Workflow automation discovery is the step of finding, capturing and documenting the repetitive tasks people already do by hand — so they can be turned into automations. It happens before you open Zapier, write a script, or brief an engineer. Instead of guessing what to automate, you record a real task as it actually runs and turn that recording into a precise, step-by-step specification a machine can follow.
Most automation projects skip this step, and it's exactly why so many of them stall. The market is full of tools that can run an automation. Almost nothing helps you answer the harder question that comes first: which workflow, and what does it actually involve?
In one line: automation tools tell you how to build. Workflow automation discovery tells you what to build — captured from real work, not a whiteboard.
Why the bottleneck isn't the tools
Ask anyone on an operations, finance or support team to list the things they do every week and they'll rattle off ten. Ask them to write one down precisely enough that someone else could automate it, and it falls apart. Which report? Filtered how? Which columns get copied where? What happens when a value is missing? What's the exact Slack message, and who gets tagged?
The knowledge lives in muscle memory. It's never written down because writing it down is tedious and the person doing it is busy actually doing it. So the workflow stays manual — not because it can't be automated, but because it was never captured in a form anything could act on.
You can't automate a workflow you can't describe. Discovery is how you describe it without sitting down to write a spec.
Discovery vs. process mining vs. task mining
These terms overlap, so it's worth being precise:
- Process mining analyses event logs from big enterprise systems (ERP, CRM) to map how a process flows across an organisation. It's powerful at scale, heavy to set up, and produces high-level diagrams rather than buildable automations.
- Task mining watches desktop activity broadly to surface patterns across many users — useful for prioritising, but noisy and privacy-sensitive.
- Workflow automation discovery is task-level and intentional. One person records one recurring task in their browser, on purpose, once. The output isn't a chart — it's a clean, editable workflow ready to export and run.
If process mining is a satellite map of the whole city, discovery is you walking one route and dropping a pin at every turn so anyone can repeat it exactly.
The four stages of workflow automation discovery
Whether you do it with sticky notes or a recorder, good discovery moves through the same four stages.
1. Spot the candidate
Look for work that is repetitive, rule-based and browser-bound. A useful filter: if you do it the same way more than once a week and a checklist would describe it, it's a candidate. Weekly pipeline reports, AR aging follow-ups, onboarding new hires, ticket triage — these are the usual suspects. Our use-case library breaks them down team by team.
2. Capture it as it really runs
This is where most teams lose the detail. The goal is to capture the actual sequence — every tab, click, filter and field — not an idealised version someone reconstructs from memory a week later. The most reliable way to do this is to record the task while you do it, so nothing gets rounded off.
3. Reconstruct the logic
A raw recording is just clicks. Discovery turns it into meaning: these three steps pull the data, this one decides whether a deal slipped, this one writes the summary. You end up with inputs, steps and decision points — an editable workflow rather than a video.
4. Hand it to a builder
Finally, the captured workflow goes to whatever will run it: an automation platform, an AI assistant, or a teammate following a guide. Because the workflow is already precise, this last step stops being a research project and becomes a copy-paste.
The payoff: when discovery is done right, the automation almost builds itself — because every ambiguous decision was already captured the moment the task was recorded.
How Spion does discovery in your browser
Spion is built around this exact idea. You hit record, do the task once the way you always do, and Spion watches the tabs, clicks and data you touch. It then reconstructs the steps, inputs and decisions into a clean workflow — and exports a ready-to-run blueprint to Claude, Workato, Make, Zapier or n8n. Not a screen recording you have to rebuild. An actual automation.
The person who knows the workflow is the person who captures it — no analyst interviews, no engineer reverse-engineering a process, no blank spec document. That's discovery collapsed into a single recording.