Build a Research
Repository Teams
Actually Use
Most research repositories become graveyards within six months. Data goes in; nothing comes back out. The problem is not the tool — it is the structure. Reports are unsearchable. Atomic insights are not.
What is a research repository?
“A research repository is a centralised, searchable home for an organisation's research — ideally stored as atomic insights (an observation + its supporting evidence + tags) rather than buried in reports — so findings stay findable, comparable, and reusable over time.”
The idea of atomic research was developed by Tomer Sharon and the atomic research community as an answer to the most common failure mode in research knowledge management: teams producing more research than they can ever retrieve or reuse. The volume of qualitative data is growing faster than any team can process. The answer is not more storage — it is better structure.
An atomic insight is a single, self-contained observation: one claim, the evidence that supports it, and the tags that make it findable in any future query. When every study produces 20–40 atomic insights instead of one 40-page report, the repository becomes a compounding knowledge asset rather than a filing cabinet.
Report library vs. atomic repository: a side-by-side comparison
| Aspect | Report library | Atomic repository |
|---|---|---|
| Unit of storage | Full report (20–94 pages) | Single observation + evidence + tags |
| Findability | Search by title or date | Search by theme, tag, product area, participant type |
| Reusability | Low — requires reading the full document | High — surfaced by relevance queries in seconds |
| Trust signal | None — all reports equal | Confidence index; deprecate stale insights |
| Evidence link | Buried in the document | Directly linked to highlight, quote, or session recording |
| Deduplication | Manual; rarely done | AI-detectable and mergeable across studies |
| Lifecycle | Static — never updated | Living — gains or loses confidence over time |
| Cross-study patterns | Impossible without re-reading everything | Surfaced automatically by tag clustering |
Six reasons repositories become graveyards
A repository graveyard is not a tool problem. Every one of these failure modes is a structural problem — solvable with the right architecture and the right habits built into the research workflow.
Report dumps nobody searches
Most teams store their research as the deliverable: the slide deck, the PDF, the Notion page. These are optimised for communicating a single study's findings — not for retrieval six months later, by someone asking a different question. The structure that makes a report compelling makes a repository unsearchable.
No evidence trail means low trust
An insight without a source is an assertion. "Users struggle with the checkout flow" — from which study? How many participants? On which segment? When evidence is buried in a report or missing entirely, stakeholders can't interrogate the claim, and stop trusting it. Low-trust insights get ignored; the repository becomes decorative.
Inconsistent tagging makes retrieval impossible
Tag systems without governance fragment over time: one person tags by feature area, another by user type, another by methodology. A query for "onboarding" returns three of twenty relevant insights because the others were tagged "signup", "first session", or "new user flow". Retrieval failure looks like repository failure — but the root cause is taxonomy drift.
Duplicates erode confidence in the whole system
When the same user behaviour has been researched three times and appears as three separate insights with slightly different wording and conflicting confidence levels, which one do you cite? Duplicates don't just waste storage — they undermine trust in every other insight in the system.
Staleness: outdated insights cited as current truth
A repository without lifecycle management becomes a liability. An insight from two years ago about a feature that has since been redesigned looks identical to a current insight. Teams cite outdated findings without knowing they're stale — and make decisions on evidence that no longer reflects reality.
Isolation from decisions and the product backlog
A repository that sits outside the product workflow is consulted at best occasionally. When insights are not connected to opportunities, backlog items, or the tools teams use daily, researchers become the only people who know the repository exists — and the only people who benefit from it.
6 principles of a repository teams actually use
These principles are not independent choices — they are a system. A repository that is atomic but not evidence-backed produces searchable assertions. One that is searchable but not deduplicated produces confident contradictions. All six are required for a repository that compounds in value rather than decaying.
Store insights, not reports
Each entry in your repository should be one observation: one thing you learned, from what evidence, tagged with what it applies to. When a study yields 30 atomic insights instead of one report, every insight is independently retrievable — even years later, by someone asking a completely different question.
Insights Repository →Every insight traceable to its source
Link every insight to the highlight, quote, or session recording that supports it. An insight without evidence is an assertion — unquotable in a stakeholder meeting and uncitable in a product decision. Evidence links transform 'we think users struggle here' into 'seven of eight participants couldn't complete this step — here is the recording timestamp.'
Insights Repository →Consistent taxonomy enforced, not suggested
A tag system without enforcement is a tag system that fragments. Define your taxonomy upfront: product areas, user types, study methods, insight types (pain point, mental model, job to be done, behaviour, expectation gap). Apply it consistently across every study. AI can assist with auto-tagging and taxonomy consistency — but the taxonomy itself needs a human owner.
Research Agents →Keep truth singular
When the same user behaviour has been observed in three separate studies, the repository should have one insight with three supporting evidence links — not three separate insights pointing in slightly different directions. Deduplication is the discipline that keeps the knowledge base coherent at scale. AI can surface overlap; researchers make the merge decision.
Research Agents →Insights gain and lose confidence over time
A repository is not an archive — it is a living knowledge base. Insights should carry a confidence state: current, uncertain, or deprecated. When a feature is redesigned, the insights about its old behaviour should be deprecated, not deleted. When new research contradicts an existing insight, it should be challenged and updated rather than silently coexisting with the old truth.
Insights Repository →Linked to decisions, accessible where teams work
An insight that lives only in the research tool has limited leverage. Connect insights to product opportunities, backlog items, and design decisions. Make the repository queryable from the tools teams already use — including via MCP for AI-assisted workflows. The closer insights are to the moment of decision, the more often they shape it.
