Desk research: check existing evidence before collecting more

Desk research uses information that already exists to answer a defined question before commissioning new collection. It can combine published research, official datasets and internal records. The essential work is evaluating where each item came from, what it measures, when it applies and how far it transfers to your decision.

A search produces documents. A useful desk-research exercise produces a dated evidence record: what is supported, what remains uncertain and which question needs an interview, observation or test. More downloads do not necessarily mean a stronger answer.

Consider a hypothetical community theatre whose manager wants to explain falling repeat attendance. The team has historic audience reports, ticket records and a proposed new loyalty scheme. This guide follows its first investigation. The setting, records and decisions are invented.

Give the search a decision boundary

The theatre’s immediate decision is whether to investigate the booking experience before designing the loyalty scheme. “What drives customer loyalty?” is too broad for that commitment. A more useful starting question is: “What do our existing records tell us about returning bookers, and which explanations of any change can they support?”

Define the period, population and outcome. Returning within a season differs from returning to a particular production. A booker purchasing tickets for several friends differs from several individual audience members. If the decision concerns local attendance, national arts participation figures may provide context without describing this theatre’s customers.

Write the question in a research design beside the intended decision. In the Double Diamond, desk research can inform discovery and problem definition. It does not replace contact with users when the records cannot explain their experience.

Set a practical boundary for the first pass: named sources, a review point and time for synthesis. The theatre might allocate a morning to locating records and an afternoon to examining the most relevant ones. That is a planning example, not a validated duration or a reason to stop if a consequential gap remains.

Build a search route you can repeat

List source families before typing phrases into a search engine. For the theatre, those might include its ticket-system exports, earlier audience studies, official local population data, published arts-participation research and documentation of changes to the booking system.

Use terms for the task and population as well as the proposed solution. Search repeat booking, return attendance, barriers to attendance and booking-channel changes rather than searching only for loyalty-programme success. A solution-centred search can make the favoured intervention appear to be the answer before the problem is understood.

Vom Brocke and colleagues’ 2009 paper on documenting literature searches argues that readers need a comprehensible search account to judge a review’s coverage. It concerns scholarly information-systems reviews. The narrower adaptation here is a search log another colleague can retrace.

Record the database or repository, query, filters, date and why a promising result was retained or excluded. Keep useful synonyms. If the same publication appears through several search services, record one underlying source rather than counting its appearances as independent support.

Use trend scouting for ongoing monitoring of emerging external developments. Desk research has a bounded question; the two activities can share source discipline while producing different outputs.

Find the original behind the headline

Follow a statistic or conclusion back to the report, dataset or study that produced it. A news story might omit the population, a commercial summary may select a convenient result and an infographic may use a different period from its headline.

Open the methods, definitions and relevant results. If only an abstract is available, record that limitation and keep the claim within it. A secondary article saying that a programme increased attendance is not enough to know whether the comparison was credible or whether the setting resembles your theatre.

Cochrane’s handbook chapter on searching and selecting studies distinguishes studies from the several reports that can describe them and emphasises search documentation. Its formal review procedures are more extensive than this business exercise. The transferable discipline is to avoid treating repeated reporting as repeated evidence.

In the fictional theatre, a national report, a trade-magazine article and a consultant presentation could all repeat one audience survey. The evidence record should show the common origin. Three citations to one survey do not become three independent investigations.

Avoid filling an inaccessible source’s gaps with assumptions. Mark the method as unknown, seek an accessible original or leave the question open.

Read the source before borrowing its conclusion

Check what the source was created to do. Operational records support their original task, which may not be your research task. The theatre’s ticket system records purchases and refunds. It may not record who actually attended, which person used each ticket or why someone did not return.

The Office for Statistics Regulation’s administrative-data quality matrix emphasises understanding collection processes, classifications, operational context and possible error. It is official-statistics guidance, not certification for a theatre spreadsheet. It gives useful questions to ask the record owner.

A current number can still contain historic definitions. Suppose the theatre changed ticket platforms halfway through the period. Guest purchases might have become separate customer records, while subscriptions were imported differently. A fall in recorded return bookers could partly reflect the migration rather than an audience change.

Inspect the transformation from raw entries to summary. Were cancelled performances excluded? Are refunds assigned to booking date or performance date? Are complimentary tickets mixed with paid attendance? These are questions to answer before interpreting the trend.

