# Atomic research: reusable findings instead of reports nobody reopens

> Atomic research stores findings as small units (observation, evidence, tags) reusable across studies. Its structure, taxonomy design, and when it fits.

*Tags: ux, research, operations, repository, knowledge*

---


> [!info] Quick Definition
> **Atomic research** is an approach for storing research findings as small, reusable units, called *nuggets*, instead of inside long reports. Each nugget is a tagged observation backed by evidence that can be combined with others to answer new questions.


## Where does it come from?

Tomer Sharon and Daniel Pidcock developed it in parallel, inspired by Brad Frost's atomic design, according to User Interviews. Sharon, then at WeWork, created Polaris, an atomic repository template. Koji reports that Pidcock presented it at UX Brighton in 2018, after his team kept re-running studies because earlier answers were hard to find.

That is the problem it solves: reports are read once, filed away, and their knowledge is lost. When a new team arrives, it repeats studies another team already ran.

## What is a nugget?

According to Sharon, a nugget is a tagged observation supported by evidence. Koji describes it as the smallest unit of research that still carries meaning: more than a quote, less than a full report. It has three parts:

- **Observation:** what was learned, both the what and the why.
- **Evidence:** the proof, such as a short video clip, a screenshot, audio, or text.
- **Tags:** keywords for filtering and searching.

Illustrative example of a nugget: "A user cannot find the cancel-subscription button because she looks for it in the help menu". It comes with a 20-second clip and tags such as "subscriptions" and "navigation".

## How do you tell a fact from an insight?

Pidcock proposes a chain of four elements, which Koji calls EFIR:

| Element | What it is | Illustrative example |
|---|---|---|
| Experiment | The study or session that produced it | Interviews with annual-plan customers |
| Fact | An observation from a single experience | "Three of five people look for cancellation in the help menu" |
| Insight | The interpretation derived from the facts | "Cancellation is not where users expect it" |
| Recommendation | The action that follows from the insight | "Add the cancel option on the account page" |

This separation avoids a frequent error: calling anything an "insight". A fact is not a conclusion. According to Koji, the levels link many-to-many, so one fact can support several insights and one insight can rest on several facts.

## How do you design a repository?

User Interviews describes a seven-step process:

1. Establish a taxonomy.
2. Do the research as usual, without changing the method.
3. Identify single-experience facts (unbiased quotes, observations, statistics).
4. Derive insights from the facts.
5. Identify opportunities or recommendations.
6. Record the nugget with observation, evidence, and tags.
7. Share it through several channels.

The tool is secondary. Koji mentions Airtable and Notion as spreadsheet or database options, and dedicated platforms such as Dovetail, Notably, Marvin, Condens, EnjoyHQ, and Glean.ly. According to Koji, most teams outgrow a spreadsheet around 500 nuggets.

The most important piece is the **taxonomy**, the vocabulary of tags. Koji calls it the most important upfront decision: with free-text tags, one theme fragments into many labels, and it recommends a small set per nugget, three to six, drawn from a controlled vocabulary. User Interviews proposes five families of tags:

- **Procedural:** date, source, research method, evidence type.
- **Demographic:** age, location.
- **Experience-oriented:** frequency or magnitude of the observation, the participant's emotional state.
- **Business-oriented:** revenue range, business unit, product line.
- **Service design-oriented:** journey, act, scene, character.

The same source warns that the taxonomy must be able to change as your customer knowledge and goals change.

## What value does it build over time?

With a repository, you can search for patterns that repeat across studies: moments of frustration or delight along a journey, for example. You can also combine facts from different studies to back an insight with more evidence. That way knowledge stops aging inside an archived report.

## When should you NOT use it?

Atomic research has costs. Koji says the most cited criticism is that tagging is slow, and it estimates that atomizing and tagging a 60-participant study adds about one and a half to two days beyond normal report writing. User Interviews gathers other criticisms: context is lost when a finding is separated from its study, converting older reports takes time, and it fits poorly with quantitative research. Sharon suggests running quantitative work in parallel to explain the why behind the metrics.

A repository can become a graveyard if nobody consults it. Without a habit of searching before researching, nuggets have no effect. And a small team that runs two studies a year probably does not need a complex repository.

Sharon identifies three situations where it works best: large, dispersed organizations, organizations with many customers, and experiences delivered across several channels. Nuggets are used alongside reports, not instead of them: the report keeps the context a nugget can lose.

## Example

Illustrative example: an insurance company runs four studies in a year, one of them about onboarding. Each study produces nuggets with tags for process, customer type, and journey stage. At year's end, searching for "renewal" and "confusion", the team finds eleven nuggets from three different studies that point to the same problem with the expiry notice. No single report showed it clearly, but the repository made it visible. The team turns the pattern into an insight and a recommendation for the next quarter.

## Common mistakes

- **Storing whole reports** in the repository and calling it atomic research.
- **Using free-text tags**, so each person uses them differently.
- **Confusing facts with insights**, and presenting opinions as data.
- **Tagging late**, when nobody remembers the context of the session.
- **Creating a huge taxonomy** from the start.
- **Not building the habit of searching** before a new study.

To capture and share findings in the moment, [presenting research findings](/en/wiki/research/presenting-research-findings/) explains how to turn them into decisions. If the research combines methods, [triangulation](/en/wiki/research/triangulation/) helps cross nuggets from different sources. And for the base method behind the observations, [user interviews](/en/wiki/research/user-interviews/) are the usual starting point.

## Sources

- [Atomic Research: The Complete Guide to Research Nuggets and Insight Repositories, Koji](https://www.koji.so/docs/atomic-research-nuggets-guide)
- [Atomic research nuggets, User Interviews (UX Research Field Guide, chapter from the 2019 version)](https://userinterviews.com/ux-research-field-guide-chapter/atomic-research-nuggets)


---

Source: https://www.fernandoux.com/en/wiki/research/atomic-research/
