Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Standard Operating Procedures

Customer Survey Wave SOP

A standard procedure for each wave of a repeated customer survey: the repeated draw, a fixed questionnaire, the launch gate, follow-ups, coding and the chart.
SOPintermediate10 min readUpdated

Owner, trigger and scope

The survey owner runs each wave, keeps its record and signs it off. The analyst draws the sample, codes the open answers, compares respondents with the frame and keeps the control chart. The customer service lead makes the follow-ups the written rule calls for and records them. The data protection owner signs off the consent, lawful basis and retention lines before the first invitation.

The trigger is a wave of a repeated customer survey falling due on the date its plan sets, such as a relationship survey of the customer base or a regular satisfaction survey of recent buyers. The voice of the customer guide sets what each survey measures, the contact rules and the inner and outer loops; this procedure is the per-wave run those pages point at. Of those, the contact rule is the one this procedure applies directly, when the sample is drawn.

Neighbouring jobs belong to other pages:

SituationWhere it is handled
Deciding whether a survey wave is needed at allCustomer research programme playbook
A survey sent after each purchase, delivery or support contactThe voice of the customer guide's transactional surveys
A tracker of category buyers rather than the business's own customersSetting up a brand tracker
A relationship score falling across several wavesRelationship survey score falling
A complaint raised in an answerService recovery and complaint handling
A request in an answer to stop marketingMarketing opt-out request SOP

The SOP library routes the other recurring tasks.

Before you start: inputs and access

InputOwnerNotes
The survey plan: the decision, the population sentence and the wave datesSurvey ownerThe survey analysis plan template holds the outputs the decision needs
The record of the last drawAnalystFields exported, cleaning and contact rules, strata, invitations and the reserve
The frozen questionnaire with its version, and the last wave's codeframeSurvey owner, analystCompared with what is in the survey tool before anything is sent
The written follow-up ruleCustomer service leadWhich answers lead to contact, who makes it and how it is recorded
The control chartAnalystBaseline, limits, run rules and annotations
The lawful basis, retention and deletion dateData protection ownerDecided by that owner, never by the survey owner

The score belongs on the retention outcome's KPI tree, as the voice of the customer guide asks: a score with no place on the tree measures nothing anyone in the business is responsible for.

The procedure

  1. Confirm the wave is due and note what changed since the last one

    The customer research programme playbook is where the need for a survey wave is decided; this step confirms that the plan's date has arrived and the decision it informs still stands. List what changed since the last wave: price moves, incidents, launches and any change to the survey itself. The voice of the customer guide keeps relationship waves clear of price rises and outages unless measuring that reaction is the point, so a wave whose date lands in one is moved to a neutral point and the move is annotated.

  2. Draw the sample by repeating the recorded draw and the contact rule

    Follow the record of the last draw step by step, as the tutorial on drawing a survey sample from a customer list writes it: the same population sentence, an export holding only the fields the draw needs with its deletion date noted, the same cleaning rules, the contact rule applied across every team that sends surveys, the same strata and a reserve held back. Count what each rule removes. A wave drawn another way is a different measurement, which the tutorial says to annotate rather than compare silently, and the sample size for each stratum starts from the business's own past waves.

  3. Keep the questionnaire fixed, or start a new series with its own baseline

    Compare the questionnaire in the survey tool with the frozen version, question by question, including the invitation. The survey and questionnaire design guide is where wording, scales and question order are decided, and the relationship score diagnosis treats any change to them as the start of a new series that needs a baseline of its own. An edit made since the last wave is reversed, or kept as a deliberate new series with its reason, its date and a pretest through the tutorial on pretesting a questionnaire, which also catches a leading question. Nielsen Norman Group's guidance, cited by the voice of the customer guide, is that every extra question costs completion.

