User Segmentation and Clustering Analysis Template

Generation Prompt

Paste this brief into WeJot AI to generate an editable first draft of this survey.

Please generate a survey questionnaire for user segmentation and clustering analysis, targeting data analysts, product managers, or marketing strategists, with 8 questions in total. The survey should cover five dimensions: variable standardization (1 question), choice of cluster number (2 questions, including 1 NPS), silhouette coefficient (1 question), segment profiling (2 questions, including 1 open-ended), and label implementation (2 questions). Question types include single-choice, multiple-choice, scale, NPS, and open-ended. Each sample question must be supported by specific cases, such as 'K-Means vs. hierarchical clustering selection', 'silhouette coefficient threshold (0.5-0.7) judgment', and 'confidence in using labels for campaign targeting', with 3-5 options per question. A recommended sample size of no less than 30 responses is suggested, emphasizing internal professional research.

Sample questions

1. 1Single choice

   Before clustering, how do you typically handle the scale differences of numerical features (e.g., spending amount, active days) during variable standardization?

   Z-score standardizationMin-Max normalizationLog transformation followed by standardizationNo standardization; cluster directly
2. 2Single choice

   When using the elbow method to determine the number of clusters K, what is your primary criterion for selecting the elbow point?

   A noticeable attenuation in the decrease of SSESelecting a slightly smaller K for better business interpretabilityRelying on the peak silhouette coefficient associated with KMostly relying on personal experience
3. 3Scale

   When adjusting K based on business complexity (e.g., changing the optimal K=5 to K=4), which situation are you most likely to accept?

   1=very unlikely to accept, 5=very likely to accept; integers only from 1 to 5
4. 4Single choice

   If the silhouette coefficient of a clustering solution is 0.45, how would you respond?

   Adopt it directly, as values above 0.3 are acceptableGo back and adjust features or algorithmIncrease the number of clusters and retestSwitch to density-based clustering and compare
5. 5Multiple choice

   For the generated clusters, which of the following methods do you typically use to validate the distinctiveness of their profiles? (Select all that apply)

   Compare the mean differences of key metrics across clustersVisualize dimensionality reduction (e.g., t-SNE) to observe separationInvite business colleagues to conduct blind labelingFocus only on statistical significance, not human judgment
6. 6Open text

   If a cluster is statistically significant but difficult to describe in business terms, what do you think is the most likely reason?

7. 7Single choice

   Before converting clustering results into business labels (e.g., 'high-value but at risk of churn') for precision marketing, which of the following do you consider most critical?

   Establishing a mechanism for label monitoring and updatesVerifying the actual impact of labels on conversion ratesEnsuring smooth integration with systems like CRMMapping the full customer journey to align with labels
8. 8NPS

   How satisfied are you with the current level of automation in your team's segmentation label implementation? (Rate on a scale of 0-10)

   Integers from 0 to 10, where 0=extremely dissatisfied and 10=extremely satisfied