What this template is for
Write an analysis plan for every survey that will inform a decision: a customer satisfaction or lapsed-customer survey, a pricing or proposition survey, a brand or awareness wave, a post-purchase survey. A single on-page poll read for themes rather than proportions does not need one. Put simply, an analysis planA survey's plan, written before the questionnaire: the decision, the figures and comparisons the report needs, which question feeds each, and small-group rules. is the list of outputs a survey must produce, fixed before its questions are written.
The plan sits between documents rather than replacing them. Before it, the research brief commissions the study, and the guide to research design and the brief explains naming the decision and choosing the method. After it come the tutorials that each tell the reader to have the analysis plan open: the survey and questionnaire design guide says to write the plan first and cut any question that does not feed it; the pretesting tutorial begins by checking every question against it; the tutorial on drawing a survey sample from a customer list asks for the decision, the figures and the comparisons before the draw; and the tutorial on coding open-text answers cuts its counts by the groups the plan names. The survey launch checklist then checks that every question maps to the plan before the survey goes live. At the end, the guide to reporting research findings shapes the readout.
When an agency runs the survey, the guide to commissioning a research agency asks what the agency's analysis plan will contain; this template is a way to answer that question or to check the agency's answer. It is not a market sizing exercise: an estimate of a market's size from several sources belongs in the market sizing worksheet, which may use a survey's results as one input.
The template
Copy the table into the study's working document and keep the field names. Fill the first two rows before anything else; if the outputs cannot be written as sentences, the survey is not ready to design.
| Field | What to write |
|---|---|
| Decision and population | The decision the survey informs and the group of people that decision concerns, copied from the research brief |
| Outputs needed | Each figure, comparison or chart the decision needs, written as the sentence it will become in the report |
| Question-to-output map | Every question listed against the output it feeds, so questions that feed nothing are cut before fieldwork |
| Groups and comparisons | The groups results will be cut by, the fields or questions that identify them and the comparisons that matter |
| Sample and weighting | Who is invited, how they are drawn, the strata, and whether and how responses will be weighted back to the population |
| Open-text coding | Which open questions are coded, who builds the code frame, and the checks on coding agreement |
| Small groups and uncertainty | How results from small groups are shown or held back, and how uncertainty is described beside each figure |
| Reporting format | The structure of the readout: what was found, what it means for the decision and what to do next |
| Data handling | How answers are kept apart from identities, who can see the raw data and when it is deleted |
Keep the plan short enough to have open beside the questionnaire. The full code frame, the draw record and the participant information can sit in attachments, linked from the field they belong to.
How to fill each section
Decision and population
Copy the decision and the population from the research brief word for word. The research design guide explains why research starts with a decision and how to name the population precisely: active customers, recent buyers, lapsed customers or people in the category who have never bought. If the plan cannot copy them because the brief is vague, fix the brief first.
Outputs needed
Write each output as the sentence it will become, with a gap for the figure: "The share of lapsed customers who name price as a reason for not returning is …". The survey design guide describes the analysis as the figures for the decision, the comparisons between segments and the trends against earlier waves; list each one you need. Mark which outputs would change the decision and which are context, so the context can be cut first when the questionnaire runs long.
Question-to-output map
List every draft question with the output it feeds. The pretesting tutorial starts with exactly this check: cut any question that feeds nothing, then confirm that each answer format can produce its output. A comparison between groups needs the grouping question; a trend needs the earlier wave's wording unchanged; a ranking needs options people can genuinely compare. Add screening and routing questions to the map too, marked as such.
Groups and comparisons
Name the groups the results will be cut by and where each comes from: a field already on the customer record, or a question in the survey. The sampling tutorial recommends grouping on fields the records already hold where possible, drawing on the SegmentationDividing a market or customer list into groups that need different offers or messages. the business already uses. Then write the comparisons that matter for the decision, not every possible cut. A long list of cuts invites fishing for a difference and produces subgroups too small to read.
Sample and weighting
Record the population, the frame that reaches it and who that frame misses, then the strata and how many will be invited in each. The sampling tutorial walks through cleaning the list, applying contact rules on a research footing, stratifying, drawing at random with a reserve and recording the draw. Decide the Sample SizeThe number of users or sessions an experiment needs to detect a given effect reliably. from the smallest group the report must read, as the survey design guide advises, not from the total. Write down now whether results will be weighted and on which few variables, so the choice is not made after seeing which version looks better; the guides advise reporting weighted and unweighted figures together when they differ.
Open-text coding
List each open question, who will build the code frame and who will act as the second coder. The coding tutorial builds the frame from a sample of answers, gives each code a definition, an exclusion and an example, has a second person code a portion and settle the differences, and counts respondents rather than mentions against a stated base. Note which outputs each coded question feeds; an open question that feeds no output is a candidate to cut.
Small groups and uncertainty
Decide in advance how a small group's result will appear. The guide to reading small numbers advises showing the count behind every rate and putting a range, such as a Confidence IntervalA range of values for an effect, such as a lift in conversion rate, that is consistent with the data at a chosen confidence level., around a figure rather than quoting it alone. Write the rule the report will follow: below what count a group is shown as a count rather than a share, and which groups will be held back or merged. The survey design guide adds that differences between groups are shown only where the sample supports them and that the report says when a subgroup is too small to read. Once fieldwork closes, the tutorial on cleaning and tabulating survey responses puts this rule to work on the export, beside completion and quality rules written down in advance, and logs every exclusion.