Platform and MCP →Repository health checklist
Use this to audit your current repository — or to set the standard before building one. Eight or nine items checked: healthy and compounding. Fewer than six: your repository is at graveyard risk.
- Insights are stored atomically — one observation per insight, not one report per study
- Every insight links to the highlight, quote, or session that supports it
- A consistent taxonomy is in use and enforced across all studies
- Duplicate insights are detected and merged on a regular cadence
- Stale or superseded insights are deprecated, not silently left alongside current ones
- Insight search returns relevant results in under 30 seconds
- The repository is queried before new research is commissioned
- Non-researchers can find insights independently without researcher help
- Insights are connected to product opportunities or backlog items
What a living repository delivers
The compounding value of an atomic, searchable repository — in numbers from customers and research industry data.
France Télévisions halved their research cycle time with Usedge — a direct result of not starting from scratch each study, and AI automation handling the processing work.
France Télévisions, Usedge customer
France Télévisions researchers were fully productive after their first study. Onboarding friction eliminated — which is what atomic structure and consistent taxonomy make possible.
France Télévisions, Usedge customer
France Télévisions expanded research access to 18 designers and 12 product managers and owners — all feeding insights into and retrieving from the same system of record.
France Télévisions, Usedge customer
Organisations with systematic research infrastructure — centralised, reusable, connected to decisions — report 2.7× better overall product outcomes versus ad-hoc research practices.
Maze, Future of User Research
A repository built on all six principles by default
Usedge's Insights Repository was designed around atomic research from the start — not retrofitted from a document storage tool. Every capability maps to a specific principle that prevents the graveyard failure mode.
Usedge stores research as atomic insights by default — not as uploaded reports. Each insight is a structured object: claim, evidence link, type, confidence score, and tags. Studies do not produce a deliverable; they produce a set of queryable knowledge units.
Every insight links to the highlight or session that produced it. Stakeholders can jump from an insight to the exact moment in the session recording where a participant demonstrated the behaviour. The evidence chain is navigable in seconds, not reconstructed from memory.
AI agents apply consistent taxonomy tags during analysis — product area, user type, insight type, study method. The taxonomy is enforced, not optional, and AI assists researchers in maintaining consistency across contributors and studies over time.
Research agents detect when a new insight semantically overlaps with an existing one and surface the match for researcher review. Researchers decide to merge, keep separate, or update confidence. Duplicates are caught systematically, not by hoping someone remembers a study from two years ago.
Insights carry lifecycle states: current, uncertain, deprecated. When a product area changes, affected insights can be marked for review or deprecated — without deleting the historical record. The repository reflects what is true now, while preserving what was true before.
Insights link to product opportunities and backlog items directly. For teams using AI-assisted workflows, Usedge exposes the repository over MCP — meaning AI agents can query the knowledge base in real time during planning, design, and development, bringing research to the moment of decision.
Quant and qual in the same system. Usedge also connects quantitative research — surveys, analytics integrations — to the same repository. Insights from qual sessions and quant data live under the same taxonomy and the same evidence chain. When qual surfaces a hypothesis and quant validates its prevalence, both pieces of evidence attach to the same insight.
Frequently asked questions
What is a research repository?
A research repository is a centralised, searchable home for an organisation's research findings — ideally structured as atomic insights (individual observations linked to their evidence) rather than a library of PDF reports. Its purpose is to make past research findable, comparable, and reusable over time, so teams stop re-researching what they already know.
What is an atomic insight?
An atomic insight is a single, self-contained observation from user research: one claim, the evidence that supports it (a highlight, quote, or session recording link), and the tags that make it findable in future queries. The concept was developed by Tomer Sharon as the answer to why research repositories become graveyards: reports answer the question they were designed for; atomic insights answer questions they were never designed for, because they are independently searchable.
Why do research repositories become graveyards?
Six failure modes, compounding: storing reports instead of searchable insights; no evidence links so insights cannot be verified; inconsistent tagging so retrieval fails; duplicate findings eroding confidence in the system; stale insights never deprecated; and no connection to product decisions so the repository stays outside the workflow teams actually use. Any one of these degrades a repository; all six together produce a graveyard.
How do you keep a research repository organised at scale?
Three disciplines, non-negotiable at scale: (1) a consistent taxonomy enforced across every study — product areas, user types, insight types — applied by AI and verified by researchers; (2) systematic deduplication on a regular cadence, AI-assisted to detect semantic overlap across studies; (3) insight lifecycle management — marking findings as current, uncertain, or deprecated as the product and organisation evolve. Without all three, scale produces chaos rather than compound knowledge.
What is the difference between a research repository and a report library?
A report library stores documents. A research repository stores knowledge. Reports are optimised for communicating a single study's findings — they are linear, context-dependent, and unsearchable by theme. Atomic insights are independently queryable: someone asking a question you never anticipated six months ago can still get a relevant answer in seconds. The structural difference is the difference between retrieval and rediscovery.
How often should a research repository be updated?
After every study — and the update should happen as part of the analysis workflow, not as a separate archiving step. If publishing to the repository is a separate task, it will be skipped. The best repositories are ones where insights are created during analysis, tagged automatically, and published as a natural output of the study process rather than an administrative overhead added at the end.
Build a repository that teams actually search and cite
Atomic insights, traceable evidence, AI deduplication, and lifecycle management — Usedge is the Insights Repository built on all six principles from the start.