The "quant or qual?" framing comes up constantly. In practice, neither alone gets you to sufficient customer understanding.
They're complements, not competitors. This article lays out the practical workflow for combining them in 5 phases.
Role separation
| Dimension | Quant (surveys, logs) | Qual (interviews, observation) |
|---|---|---|
| Strength | Scale, objectivity, repeatability | Context, emotion, story |
| Questions answered | "What, how much" | "Why, how" |
| Sample size | Hundreds to tens of thousands | A few to a few dozen |
| Capture time | Short / large-scale | Long per respondent |
| Bias type | Question-design bias | Interviewer bias |
| Analysis | Stats, pivot | Interpretation, storytelling |
"Quant for the whole picture, qual for depth" is the basic pattern. The realistic flow is moving back and forth between them.
5-phase combined workflow
Phase 1: Discovery (quant)
Quant data surfaces "huh?" moments.
Trigger data
- NPS / CSAT trends
- Segment-level numerical differences
- Open-text theme distribution
- Behavioral logs (drop-off points, frequency)
- Revenue / retention changes
Example
Overall NPS averages 32 (up 4 from last period).
But 3rd-year-customer NPS dropped from 45 to 22.
Something is happening in that segment.
A numerical anomaly triggers "why?" — the cue to move to qual.
Phase 2: Deep-dive (qual)
n=5–10 user interviews to probe the "why."
Selection
- 5–10 people from the anomalous segment
- Diverse where possible (attributes, usage intensity, satisfaction)
Question design
1. Recent usage (factual)
2. Current satisfaction (self-rating)
3. Comparison to past (perceived change)
4. Hypothesis floats (probe)
5. Open space (unstructured)
Hypothesis generation
Hypothesis from interviews:
"By year 3, the features that drew them initially have become routine.
New features don't register as compelling.
The customer can't rediscover the product's value."
n=5 hypotheses are still unvalidated hypotheses. Next phase: validate at scale.
Phase 3: Validation (quant)
Design a follow-up survey to confirm the hypothesis at scale.
Validation survey
Q1: Have you tried any new features in the past 6 months?
Q2: What's the main driver of your current satisfaction? (Multiple)
- Original core features
- Recently added features
- Support
- Pricing
- Other
Q3: "The product's value feels lower than when I started" —
1 (not at all) to 5 (strongly)
Direct hypothesis-confirmation questions.
Result
3rd-year segment:
- Tried new features: 23% (overall avg 45%)
- "Value feels lower": 42%
→ Hypothesis confirmed
Validated insight gets promoted from hypothesis to fact.
Phase 4: Action
Design actions on the validated insight.
Example actions
Insight: 3rd-year customers aren't trying new features
→ Action: "New feature tour" campaign for 3rd-year customers
- Individual email introducing new features
- CS-led online workshop
- Adoption bonus
Action design should directly map to insight. When the mapping is fuzzy, effectiveness verification breaks.
Phase 5: Re-measurement (quant)
3 months later, re-run the same survey to verify effect.
Effectiveness check
- NPS change in the target segment
- New-feature usage rate change
- "Value lower" sentiment change
- Open-text tone change
Next action by result
- Worked → roll out to other segments
- Didn't work → revise hypothesis, qual deep-dive again
- Unexpected → check for side effects
One quant×qual cycle complete. Loop.
The hard part — five anti-patterns
Anti-pattern 1: Quant-only conclusions
Moving on the numbers without qual confirmation. Classic correlation/causation confusion.
✗ "NPS dropped → 'pricing' mentions rose → cut price"
○ "NPS dropped → probe pricing complaints → real issue is feature gap → improve features"
Anti-pattern 2: Qual-only conclusions
Strong opinions from n=5 interviews drive major investment without scale validation. Risk of being whipsawed by a vocal few.
✗ "3 heavy users want feature A → 6-month dev investment"
○ "3 heavy users want feature A → survey all customers → 70% want it → invest"
Anti-pattern 3: Quant and qual run in separate silos
Research team does qual; data team does quant; no integration venue exists. Neither side's insight lands.
Counters:
- Make quant×qual pairing a habit
- Reports include both
- Decision meetings present both
Anti-pattern 4: Forcing qual findings into numbers
Stories and emotional discoveries forced into quantification lose their essence.
Counters:
- Keep qual as qual (don't force into numbers)
- Use direct quotes in reports
- Read customer statements aloud in exec meetings
Anti-pattern 5: Locking into "quant-first"
Starting every project by designing a survey. Qual hypothesis-then-quant validation produces deeper findings.
Recommended flow:
1. Spot an anomaly in existing data (quant)
↓
2. Interview deep (qual)
↓
3. Build hypothesis (qual → quant bridge)
↓
4. Validate by survey (quant)
↓
5. Execute action
↓
6. Re-measure (quant)
↓
7. More interviews if needed (qual)
Practical patterns
Pattern 1: Anomaly-trigger
Quant anomaly → qual deep-dive → quant validation → action → quant re-measure
Most standard. Monthly NPS triggers an anomaly, the cycle starts.
Pattern 2: Hypothesis-first
Qual hypothesis (n=5) → quant validation → action → quant re-measure → qual retrospective
For new business / new feature ideation. Qual-triggered.
Pattern 3: Parallel
Quant + qual run simultaneously, reconciled at analysis
Time-constrained efficiency. Disagreements between the two often surface new findings.
Pattern 4: Continuous observation
Quant monthly + qualitative interview once per quarter
Mature-product continuous improvement. The steady-state pattern.
Team composition
In orgs running both, role separation that works:
- Data analyst — quant aggregation / visualization
- UX researcher — qual interview design / execution
- PMM / PdM — integrate both into decisions
- Marketer — design and execute actions
- Exec / department head — make calls informed by both
Roles can overlap, but "someone looking at both" must exist — otherwise integration never happens.
Where Repoan fits
Repoan is designed to seamlessly bridge quant and qual:
- Natural combination of quant questions + open text — AI question generation balances both
- AI theme extraction on open text — structure qualitative data like quantitative
- Numerical change + open-text change together — combined dashboard
- Continuous surveys — track both axes over time
- Pre / post-interview supplemental surveys — qual interview design support
Summary
Combining quant and qual:
- 5 phases (discovery → deep-dive → validation → action → re-measure)
- Quant surfaces anomalies; qual explains; quant validates at scale
- Don't conclude from either side alone
- Avoid the 5 anti-patterns (quant-only, qual-only, siloed, forced quantification, quant-only-first)
- Need someone with org-level responsibility for both
The right frame isn't "quant vs. qual" — it's "quant and qual." Orgs that balance both reach the real customer understanding the AI era requires.