Why research gets run twice
Research gets run twice for familiar reasons. Nobody knows the first study happened, because its findings live in a deck in one team's drive. Somebody finds the deck but cannot tell which slide is a finding and which is a hunch, or whether it still holds. Or the findings are sound but tied to the decision they served, so a team asking a related question does not recognise them as relevant. A study that changed nothing the first time is the one most likely to be run again, and the diagnosis for research findings that are not acted on traces that failure back through the brief and the report before it reaches the repository. Where the repeated study is one of several things two teams have done twice, the diagnosis for marketing teams duplicating work traces each duplicate to the join between the teams that failed.
The site's pages each keep part of the record. The customer interviews tutorial ends by storing summaries where future teams can find them. The research design guide treats the brief and the findings together as the record of a study, and the guide to reporting research findings ends each study with a record of the decision taken. The audience research guide refreshes research on a cycle, and the content audience research guide dates every entry in its question bank so recent evidence can outweigh old. The segmentation guide asks for profiles written from research with an owner and a review date. Tests have their own memory: the experimentation programme playbook keeps a learning log, and the learning agenda tutorial feeds its answers back into the plan. None of these builds the store for research findings themselves, across studies and teams. That is the repository's job, and it closes the loop the methods section opens: every study starts from a decision and should end as evidence the next decision can find. Once research runs continuously rather study by study, the customer research programme playbook makes the repository one of its standing records, read at every monthly intake and reviewed once a year.
Before you start
Gather what exists: briefs, findings documents, decks, summaries and evidence tables from past studies, with the date and the team that ran each. Do not gather raw recordings or participant lists; they stay where they are.
Talk to whoever owns data protection. The repository will hold material derived from studies of people, and they need to agree what may go into it, who may read it and how the deletion dates set for each study's identifiable material are honoured.
Agree an owner. A repository without one fills with untagged material and is abandoned. The owner need not write every record, but they keep the tag set, check new records and run reviews.
Choose where it lives. A shared document or a simple database is enough to start; the structure matters more than the software. As general practice, choose a place everyone who briefs research can already reach.
The steps
Decide what the repository is for and who owns it
Write the purpose in a sentence: "so that anyone briefing research or making a customer decision can find what we already know, how strong it is and how old it is". Name the owner and the people who add records, usually whoever runs each study. Decide the scope: customer and market research findings, from interviews, surveys, diary studies, groups, desk researchResearch from existing sources, the business's own and published ones, done first so that interviews and surveys cover only what nobody has answered. and analysis. Test results stay in the learning log the experimentation playbook describes, linked rather than copied.
Set what each study leaves behind
Decide what every study deposits when it closes. As general practice, that is the brief, a short study record (decision, questions, method, who took part described by role or group, dates, who ran it, limits), the findings as separate records, the de-identified excerpts each finding rests on, and the evidence table if one was built, as the AI for research synthesis guide describes. Name these in the deliverables field of each brief, the field the research brief template already has, so every new study commits to leaving them behind. Equally, decide what each study does not leave: recordings, transcripts with names, contact details, screener answers and the recruitment log.
Keep personal data out of the findings
Recordings, contact details and anything else that identifies a participant are Personal DataAny information relating to an identified or identifiable living person, such as a name, email address, location data or an online identifier., as the personal data glossary entry explains. The principles the GDPR glossary entry summarises shape what the repository holds: data collected for a stated purpose and not reused in ways that purpose does not cover, limited to what is needed, kept in identifiable form no longer than necessary, with the business able to show that it complies. In practice, the repository holds findings and excerpts with names and identifying details removed, attributed by participant code, role or group, as the interviews tutorial's checklist does for quotes, and as the confidentiality principle the site's research guides set out from the MRS requires. Identifiable material stays in the study's restricted store and is deleted on the date set when consent was taken, which the tutorial on participant information and consent writes into the participant information and the study's records. The repository never sets that date; whoever owns data protection does.
Write each finding as its own record
A finding record states one thing the research found, the evidence for it, and how strong that evidence is. Write the finding as an observation in plain words: "adult learners who stop attending usually stop after a change in work hours, not after a poor lesson". Add the evidence: the excerpts or figures it rests on, with participant codes or table references, and the counter-evidence. Add strength in words: how many participants or respondents it came from, which groups, and whether it was a pattern or a single voice. One finding per record lets the same finding be found from different questions, which a paragraph in a deck never can.
Tag every record the same way
Tags are how records are found, and they work only if everyone uses the same ones, as a Naming ConventionAn agreed pattern for naming campaigns, ads, events and tracking parameters so that reports can group them reliably. does for campaign names: one agreed set of fields, each with a closed list of allowed values. A workable set is topic, segment, job, journey stage, method, date and strength. Take the segment values from the business's SegmentationDividing a market or customer list into groups that need different offers or messages. and its personasA research-based profile of a typical customer type: goals, problems and buying behaviour. as their pages define them, and the job values from the jobs-to-be-done framework, written as the jobs customers hire the product for. Keep the lists short, let only the owner add values, and tag every record in every field, using "not specific" rather than leaving a field blank.
