An order can arrive on time and still be incomplete. A survey can report happy customers while others quietly leave without responding. Quality measurement must account for both.

Satisfaction describes how people evaluate an experience; service quality also involves observable commitments and verifiable failures. The American Society for Quality (ASQ) describes customer satisfaction information as a basis for improvement [R01]. The UK Government Digital Service recommends collecting feedback at different points in the service journey and comparing surveys with support and other feedback channels [R02]. Its reporting obligations apply to certain UK public services, not automatically to a neighborhood café. This article builds a proportionate measurement approach for small businesses.

1. Separate perception, effort, conformity, and recovery

Four questions require different evidence:

Dimension Question Appropriate evidence
Satisfaction How did the customer rate the experience? Voluntary survey, contextual feedback
Effort How hard was it to complete a task or resolve a problem? CES with a declared scale; observed journey
Conformity and fulfillment Did the customer receive everything agreed, correctly and on time? Order, inspection, delivery and defect records
Recovery Was the harm remedied and the case demonstrably closed? Case record, executed remedy, confirmation and reopenings

A positive rating does not prove compliance with critical requirements; a technically conforming delivery does not guarantee the customer understood the process. To define required quality attributes, see the product and service quality criteria. To handle an individual complaint, use the complaints and service recovery process. This guide measures aggregate patterns rather than repeating those processes.

2. Define CSAT and CES with clear questions, scales, and denominators

CSAT can ask, “How satisfied were you with the order you received?” on a scale from 1 (very dissatisfied) to 5 (very satisfied). Under the top-two-box convention, count responses of 4 or 5 and divide by valid responses: CSAT (%) = responses rated 4–5 / valid responses × 100. Qualtrics describes this convention [R03]. Always state the question and its timing: satisfaction with a purchase, a delivery, and the entire customer relationship are different constructs.

CES assesses perceived ease of an interaction. We might ask, “How easy was it to resolve your issue?” with 1 = very difficult and 7 = very easy. In this teaching example, favorable ease share = responses rated 5–7 / valid responses × 100. You could instead report the 1–7 average, but do not combine the two methods or reverse the scale without saying so. Qualtrics describes CES as an effort measure for completing a task or resolving a case, not as a single mandatory formula [R04]. Choose and preserve a method; retain optional written comments to explain the rating.

Do not automatically request CSAT and CES after every action. A short receipt QR code, an accessible form, or a question after resolving a complaint may be sufficient. Never condition a refund on survey participation.

3. Use NPS only when recommendation intent answers a useful question

NPS, optionally, asks how likely a person is to recommend a business, from 0 to 10. Under Bain's definition [R05], promoters score 9–10, passives 7–8, and detractors 0–6; NPS = % promoters − % detractors. The result ranges from −100 to +100 points, not a percentage of satisfied customers. If 12 of 30 responses are promoters and 9 are detractors, 40% − 30% = +10 points.

This does not replace evidence of complete deliveries, effective resolution, or actual return purchases. Do not compare NPS across differently selected audiences and survey moments as if it were a definitive competition. Observable loyalty is addressed in retention and repeat-purchase measurement.

4. Calculate defects, complaints, and genuine closure from recorded events

First define the unit to be counted. For a café it may be delivered orders. Give each order a transaction identifier and record multiple associated incidents without counting the same order several times in an affected-order rate.

  • On-time deliveries = deliveries within the promise / deliveries with a measurable time promise.
  • Complete orders = orders with no missing items on receipt / orders delivered.
  • Complained-about orders = orders with one or more complaints / orders delivered.
  • Order defect rate = orders with at least one defined defect / inspected or delivered orders, consistently defined.
  • Verified case closure = cases with a completed remedy and confirmed closure / cases opened in the defined cohort, with reporting cutoff and age stated.

A complaint is not necessarily a defect: a defect may be caught internally or never reported; a customer can complain about treatment even when the order is complete. Record severity (informational, service failure, serious risk), recurrence, opened-at time, first response, and actual resolution. An automated acknowledgment is not a completed remedy. Consider medians and ranges when a few extreme cases distort average resolution times.

5. Account for nonrespondents and missing coverage

A voluntary survey is rarely a census. Record invitations, valid responses, response rate, channel, timing, and segment. Pew Research Center explains that nonresponse bias can arise when respondents systematically differ from nonrespondents [R06]. We cannot assume “65% CSAT” means 65% of all customers were satisfied when only one fifth answered.