If the team needs a wider environmental view, PESTEL can organise contextual signals. Keep those signals separate from evidence about the specific booking pattern. An economic explanation becomes plausible through a mechanism and relevant data, not because “economic” appears in a table heading.

Use a small appraisal table

For each important source, complete a record like the following. The entries are hypothetical and do not describe actual theatre results.

Source and purpose	Fit to the question	Limitation or next check
Current ticket export, created for sales operations	Can identify purchases under current account rules	Confirm migration and guest-account treatment
Old audience questionnaire, created for a season review	Describes respondents’ stated reasons for attending	Check recruitment, response and wording
Local official population estimate	Context for the catchment area	Residents are not the same population as bookers
Published study of subscription schemes	May suggest a mechanism to investigate	Different venue, offer, period and comparison

Add the source link or file reference, publication date, retrieval date, population, definitions and the exact claim you can support. Distinguish a data-quality problem from limited transferability. An excellent study in another setting can be accurately conducted and still answer a different question.

Use assumption mapping when appraisals reveal important beliefs supported only by repetition. “Audiences no longer want subscriptions” may belong on the question list if the evidence only concerns fewer account renewals after a system change.

Do not reduce appraisal to a colour assigned to the author’s reputation. Authority matters, but it cannot rescue an irrelevant population or an undefined outcome. State why the information deserves weight for this decision.

Compare claims without forcing agreement

Two apparently conflicting figures may have different denominators. One could count people, another households and another transactions. Before combining them, write the definitions side by side. If they cannot be reconciled, retain them as separate evidence.

The theatre might find stable ticket sales but fewer repeat accounts. That does not automatically imply a shift to first-time visitors. A production change, group booking or account migration could produce the same combination. The next useful step is to inspect those pathways, not invent a single story that explains every number.

Look for information that would weaken the team’s preferred explanation. If the proposed loyalty scheme depends on discount sensitivity, what evidence suggests booking difficulty, unsuitable programme timing or an inability to attend? Search for contrary cases deliberately rather than waiting for them to appear.

Keep descriptions apart from causal explanations. A period of lower repeat booking after a price change establishes temporal order if the dates are sound. It does not by itself establish that the price caused the decline.

For strategic choices, opportunity fields can organise the resulting questions into useful areas for investigation. Do that after appraisal, so the fields reflect the evidence rather than the headings of downloaded reports.

Turn gaps into fieldwork questions

GOV.UK’s discovery-research guidance includes reviewing existing evidence and identifying what still needs to be understood about users and their context. Treat desk research as preparation for those unresolved questions.

In the hypothetical theatre, the team might conclude that ticket records can describe booking pathways but cannot explain why some previous bookers left. It can then use customer interviews to explore recent attendance decisions with relevant participants, including people who did not return.

If existing accounts suggest trouble at the box office, observation can examine what happens during an agreed booking or entry task. The desk research should supply a question and a scope, not a prediction the observer is expected to confirm.

A later survey needs meaningful categories and a suitable participant frame. If the desk review shows that “returning customer” is ambiguous, resolve that definition before estimating percentages. More responses cannot repair a poorly defined measure.

Close the search with an accountable handover

Write a short synthesis in three parts: what the evidence supports, what it does not settle and which next decision follows. Link each material statement to its source record. Preserve excluded sources and the reason for exclusion where they could otherwise resurface.

For the theatre, the handover might say that account comparisons across the migration are unreliable, that a narrower post-migration analysis is possible and that attendance decisions need direct investigation. The manager can then choose a record-cleaning task and a bounded interview round before funding a loyalty concept.

Assign owners and review triggers. If a dataset is released annually, note when the next release could matter. If a missing methods appendix arrives, revisit the affected claim. An evidence record should be maintained when its decision context changes, not left permanently labelled “research complete”.

Before the next meeting, ask whether a colleague could reconstruct the reasoning without being guided by the researcher. If they can find the documents but cannot see how the decision followed, improve the synthesis.

Evidence and limitations

A bounded desk search cannot guarantee that all relevant information has been found. Publication, availability, search terminology and organisational record-keeping shape what is visible. A clear search log makes those boundaries examinable; it does not eliminate them.

This workflow is an original practical synthesis of review methodology and institutional guidance. It is not a systematic-review protocol or a substitute for specialist statistical appraisal. The theatre records and proposed outcomes are fictional.

Use the exercise to reduce unnecessary collection and sharpen the research that remains. Its best result may be discovering that an apparently confident number cannot support the decision being asked of it.