Link findings to the decisions they informed
Give each finding a field for the decisions it informed: what was decided, by whom and when, with a link to the decision record or the plan. Where a finding led to a test, link the learning log entry; where a test result later confirmed or contradicted the finding, note it on the record. These links show which findings have shaped the business, which have never been used and which have been challenged, and they let a reader see the consequence of revising a finding.
Set access and a review date for each finding
Decide who may read and who may add. Most findings can be open to everyone who makes customer decisions, because the point is that they are found; some, such as findings about a named key account, need narrower access. Give every finding a review date and an owner. At review, the owner confirms the finding, updates it with newer evidence, or retires it with a note saying why. A retired finding stays visible, marked as retired, so nobody rediscovers it as new. The review cadence is the team's choice, set from how quickly its market and customers change.
Add the past studies worth keeping
Go through the material gathered before you started, newest first. For each study, write the study record and turn its findings into records with the same tags. Where a past study cannot show its evidence, or used a method its findings cannot support, record the findings as weak or leave them out with a note. Check that nothing identifiable comes across from old decks: names in quotes, screenshots of participants, contact lists in appendices. Old recordings found along the way go to whoever owns data protection, not into the repository.
Make a search of it the first step of every brief
The repository works only if people meet it at the moment they are about to commission research. Add a line to the brief: "what the repository already says", with links to the records found and their strength and dates. If the search answers the question well enough, the study may be smaller or unnecessary; if it shows weak or old evidence, the brief can say exactly what the new study must strengthen. When the study closes, it leaves behind its records, and the next brief finds them.
A finding record
The table shows the fields of a finding record and what each holds. Adapt the field names to the business, but keep one set across all records.
| Field | What it holds | Why it matters |
|---|---|---|
| Finding | One observation, in plain words | Can be found and reused on its own |
| Evidence | De-identified excerpts or figures, with participant codes or table references | Lets a reader check the finding |
| Counter-evidence | What the study found that cut the other way | Stops a finding reading stronger than it is |
| Strength | How many participants or respondents, which groups, pattern or single voice | Shows how much weight it can bear |
| Study | Link to the study record and the brief | Gives method, dates and limits |
| Tags | Topic, segment, job, journey stage, method, date | Makes it findable from different questions |
| Decisions informed | What was decided, by whom, when, with a link | Shows use and the cost of revising it |
| Related tests | Links to learning log entries | Connects research with what was tested |
| Access | Who may read it | Keeps sensitive findings narrow |
| Owner and review date | Who checks it and when | Keeps the store current |
| Status | Current, updated or retired, with a note | Old findings stay visible as old |
A language school builds its repository (illustrative)
Ezvantor Language School stops repeating its drop-out research
Ezvantor Language School, an invented school teaching adults, finds that its course team and its marketing team each commissioned interviews on why adult learners drop out, 11 months apart. The two studies interviewed 14 and 12 learners, asked similar questions, and reached overlapping findings, but neither team knew of the other's study until a planning meeting.
The head of marketing becomes the repository owner. A shared database gets one table of study records and one of finding records, with tags for topic, learner segment (the school's 4 segments), job (from a short list of 5 jobs written with the course team), journey stage, method, date and strength. Adding the two drop-out studies produces 19 finding records; 6 appear in both studies and are merged, with the evidence from each, and marked as stronger. Old decks contain learners' first names in quotes, so the owner replaces them with codes.
Recordings from the older study are 2 months past the deletion date set when consent was taken. They are deleted, and the data protection owner adds a check of deletion dates to the owner's review. Each finding gets a review date 12 months ahead. The research brief template gains a first line, "what the repository already says", and the next brief, on why learners choose evening rather than weekend classes, finds 3 relevant records and narrows its questions to what they do not answer.
Common mistakes
- A folder of decks. Findings buried in slides cannot be found from another question; write each as its own record.
- Recordings and names in the repository. Identifiable material belongs in the study's restricted store, deleted on its date, not in a library everyone reads.
- Tags invented per study. Tags only find records if every record uses the same closed lists.
- Findings with no strength. A single vivid voice and a pattern across groups look the same unless the record says which it is.
- No review dates. Old findings are reused as current, or quietly ignored, with no way to tell which.
- Deleting retired findings. A finding that no longer holds should stay visible as retired, so nobody rediscovers it.
- Nobody searches it. A repository that is not the first step of every brief is a filing exercise.
- Copying the learning log. Test results have their own log; link it rather than keeping two versions.
