Patterns Mimir surfaces across your sources, ranked by how often they appear and how much they matter.
Insights are recurring patterns Mimir surfaces across your sources. Rather than matching keywords, Mimir reads everything you upload, pulls out the important signals (pain points, feature requests, observations, metrics, and quotes), and groups related ones into themes.
That means similar signals from different sources get grouped together even when they use completely different words. A customer saying “I can never find the export button” in an interview and another writing “downloading reports is painful” in a survey will both appear under the same pattern — because the underlying signal is the same.
Every insight is backed by traceable evidence from your original sources. No vague summaries — everything links back to what real people actually said.
Insights are generated automatically after you upload sources and run analysis. There are two steps.
Reading your sources
Finding patterns
The insights page shows all patterns Mimir found in your project, ranked by severity. Each card gives you three pieces of information at a glance.
Severity
A badge showing critical, high, medium, or low. Severity reflects how significant the underlying problem or opportunity is based on what your sources describe. Critical patterns represent issues that are actively blocking users or causing churn.
Frequency
How many of your sources mention this pattern. “Mimir found this pattern in 4 sources” is a much stronger signal than 1. Frequency tells you how widespread a pattern is across your data.
Summary
A concise description of the pattern Mimir found. Written in plain language — you should be able to read it and immediately understand what your users are experiencing. Click any insight to see the full evidence.
Click any insight from the list to open its detail page. This is where you go from “something is happening” to “here's exactly what people said.”
Evidence drill-down
The core of the detail page. Every piece of evidence is an exact quote or observation traced back to its original source. You can see which source it came from, what kind of signal it is (pain point, feature request, observation, metric, or quote), and the full context of what was said. No paraphrasing — these are the actual words from your data.
Source attribution
Each piece of evidence is linked to its source. If you uploaded an interview transcript called “Customer call — Jan 15”, you'll see that name next to every quote pulled from it. This makes it easy to trace any pattern back to its origin and verify the context.
Linked recommendations
At the bottom of the detail page, you'll see which recommendations address this pattern. This closes the loop from “what's happening” to “what should we build.” Click any linked recommendation to jump to its detail page.
Breadcrumb navigation
A breadcrumb trail at the top of the page lets you navigate back to the insights list without losing your place. The pattern is consistent across all detail pages in Mimir.
Insights aren't static. When you add new sources and re-run analysis (using /analyze in chat), Mimir compares the new results against your previous set and surfaces what changed. This is how you track your evolving understanding of users over time.
New patterns
Patterns that didn't exist in your previous analysis. Often emerge when you bring in a new type of source or feedback from a different user segment.
Strengthened patterns
Patterns that gained more evidence — higher severity, higher frequency, or both. When a pattern strengthens across multiple re-analyses, the signal is getting louder and it's probably worth acting on.
Weakened patterns
Patterns that lost severity or frequency. New sources might contradict earlier ones, or the issue may have been more niche than it initially appeared.
Resolved patterns
Patterns that dropped out entirely. This usually means you shipped something that fixed the underlying problem, or the concern turned out to be temporary. The best kind of change.
Start with at least 3 diverse sources
A single interview gives you data points. Three or more sources from different channels (interviews, surveys, support tickets, usage data) give you patterns. Comparing across sources is where Mimir adds the most value.
Re-analyze as you gather more data
Don't treat analysis as a one-time event. Add new sources as they come in — a fresh batch of support tickets, a new round of interviews, updated analytics — and re-run to see how your understanding evolves. Use /analyze in chat to trigger a re-analysis.
Use evidence quotes with stakeholders
When making the case for a product decision, link back to the evidence on insight detail pages. Direct quotes from real users are more persuasive than any summary. “Mimir found this pattern in 8 out of 12 sources” with the actual quotes to back it up is a strong argument.
Prioritize high severity + high frequency
Patterns with both high severity and high frequency are your strongest signals. A critical issue mentioned across most of your sources is almost certainly worth addressing. These are the patterns that should flow directly into recommendations.