2.3 Quantitative Research Methods

If qualitative research discovers mechanisms, quantitative research estimates how widespread those mechanisms are, whether scores are moving, and which segments differ. Domain 1 expects CCXP professionals to design and interpret surveys and related measures with enough statistical judgment to avoid false confidence — without turning the exam into a pure statistics test.

Why Quantitative Methods Matter

Executives fund what they can size. Quantitative evidence supports prioritisation (“this friction affects 28% of onboarders”), governance (“relationship health fell two points in the high-value segment”), and ROI stories later in the metrics domain. Poor quant design, however, creates precise wrong answers: beautiful charts built on biased samples or ambiguous questions.

---

Survey Design Foundations

A survey is a measurement instrument. Design starts with the decision the data must inform, then the construct (what concept you measure), then the items (questions), then sampling and fielding.

Constructs and items

Define constructs in operational language: overall relationship loyalty, interaction satisfaction, effort, goal completion, trust, or recommendation intention. Each item should map to one idea. Avoid double-barreled questions (“How satisfied are you with speed and friendliness?”) and leading wording (“How excellent was our award-winning support?”).

Scales

Common CX scales include:

Use the same scale endpoints and labels over time if you intend to trend results. Changing from 5-point to 7-point mid-programme breaks comparability.

Questionnaire architecture

• Open with easy, relevant items; place sensitive demographics later
• Put the primary metric early enough to reduce drop-off effects on the KPI
• Limit length; every extra question costs completion and attention
• Use display logic so customers only see relevant path questions
• Close with an optional open-end for unexplained residual pain

---

Relationship vs Transactional Surveys

A mature programme uses both. Relationship results set direction; transactional results point to process owners. Do not average every transactional survey into a fake “company NPS” without a defined methodology — the exam rewards method clarity.

---

Sampling, Response Rates, and Bias

Sampling approaches

• Census invitation — invite all eligible customers in a window (common for transactional programmes)
• Probability / random sample — each customer has a known chance of selection (strong for relationship tracking)
• Stratified sample — deliberate representation across segments (value tier, region, product, tenure)
• Convenience sample — whoever is easy to reach (fast but often biased; label limitations)

For CX decisions that reallocate investment across the base, prefer designs that protect segment coverage, not only total completes.

Response rates

Response rate = completes ÷ eligible invitations (define eligibility carefully). Higher is not automatically better if incentives attract only deal-seekers or if only promoters respond. Track nonresponse bias: do responders differ from non-responders on value, tenure, complaint history, or channel?

Tactics that often improve quality without distorting the base:

• Shorter instruments and mobile-friendly design
• Clear purpose and realistic time estimate
• Sensible timing (soon after the interaction; avoid fatigue windows)
• Rotation rules so the same customers are not hammered
• Multi-mode options when one channel excludes a segment (e.g., SMS vs email vs in-app)

---

Practical Statistical Concepts for CX Pros

You do not need to derive formulas on the CCXP, but you must avoid naïve interpretation.

Statistical significance (practical view)

A difference is statistically significant when it is unlikely to be pure sampling noise under a stated model and threshold (often p < 0.05 in business practice). Significance answers: “Is this difference likely real in the data-generating sense?” It does not automatically answer: “Is this difference large enough to care about?”

Effect size and practical significance

A one-point CSAT move on a huge sample can be “significant” yet operationally trivial. Always ask:

• How large is the change relative to historical volatility?
• Does it concentrate in high-value or at-risk segments?
• Would acting on it pass a cost/benefit screen?

Confidence and sample size intuition

Small samples produce wide uncertainty. Segment cuts (especially rare segments) need enough completes before you declare a crisis. When n is thin, treat results as directional and confirm with additional data or qualitative follow-up.

Correlation vs causation

Driver models and cross-tabs show association. “Customers who use chat have higher CSAT” does not prove chat causes higher CSAT — chat users may have simpler issues. Use design of experiments, staged rollouts, or strong quasi-experimental logic when claiming causal impact.

Multiple comparisons caution

Slicing a dashboard fifty ways will produce some “significant” differences by chance. Pre-register key cuts where possible, and require replication or corroborating operational evidence before major investments.

---

Representativeness Across Channels

Customers who answer email surveys are not always like customers who only use the mobile app, visit stores, or call support. Channel representativeness means your measured sample resembles the population about which you want to decide.

Design practices:

• Map who uses which channel in the journey before choosing invitation mode
• Offer accessible modes for older, rural, or low-bandwidth segments when they matter strategically
• Weight results when the sample mix systematically differs from the base (document the method)
• Compare survey-derived reasons for contact with operational reason codes as a sanity check
• Report coverage gaps explicitly (“digital-only pulse; store-only customers underrepresented”)

Scenario: insurance claims measurement

An insurer launches a post-claim CSAT survey by email only. Response rate looks strong among young digital customers, and scores rise after a portal update. Later analysis shows older high-severity claimants — a strategically critical group — rarely respond and still call repeatedly about status. Quant lesson: channel-skewed measurement can celebrate a local digital win while missing relationship risk. Fix by adding SMS/voice options for that segment and stratifying reports by severity and age band.

---

Putting Quant and Qual Together

Best practice is sequential or concurrent mixed methods:

1. Qualitative discovery → item design and hypotheses
2. Quantitative sizing and tracking → prioritisation
3. Qualitative follow-up on surprising segments → mechanism refresh

On the exam, choose the method that matches the decision. Need prevalence and trend? Quantitative. Need meaning and context? Qualitative. Need both for a major investment case? Mixed methods with clear roles for each.

---

Exam Anchors

Expect items that contrast relationship vs transactional surveys, flag bad question wording, test response-rate vs representativeness judgment, and ask whether a “significant” change is automatically actionable. Strong CCXP answers treat quantitative research as disciplined measurement for decisions, not vanity scorekeeping.