy process doesn't remove ambiguity - it hides it behind confident-sounding output. Whiteboard first. Then automate. Then, when the boring path runs itself, let AI handle the edges with a human in the loop.
We eventually did automate that leave flow - and a few hundred companies use that same flow in Papervee today. But the code was the easy part. The whiteboard was the product.