  4. Pass the survey launch checklist before the first invitation

    The survey launch checklist is run again ahead of every wave of a repeated survey, by the survey owner with the analyst and, on the consent and data lines, the data protection owner. It checks that the invitation stays on a research footing with nothing promoted, names the sender, the purpose, the time it takes and the use of the answers, and that the lawful basis and retention are agreed. The voice of the customer guide gives the general direction behind those lines: ESOMAR's codes ask researchers to be honest with respondents, keep taking part voluntary and handle personal data with care; for customers in the European Union, the GDPR sets the rules for collecting, using and keeping answers linked to identifiable customers; and answers used to contact a person or linked to their account count as personal data, so customers are told how they will be used. That is general direction, not legal advice. The tutorial on participant information and consent writes the wording.

  5. Send the invitations and read the completion funnel during fieldwork

    Send on the planned date. When the invitation goes by email, take the test, approval and collision habits of the email campaign send SOP while keeping the message research rather than marketing. Then read the completion funnel the diagnosis for a low survey response rate describes: invited, arrived, started and finished, read on the first morning while a fix is still cheap. Where the fall sits names the cause. Release reserve invitations only in a stratum that is answering thinly, after the funnel has been read, and record when and how many. A question changed mid-wave splits the answers into separate sets, so it is a last resort.

  6. Follow up the answers the written rule says need a person

    The voice of the customer guide calls this the inner loop: a person contacts each customer whose answer the written rule names, acknowledges what they said and fixes it where that is possible. Log every follow-up so a missed one shows. Contact only customers who said they were willing to be contacted, and never trace an answer from an anonymous survey. A complaint enters the complaints process in the service recovery guide, and an answer asking for no more marketing goes to the opt-out SOP.

  7. Compare respondents with the frame and code the open answers

    When fieldwork closes, the analyst sets the share of the frame beside the share of respondents for each group, as the sampling tutorial's last step does, weights on a few variables where they differ and names any group that answered too thinly to read. Those comparisons run on the export only after it has been cleaned by the completion and quality rules written for this survey, as the tutorial on cleaning and tabulating survey responses describes, with what each rule removed logged. If respondents look unlike the people invited, the diagnosis for survey results that are not representative separates the causes. Then delete the extract and record the survey date against each respondent in the contact history. Code the open answers with the tutorial on coding open-text answers: personal details removed first, last wave's codeframe applied, new codes added with a date rather than old ones redefined, a second person coding a share of the answers, and each code counted by respondent against the base.

  8. Plot the score on the control chart and apply the agreed rules

    The relationship score diagnosis asks for every wave to be read on a control chart carrying its baseline, limits, run rules and notes, and for the agreed rules to settle whether a wave counts as a signal before anyone explains it. A wave inside the limits with no rule fired is recorded as no change shown. If a rule fires, check in a set order: the data and the sample, then the business's calendar of price changes, incidents and launches, and the open text and records last. Once a year, or when the product or the survey changes, the tutorial on testing whether survey scores predict churn joins a wave to what customers did next, which shows whether the score works as a leading indicator, warning of customers leaving.

  9. Take the themes to their owners and store the wave

    The outer loop sends the themes most common among low scorers to the people who own those parts of the experience, with actions, owners and dates that the next review checks first. Set the themes beside the contact reasons from tagging support contacts by reason, and write the readout as the guide to reporting research findings advises, leading with the answer to the decision. Store the wave as the research repository tutorial sets out: findings and de-identified excerpts in the repository, identifiable material in the study's restricted store until the deletion date the data protection owner set.

A boiler servicing company runs one wave of its customer survey (illustrative)

Illustrative example — not a real result

One wave: a restored questionnaire, a repeated draw and a theme sent to scheduling

Kweltrop Heating, an invented boiler servicing company, surveys its service-plan customers each quarter. Before the November wave, the survey owner found a question about a new thermostat added in the survey tool since August; the frozen version 3 was restored and the thermostat question moved to a separate study. The analyst repeated the recorded draw: 8,840 customers matched the population sentence, cleaning removed 120 duplicate, staff and test records, and the 90-day contact rule removed 640 customers surveyed after a recent repair, leaving 8,080. In proportion to three tenure strata, 1,500 were invited and 300 held in reserve, none of it released. 420 customers answered: 188 scored 9 or 10 and 58 scored 6 or below, a relationship score of +31. The chart's centre line, drawn from 12 earlier waves, was +33 with limits at +24 and +42, and no run rule fired, so the review recorded no change shown and nobody launched a fix for noise. Of the 58 low scorers, 23 had ticked that they wanted a call; the customer service lead called all 23 and recorded each, and 4 complaints went into the complaints process. Customers in their first plan year were 33% of the frame but 21% of respondents, 88 of 420, so the readout reported them separately and named them as read only roughly. Coding the 310 comments with codeframe version 4 found missed appointment windows the most common theme among low scorers, mentioned by 19 of the 58. The theme went to the scheduling team lead, who agreed to compare engineers' arrival times with booked windows and report at the February wave review.