Compare new and repeat customers where feasible and lawful, counter sales and delivery, peak and quiet days, and transactions with and without incidents. Anonymous purchases do not authorize intrusive identification. A silent customer is not necessarily happy; reduced purchasing does not by itself demonstrate dissatisfaction.

Small samples demand restraint: five good ratings in a day do not establish sustained improvement. Keep the raw counts beside every percentage. State which population was invited when reporting survey results.

6. Account for failure costs without inventing reputational losses

ASQ groups quality costs into prevention, appraisal, internal failures, and external failures [R07]. In services, track rework, refunds, replacements, additional delivery, support time, and warranties separately. Training and preventive checks are costs too, but they are not failure costs.

Count incremental, nonduplicated costs. If a refunded order was also remade for another customer, do not book the same expense twice. We do not assign fabricated monetary values to reputation, future churn, or emotional harm; these may remain identified, unpriced risks. A specific initiative's economics belong in the improvement profitability evaluation.

7. Worked example: strong punctuality, incomplete orders

Entirely hypothetical “Almuerzo Cerca” scenario, separate from the B01–B03 teaching data. During April the café delivered 200 orders, each with a measurable delivery promise; 180 were on time, but 22 were incomplete. It recorded 15 complaints from different orders—some may overlap with incomplete orders—and verified the resolution of 12 by the reporting cutoff. It sent 200 CSAT invitations, received 40 valid responses, and 26 rated 4 or 5. CES produced 32 valid responses, with 20 rated 5–7. A separate optional NPS survey received 30 responses: 12 promoters, 9 passives, and 9 detractors.

Measure Calculation Result What it does not show
On-time delivery 180/200 90% That every item arrived
Complete orders 178/200 89% Overall satisfaction
Orders with complaints 15/200 7.5% That there were only 15 failures
Verified case resolution at cutoff 12/15 80% That the other three are resolved
CSAT among respondents 26/40 65% Satisfaction across all 200 customers
CSAT response rate 40/200 20% Absence of selection bias
Favorable CES ease 20/32 62.5% Ease for everyone
Optional NPS 12/30 − 9/30 +10 points Actual customer retention

In separate records, the incremental cost of remaking eight orders was 8 × S/12 = S/96; five refunds for different cases cost 5 × S/30 = S/150; six additional support hours at S/15/hour cost S/90. Total tracked extra spending was S/336, assuming disjoint cases and excluding support hours from the S/12 remake cost. This is not a complete cost-of-quality estimate and does not claim any future lost margin.

The correct diagnosis is not “90% on time, everything is fine.” Rather: most deadlines were met, but incompleteness remains material, survey participation is partial, and some complaints have no verified closure. First check whether missing items cluster around preparation, packing, or delivery instead of redesigning a survey to hide them.

8. Prioritize severity, recurrence, and exposure over averages

Categorize each problem using safety or compliance risk, number of affected orders, customer impact, recurrence, recorded cost, supporting evidence, owner, and next check. A food-safety or injury risk does not wait for CSAT to drop: it requires immediate containment and applicable controls for the activity and jurisdiction. A generic traffic light is not permission to operate.

If incomplete deliveries recur, start a root-cause investigation; if there is a plausible, safe change, design an operational experiment. Review cadence and visualization belong in the dashboard guide. Here the objective is defining and interpreting quality evidence.

9. Minimum survey and scorecard template

Short post-order survey: “How satisfied were you with this order? (1–5).” “Did you receive all items ordered? (yes/no/cannot tell).” Optional comment: “What should we improve?” After an incident, a separate CES question: “How easy was it to resolve your issue? (1 very difficult–7 very easy).” Do not request a phone number or address in a feedback form without a genuine need.

Scorecard field Definition to document
Reporting period/cohort Delivery dates and complaint cutoff
Population Delivered orders versus valid survey responses
Result Numerator, denominator, percentage or NPS points
Critical quality Severe defects, controls and urgent actions
Cost Rework, refunds and hours without double counting
Investigation Issue, evidence, owner and next review

The companion CSV resources include a scorecard, survey sheet, and issue-priority matrix. They are starter records, not replacements for a POS, accounting system, or privacy controls.

What you should have by the end

A reproducible combination of customer opinion, objective conformity, actual resolution, and observed cost, with honest denominators, disclosed survey limitations, and a clear action for each signal. One favorable metric will no longer count as proof of overall success.

References