Reporting format
Fix the readout's shape before the data arrives. The guide to reporting research findings puts the answer to the decision first, then findings with their evidence and how strong each is, then what the study cannot say, then recommendations with owners. The what, so what, now what framework gives a compact structure for each finding: what was found, what it means for the decision and what will change. Note the bases and the exact question wording that will sit beside each figure.
Data handling
Surveys collect Personal DataAny information relating to an identified or identifiable living person, such as a name, email address, location data or an online identifier. whenever answers can be linked to a person, which includes a tracked invitation link. The survey design guide records that the professional codes published by the MRS and ESOMAR expect participation to be voluntary and informed, respondents to be told who is asking and why, and research never to be used as cover for selling. The tutorial on participant information and consent gives every participant a code, keeps the key that links codes to names restricted, collects only what the study needs and sets a deletion date agreed with whoever owns data protection. Record those choices here, and say whether answers are anonymous, confidential or attributable. This is general direction, not legal advice.
Filled example
Below, the plan is completed for an invented business. The figures, dates and results are illustrative.
Celvidrun Cycle Hire: a survey of customers who never came back
Celvidrun Cycle Hire is an invented cycle hire business with an online booking system. Its records showed that most customers hired once and never returned. The marketing lead wrote this plan in March, before any question was drafted, for a survey of one-time customers from the previous two seasons.
| Field | Celvidrun's plan |
|---|---|
| Decision and population | Whether to change prices or rebuild the booking journey first this season. Population: customers who made exactly one booking in the last two seasons and have not booked since, about 4,800 people on the customer list |
| Outputs needed | O1: the share who name price as a main reason for not returning. O2: the share who name the booking journey. O3: O1 and O2 compared between customers whose first hire was in summer and in other seasons. O4: O1 and O2 compared between solo hirers and groups. O5: the share who would consider hiring again, by group size. O6 (context): what they did instead |
| Question-to-output map | Q1 screening (confirms they hired once); Q2 main reasons for not returning, closed list with "other" → O1, O2; Q3 open: "What, if anything, would bring you back?" → O1, O2 via coding; Q4 likelihood to hire again → O5; Q5 what they did instead → O6. Draft questions on helmet comfort and café interest fed no output and were cut |
| Groups and comparisons | First-hire season and group size, both taken from the booking record, not asked. Comparisons that matter: O1 and O2 by season and by group size. No other cuts planned |
| Sample and weighting | Frame: the booking system export, with staff, test bookings and anyone who opted out removed. Strata: season of first hire × solo or group (four strata). Invite 600 per stratum, all of a stratum where fewer exist; a reserve of 200 per stratum held back. Weight to the frame's shares of the four strata only, and report weighted and unweighted figures together if they differ |
| Open-text coding | Q3 coded by the marketing lead from a first read of 60 answers; the operations manager codes a random 50 independently and differences are settled by sharpening definitions. Counts are respondents per code against the base of those who answered Q3 |
| Small groups and uncertainty | Any group with fewer than 50 answers shown as counts, not percentages. Each share given with its range. Groups with fewer than 20 answers merged with the nearest group or held back, and the report says which |
| Reporting format | One page: the answer to the decision first, then O1 to O5 with bases and question wording, then what the survey cannot say (it misses customers with no email on file), then a recommendation with an owner. Each finding written as what, so what, now what |
| Data handling | Invitations sent from the booking system with a statement that the survey is research, nothing will be sold on the back of it, and answers are reported only in groups. Respondent IDs replace emails in the analysis file; the key is held by the operations manager. Raw answers deleted at the end of the season, as agreed with the business's data protection owner |
The survey closed with 640 answers. Booking journey problems were named by 41% and price by 23%, with the gap widest among groups, who described the multi-bike booking form as confusing. Winter first-hirers in groups gave only 17 answers, so they were merged with spring as the plan required and shown as counts. The lead recommended rebuilding the group booking flow before reviewing prices.
Common mistakes
- Writing questions before the plan. A questionnaire drafted first collects answers nobody planned to use and misses the question that identifies a group the decision compares.
- Outputs as topics, not sentences. "Attitudes to price" cannot be checked against a question. "The share who name price as a main reason" can.
- Every possible cut. A long list of comparisons invites fishing for a difference and produces groups too small to read. Name the comparisons the decision needs.
- Asking what the records already hold. Grouping fields already on the customer record should be attached, not asked again, which shortens the survey and avoids misremembered answers.
- Deciding on weighting after seeing the results. The choice of whether and how to weight belongs in the plan, before anyone knows which version supports the preferred answer.
- No owner for open answers. Open questions without a named coder and second coder are often summarised by whichever quotes stand out, which is not the same as counting them.
- Percentages for a handful of people. Without a rule written in advance, small groups end up quoted as shares. Decide the rule in the plan and follow it in the report.
- Data handling left to the end. How answers are kept apart from identities and when they are deleted has to be settled before the first invitation, not after the export.
The research brief that comes before this plan, and the other research and planning documents, are in the templates library.
