Customer experience survey: Which part of the delivery should you fix?
A customer who never contacted support cannot rate its resolution. Follow a recent order through a routed survey and use the right base for each result.
Build a customer experience survey around one recent episode, then ask people only about the parts they experienced. For an online grocer, ordering, delivery, item condition and support belong to the same customer journey, but each question may have a different group of eligible respondents.
ElmCart wants to understand falling repeat orders before choosing between a checkout improvement, a delivery change and more support capacity. This illustrative study follows each customer's most recent order scheduled for delivery between three and fourteen days before the survey. It includes cancelled and missing deliveries as well as successful ones. Limiting the sample to completed deliveries would remove some of the experiences the team most needs to understand.
Give every answer an order to refer to
Draw the sample from the order register and select one index order per customer. Include people who have not ordered again. Show its date and a recognizable order reference, then ask whether the recipient was involved in ordering, receiving or dealing with a problem. The experience of a different household member should not be silently substituted.
Keep the scheduled delivery date, survey date, recorded outcome and any permitted links to support cases with the study ID. The three-to-fourteen-day interval is this study's choice, not a universal rule. Report elapsed time because a support case may still be open when someone answers.
Use existing operational records where they answer the question directly, then collect the customer's account of what happened. AAPOR recommends considering relevant administrative records alongside survey evidence. Its guidance also emphasizes matching the method to the research objective.
Start with involvement, then route through the experience
Insert the grocer's name, order date and reference before fieldwork. Allow a respondent to skip a substantive question. End the survey if Q1 is unanswered, because eligibility has not been established. If Q2 is skipped, go to Q4. If Q4, Q6 or Q7 is skipped, go to Q9. If Q9 is skipped, go to Q13. Other skipped items continue to the next numbered question. Keep missing answers distinct from explicit Don't know or Cannot judge answers.
Question | Exact wording, answers and route |
|---|---|
Q1. Index order | Thinking about your [grocer] order scheduled for [date], reference [ID], were you personally involved in placing it, receiving it or dealing with a problem with it? Yes → Q2. No or Don't remember → end. |
Q2. Ordering involvement | Did you personally place or try to place this order? Yes → Q3. No or Don't remember → Q4. |
Q3. Ordering task | How easy or difficult was it to submit the order? Very difficult. Somewhat difficult. Neither easy nor difficult. Somewhat easy. Very easy. I could not complete the task. Don't remember. All → Q4. Keep inability to complete as a separate outcome. |
Q4. Order outcome | What happened to this order? All of the groceries were delivered → Q5. Some were delivered → Q5. It was cancelled before any groceries arrived → Q9. No groceries have arrived and the promised window has passed → Q9. Don't remember → Q9. |
Q5. Delivery timing | When did the first delivery arrive compared with the latest delivery window you had been promised? Before the window. Within the window. After the window. I had not been given a window. Don't remember. All → Q6. |
Q6. Checking items | Did you personally check any of the groceries that arrived? Yes → Q7. No or Don't remember → Q9. |
Q7. Usable condition | Were all of the groceries you checked in a condition in which you could use them? Yes → Q9. No → Q8. Don't know → Q9. |
Q8. Condition detail | Which groceries could you not use, and what was wrong with them? Open text. Don't remember is valid. Continue to Q9. |
Q7 concerns the items that arrived and were checked. It does not measure missing products. Keep Q4's partial-delivery outcome and the order's missing-item records alongside it. A high condition score can coexist with an incomplete order.
Ask support questions only after a support contact
Support ratings need a contact, a task and a point in time. A customer who never needed help has no resolution experience to rate. Someone awaiting a reply has a different outcome from someone whose case has been closed without a solution. If someone raised several issues, ask them to use the one that mattered most for Q10–Q12.
Question | Exact wording, answers and route |
|---|---|
Q9. Support contact | Did you contact [grocer]'s customer support about this order? Yes → Q10. No or Don't remember → Q13. |
Q10. Contact route | How did you first contact customer support about the issue that mattered most to you? In-app or website chat. Email or web form. Telephone. Another route, please describe. Don't remember. All → Q11. |
Q11. Contact effort | How easy or difficult was it to make that first contact? Very difficult. Somewhat difficult. Neither easy nor difficult. Somewhat easy. Very easy. Cannot judge. All → Q12. |
Q12. Resolution | As of today, what is the status of that issue? Fully resolved. Partly resolved. Not resolved after a response. Still waiting for a response. Don't know. All → Q13. |
Q13. Overall experience | Overall, how satisfied or dissatisfied are you with how [grocer] handled this order? Very dissatisfied. Somewhat dissatisfied. Neither satisfied nor dissatisfied. Somewhat satisfied. Very satisfied. Cannot judge. All → Q14. |
Q14. Improvement | What, if anything, would you change about your experience with this order? Open text. Nothing is a valid answer. End. |
Keep the ordered response scales in their displayed order. Pretest what customers mean by the delivery window, usable condition and resolution before launching the survey. Pew Research Center recommends testing the questionnaire and explains how wording and question sequence affect answers. Source: Pew's survey-design guidance.
The overall question includes people whose order never arrived. It asks about handling that order, so a cancellation remains a legitimate experience. A product-quality question has a narrower base because someone must have received and checked products to answer it.
Put the denominator beside every result
In the worked readout below, ElmCart sends 200 invitations and receives 40 survey returns. Thirty-six people confirm involvement in the index order.
Measure | Worked result and eligible base |
|---|---|
Delivery outcome | 28 report some or all groceries delivered, 5 report cancellation and 3 report non-arrival. Base: all 36 eligible respondents with a Q4 answer. |
Late first delivery | 8 of 26 answering Before, Within or After in Q5 report After. Two of the 28 recipients cannot identify the promised timing. Keep those two visible separately. |
Item-condition problem | 7 of 24 people who checked groceries give a usable No answer in Q7. The other 17 answer Yes. Four recipients did not check the groceries. |
Support contact | 9 of 36 eligible respondents report contacting support. The other 27 answer No. |
Support resolution | Among 8 usable Q12 answers, 3 are fully resolved, 2 partly resolved, 2 unresolved after a response and 1 awaiting a response. One of the 9 contacts answers Don't know. |
Overall satisfaction | 26 of 34 usable satisfaction answers are Somewhat or Very satisfied. Two eligible respondents choose Cannot judge. Return the full five-category distribution as well. |
Dividing the three fully resolved cases by all 36 respondents would mix people who needed help with those who never contacted support. Treating the four customers who did not check items as satisfied would inflate a different result. Each number earns its meaning from its question and base.
These responses describe participating customers. Compare the order outcomes and customer mix of respondents with those invited before generalizing to the whole service. Preserve missing answers, cancelled orders and open cases in the file shared with operations.
Assign the next operational check
ElmCart has seven item-condition reports to investigate. The fulfilment team can inspect the named products, picking and transport records, and any linked photographs already supplied through the service process. Support can review its eight reported case outcomes alongside the time since contact. One response does not establish what happened across every order on that route.
Our customer-journey research guide explains how to connect an event with the customer's account of it. Use that connection to give an owner a bounded task, such as checking one product's handling or clarifying when an unresolved case will next receive an update.
To investigate repeat ordering, follow the relevant customer cohort in permitted transaction records over a defined period and account for whether another grocery order was due. A satisfaction answer alone cannot establish why someone did or did not return.
Once ElmCart can deliver a revised service promise, Sapien can explore simulated responses to alternative delivery explanations or support messages. The operational survey measures the service people received. The modeled comparison helps shape a proposed change before the team tests it with customers. Source: Sapien's research approach.