QA checks

The survey owner confirms each check before the wave is signed off. A failed check goes back to the step that owns it, and the wave record says what was found.

  • The sample followed the recorded draw, and nobody inside the contact rule's interval was invited.
  • The questionnaire matches the frozen version, or a change started a new series with its own baseline.
  • Every answer that met the follow-up rule has a recorded follow-up.
  • The score was read on the control chart against the agreed rules before anyone explained it.
  • Respondents were compared with the frame and the readout names any group that can be read only roughly.
  • The draw, the questionnaire version, the coded answers and the agreed actions are stored where the next wave will look.

Hand-offs and records

The owners of each theme receive it with the evidence, an action, a date and the next review that will check it. The customer service lead hands back the follow-up log, and complaints continue in the complaints process. The analyst passes the chart and the themes to the monthly reporting pack for the month the wave closed, so the score is reported with its limits rather than as a single number. The data protection owner receives confirmation that the frame extract was deleted. When retention itself is falling, the diagnosis for a rising customer churn rate reads the behaviour behind the churn rate.

The lasting record is a wave record in the research repository: the population sentence, the draw with every rule and what it removed, the questionnaire version, invitations and responses by stratum, the respondent comparison, the codeframe version, the chart with any rule that fired, the follow-ups made and the actions owners agreed, with the frame extract deleted on its date.

Common mistakes

  • Drawing each wave from scratch. A different draw is a different measurement, and the chart then compares unlike with unlike.
  • Editing the questionnaire between waves. A small wording change breaks the series; start a new one on purpose or leave it alone.
  • Explaining the number before reading the chart. Most movements sit inside the limits and need no story.
  • Calling customers who never agreed to contact. Follow-ups are for those who said yes; anonymous answers stay anonymous.
  • Recoding the comments with a new frame. Themes can no longer be compared with the last wave.
  • Agreeing actions nobody checks. The next review starts with last wave's actions, or the outer loop never closes.

Frequently asked questions

How often should a wave of our customer survey run?

No interval is set here, because the right one depends on the decision the survey informs and on how often customers are already asked. The survey plan sets the wave dates; the contact rule, agreed across every team that sends surveys, stops the same customers being asked again too soon; and a wave that would land just after a price rise or during an outage is moved unless that reaction is what the business wants to measure. If customers answer less with each wave, read the completion funnel and the contact history before adding waves.

Can we improve a question between waves?

Only by accepting that the series starts again. A change to the wording, the scale, the order or the invitation creates a new measurement that cannot be compared with the old one, so the control chart needs a new baseline from the wave the change took effect in. Pretest the new version, record the change with its reason and the wave it took effect in, and if a question must be replaced, run the old and new versions side by side for a wave so the break can be read. A wording fault explains a score that was always odd, not one that recently moved.

Should every low score get a call?

Only the answers the written follow-up rule names, and only from customers who agreed to be contacted about their feedback. An anonymous survey stays anonymous, however worrying a comment is. The rule says which answers lead to contact, who makes it and how it is recorded, so the team can see afterwards whether every follow-up happened. A complaint found in an answer goes into the complaints process, and a request to stop being contacted goes to whoever handles opt-outs.

Terms used on this page

Frameworks

Where this page is used

Sources

  1. ESOMAR — ESOMAR
  2. Nielsen Norman Group articles — Nielsen Norman Group
  3. Regulation (EU) 2016/679 (GDPR) — EUR-Lex

Related