THE INTERVIEW METHOD
DEFINING THE INTERVIEW METHOD
The interview method is a way of collecting primary data by asking questions
directly to people. The researcher interacts face-to-face, by phone, or online to
gather information.
TYPES OF INTERVIEW METHOD
STRUCTURED VS. UNSTRUCTURED
• Structured Interviews
Follows a rigorous, pre-defined set of questions. This allows for quantitative aggregation of
qualitative data, making it easier to transform responses into categorical features for
models.
• Unstructured Interviews
Open-ended conversations that allow for deep [Link] are vital in the early
discovery phase to identify entirely new areas of predictive interest or risk.
SEMI-STRUCTURED & FOCUS GROUPS
Semi-Structured: Combines a guide with the flexibility to probe deeper. Ideal for
interviewing individual power-users or high-churn risk clients.
Focus Groups: Involves multiple participants to observe group dynamics and
consensus. Useful for predicting market reception to new product features or pricing
changes.
PRIMARY ADVANTAGES
• Contextual Nuance
Provides the emotional and situational context that raw CRM numbers often lack.
• High Data Quality
Misunderstandings are corrected in real-time, ensuring high response accuracy.
• Predictive Enrichment
Directly identifies "Soft Features" like customer sentiment or intent.
CRITICAL DISADVANTAGES
• Scalability & Bias Issues
• Interviewer Bias: The tone or phrasing of the interviewer can skew responses, leading
to "noisy" data in the model.
• Scalability: Unlike automated data scrapers, interviews are slow and resource-heavy,
making it difficult to achieve large sample sizes.
• Standardization: Converting free-flowing text into numerical inputs (NLP) adds
significant complexity to the pipeline.
EXAMPLE: CHURN PREDICTION
• The Scenario
A SaaS firm noticed their predictive model was missing 40% of churn cases. They initiated
semi-structured interviews with 50 high-profile departing clients.
• The Discovery
Interviews revealed that "Poor onboarding communication" was a primary driver,
which wasn't being tracked as a metric. By adding an "Onboarding Sentiment" feature
derived from interviews, model accuracy jumped to 85%.