NPS surveys: how to automate response collection and CRM actions

A store sends customers a survey, but the responses stay in a spreadsheet and no one takes ownership of complaints. It makes sense to automate NPS surveys together with the action that follows: tie the invitation to the right event, save the score in CRM and assign the issue to a person. Otherwise, the business gets a number that does not explain what to change in its service.

First, agree on whom you are surveying, which experience you are asking about and when, as well as who responds to comments. Without consistent rules, results from different periods may not be comparable.

What NPS measures and how to calculate the result

Net Promoter Score measures willingness to recommend a company. The basic question is: “How likely are you to recommend our store to a friend or colleague?” The respondent chooses a score from 0 to 10. Under Bain’s calculation methodology, scores of 9–10 indicate promoters, 7–8 indicate passives and 0–6 indicate detractors. These are response categories, not a definitive diagnosis of a customer’s behaviour.

The formula is the percentage of promoters minus the percentage of detractors among all valid responses. Passives remain in the denominator. The result ranges from −100 to +100; it should not be presented as the percentage of satisfied customers. The average rating on the scale is not NPS either.

A hypothetical example for an online store

Suppose you invited 1 000 customers and received 200 valid responses: 110 promoters, 50 passives and 40 detractors. Promoters account for 55% and detractors for 20%. NPS is therefore +35. The response rate among those invited is 20%: this is a different metric, not part of the NPS formula.

Do not take a simple average of the scores for groups of different sizes. Recalculate overall NPS from the numbers of promoters, detractors and all respondents. Keep the underlying counts for verification.

Define which relationship you are asking about

Regular measurement of attitudes towards a brand requires a consistent survey wave of customers who already have experience with it. A separate event-triggered survey can be used to assess a particular order. But these streams should not be combined in one chart without explanation: changing the timing of the invitation changes the context of the response.

A survey immediately after payment is premature if the product has not yet been used. Set a practical interval based on the product and objective, test it in a pilot and document it. There is no single right day for every store.

Do not hide incomplete or problematic orders just to improve the number. If your methodology excludes them from a particular wave, show the reason and the size of the excluded group. A separate stream for cancellations or returns will help avoid losing important signals.

Bring events, invitations and responses into one process

Survey automation starts with the event source. The store, delivery service and CRM may interpret the status “completed” differently. Agree on which system confirms receipt of the product, what partial delivery means and how a status change after a return is handled. A manager’s checkbox without a clear rule is a weak trigger.

A practical flow is: delivery event → invitation eligibility check → scheduled sending → response capture → CRM record → task for the person responsible. Duplicate control is needed between the event and the email. A technical signal received again must not create another survey for the same experience.

Which fields to agree on before development

For the invitation, store the customer ID, order or wave ID, survey type, event time, scheduled and actual sending times, channel and message delivery status. For the response, store the invitation ID, rating, optional comment, date and survey version. Contact details do not need to be duplicated everywhere: a controlled link to CRM is enough.

Define the rule for repeat responses: whether the first or last counts and how the history is retained. In B2B, choose the unit of measurement—contact or company: several employees of one customer company should not silently change its weighting.

An off-the-shelf service may cover the survey and basic workflows. HubSpot documentation, for example, describes NPS responses in contact properties and the triggering of related workflows. Check the features and subscription tier: the delivery and duplicate-handling logic you need may require custom development.

Assign the comment to someone, not just to a dashboard

CRM integration is useful when a response becomes a support case that can be managed. A low score may create a task for support, but it does not automatically explain the reason. Ask: “What had the greatest influence on your score?” Keep the comment optional so you can receive a rating without forcing people to write a long text.

The task should contain links to the response and order, an owner, an agreed deadline for the first response and a resolution status. If the contact already has an open case, it is better to add context to it than create parallel tickets. If no owner is available, a fallback queue is needed rather than a process that ends silently.

Do not ask customers to change their score in exchange for help. Resolving the issue is a separate process; the rating should remain an honest signal. For more on dialogue, see the article on quality customer interactions.

Do not call a survey anonymous if you link responses to a contact. Do not post a comment on the website without permission: an internal rating and a public review are different things. The latter is covered in the article on managing reviews and trust in your brand.

How to read the score without fooling yourself

The +35 in the hypothetical example describes responses from 200 people, not the proven attitude of the entire customer base. Those who responded may differ from those who stayed silent. Check coverage of new and repeat customers, regions, sales channels and delivery types. Unreachable contacts and unsubscribes also shape the limits of the sample.

A high response rate alone does not guarantee the absence of bias. Equally, a low rate does not prove that the entire result is wrong. Compare the composition of invitees and respondents using the characteristics available. If a channel change means that mainly active account users now respond, a rise in the score may be explained by the group’s composition.

Assess the uncertainty in the difference between survey waves. In our hypothetical dataset, moving one response from detractor to promoter changes NPS by one point. In small segments, the impact is even greater. The number of completed surveys needed depends on precision and sample structure, not a universal minimum.

NPS does not prove a causal effect on revenue. Compare it with repeat orders, returns and support cases for the same groups and periods. Simultaneous growth in sales and the score does not imply causation: a separate evaluation design is needed to test an improvement.

What to test before sending at scale

The pilot should test the entire journey from event to task and report, including errors and delays.

1. Create test events: standard and partial delivery, a duplicate signal, cancellation and return. Check who receives an invitation and when.
2. Test unsubscribing, frequency limits and a repeat permission check immediately before sending. A scheduled email must not ignore a newly recorded restriction.
3. Test all boundary scores: 0, 6, 7, 8, 9 and 10. Do not count a blank response as zero; reject values outside the scale.
4. Test reopening the link, duplicate responses and delayed data exchange with CRM. Once the connection is restored, the history must not be duplicated.
5. Verify the formula using a control dataset, along with the time zone and period boundary. Compare raw responses, the service’s report and your own dashboard.
6. Assign a test complaint to its owner and take it through to closure. Check the fallback queue, access permissions and failure notification.

The acceptance criterion is: one eligible event creates one invitation, the response is counted according to the agreed rule and the issue has an owner. Expand coverage only after checking the data for a limited group.

Timelines, costs and the limits of automation

The plan consists of agreeing on the methodology, checking events and contacts, building a survey prototype, integration, testing and a pilot. The duration depends on the availability of delivery data, CRM capabilities, the number of channels and the speed of approvals. Until at least one cycle of response collection and processing is complete, a technically ready form is not yet a tested process.

In the estimate, calculate analysis and design, the service licence and user seats, message sending, integration development, history storage, reports and monitoring separately. Add testing, support and employees’ time spent responding to customers. Compare not only launch costs but also the costs of maintaining rules and changing integrations.

The main risks are survey fatigue, incorrect invitations, leaked comments and a backlog of unhandled complaints. Restrict access by role, agree on a retention period and define the data deletion procedure. Do not send open-ended comments to third-party analytics services without reviewing their terms. The person responsible should determine the legal grounds for processing and sending, taking your jurisdiction into account.

If there are no resources to act on responses, establish the process manually with a small group. Automation cannot replace a decision about compensation or fixing a delivery issue.

A useful outcome is not the highest score, but a repeatable measurement and a closed loop from comment to action. Prepare your order statuses, invitation rules and a sample report. Take these to GL.ua (Glyanets) to discuss integrating your online store with CRM and estimate a pilot for automated feedback.