n8n Automation Mining Skill Improves Task Discovery in Synthetic Company Tests
n8n Labs reports that its updated Automation Mining skill found more recurring tasks in simulated company activity, though the results are not production time-saving data.
n8n Labs has published an updated Automation Mining skill that improved how well an AI-assisted process identified recurring tasks in a controlled test. In an evaluation involving 57 synthetic companies, verified task recall increased from 28.2% to 51.7%. The result suggests that AI agents may become more useful for finding automation candidates before a team has to map every process manually.
The important qualifier is that this is an n8n Labs experiment, not a production benchmark. The test used synthetic activity rather than live company logs, so it does not establish real-world time savings, cost reductions, or workflow reliability. Still, the experiment offers a concrete look at how n8n is approaching one of the hardest parts of automation: deciding what is worth automating in the first place.
According to the official Automation Mining skill repository, the method draws on a taxonomy with 907,353 workflow rows to define tasks and synthetic activity. Two AI models read the simulated-company data and generated automation proposals using the updated procedure.
What the experiment measured
Automation projects often begin with discovery: identifying repetitive work, understanding which systems are involved, and separating high-value opportunities from tasks that are too irregular or ambiguous to automate safely. n8n's experiment evaluates whether an AI-driven skill can identify known tasks from activity data and generate proposals that are useful.
The published results cover two metrics:
- Verified task recall: the share of known recurring tasks the procedure found.
- Useful proposal quality: the share of proposals assessed as useful in the experiment.
- Test environment: 57 synthetic companies, using synthetic activity read by two AI models.
| Metric | Earlier procedure | Updated procedure | Reported change |
|---|---|---|---|
| Verified task recall | 28.2% | 51.7% | 23.5 percentage points |
| Useful proposal quality | 44.7% | 53.0% | 8.3 percentage points |
The recall improvement is the headline result. Put simply, the updated procedure found roughly half of the recurring tasks included in the synthetic test, compared with a little over one quarter under the earlier procedure. The quality metric also improved, which matters because finding more tasks has limited value if the resulting recommendations are not useful.
What the result does and does not show
The findings support a narrow conclusion: within this synthetic evaluation, the updated procedure performed better on the reported task-discovery and proposal-quality measures. They do not show that every organization can expect the same level of recall, nor that automating the identified tasks will save a defined number of hours or dollars.
That distinction matters for teams evaluating automation tools. Real operational data can be incomplete, inconsistent, or spread across email, spreadsheets, CRM systems, finance software, and other applications. A task may also be technically automatable but still require human judgment, approval, or exception handling. The repository's accompanying whitepaper explicitly frames the work as a methodology and reports results from synthetic activity, rather than claiming measured production outcomes.
How the skill is available
The Automation Mining skill can be installed using npx skills@latest add n8n-io/automation-mining-skill --skill automation-mining. The repository says it supports multiple AI-agent environments, including Claude Code, Codex, Cursor, and Pi.
The supplied materials provide installation guidance but do not provide separate pricing details for the skill. Businesses considering it should also distinguish between the discovery method itself and the subsequent work of designing, building, testing, monitoring, and maintaining the workflows that a discovery exercise may identify.
Practical implications for workflow automation
For managers, the experiment's value is not a promise of automatic savings. It is evidence that structured task discovery can improve when an AI agent is given a taxonomy and a more refined procedure. That could make early automation assessment less dependent on lengthy workshops or ad hoc guesses about where repetitive work exists.
A sensible application would be to use an automated discovery output as an initial shortlist, then validate each proposal against the actual process. Teams should ask whether the task is frequent enough to matter, whether its inputs are reliable, what happens when data is missing, and where a person must remain in control.
Potential candidates may include repetitive information transfers between business applications, routine status updates, lead routing, document-handling steps, and notifications triggered by defined events. Whether any of these are appropriate depends on a company's own systems and process rules. The n8n experiment does not claim that these use cases will work universally.
Scalevise can help turn an automation shortlist into workflows that reduce manual handoffs without creating fragile processes. From mapping data flows and exceptions to building n8n integrations with monitoring, the team can help focus effort on opportunities that are practical to operate. Explore Scalevise's n8n setup service to discuss an n8n automation project with Scalevise.
Frequently Asked Questions
What is n8n Automation Mining?
n8n Automation Mining is an n8n Labs experiment and skill for helping AI agents identify recurring tasks and generate automation proposals from activity data.
How much did the updated procedure improve task recall?
In the published synthetic test, verified task recall increased from 28.2% to 51.7%, a gain of 23.5 percentage points.
Were the Automation Mining results measured in real companies?
No. The reported evaluation used synthetic activity from 57 simulated companies, so it does not measure real-world production savings or outcomes.
Which AI-agent environments can use the skill?
The repository lists support for multiple AI-agent environments, including Claude Code, Codex, Cursor, and Pi.
Conclusion
n8n's Automation Mining experiment shows a meaningful improvement in task discovery within its synthetic evaluation, particularly on verified task recall. The result is promising for teams seeking a more structured way to surface automation opportunities, but it should be treated as a starting point for validation rather than proof of production savings. The real test will be whether the method produces useful, maintainable workflow candidates when applied to live business processes.