What Is Key Driver Analysis?
Key driver analysis (KDA) is a statistical technique that identifies which attributes, features, or experiences have the greatest impact on an overall outcome, typically customer satisfaction, loyalty, likelihood to recommend, or purchase intent. Instead of asking customers what's important to them (stated importance), KDA calculates importance from the data itself (derived importance) by measuring how strongly each attribute correlates with or predicts the outcome variable. This distinction matters because what people say drives their satisfaction and what actually drives it are often different things. KDA reveals the true drivers, not just the ones respondents think matter most.
Why Key Driver Analysis Matters
Resources are finite. You can't improve everything simultaneously, so you need to know which improvements will move the needle most. KDA answers that question empirically. It replaces intuition-based prioritization ("we think service quality matters most") with evidence-based prioritization ("a one-point improvement in service quality produces the largest increase in overall satisfaction"). It's the bridge between measuring performance and deciding where to invest.
How Key Driver Analysis Works
The Core Logic
KDA examines the statistical relationship between a set of predictor variables (attribute ratings) and a single outcome variable (overall satisfaction, NPS, purchase intent). Attributes that have a strong positive relationship with the outcome are "key drivers", when they improve, the outcome improves. Attributes with weak relationships are less influential, regardless of how well or poorly you perform on them.
Common Methods
Correlation-based analysis is the simplest approach. Calculate the Pearson correlation between each attribute and the outcome variable. Higher correlations indicate stronger drivers. This is fast and intuitive but doesn't account for relationships between the attributes themselves.
Multiple regression is the most common KDA method. It regresses the outcome variable on all attributes simultaneously, producing standardized coefficients (beta weights) that reflect each attribute's unique contribution while controlling for the others. This handles correlated attributes better than simple correlation but can produce unstable results when attributes are highly collinear.
Relative importance analysis (also called relative weights or dominance analysis) addresses the collinearity problem directly. It decomposes the total R-squared across all predictors, assigning each attribute its proportional contribution to the explained variance. This produces more stable driver importance scores than raw regression coefficients, especially when attributes are correlated, which they almost always are in satisfaction research.
Shapley value regression uses game-theory concepts to calculate each attribute's marginal contribution across all possible combinations of predictors. It's computationally intensive but produces the most defensible decomposition of explained variance.
The Priority Matrix
KDA findings are most actionable when displayed on a priority matrix (also called an importance-performance grid). The horizontal axis shows derived importance (from the KDA). The vertical axis shows current performance (mean rating for each attribute). The resulting quadrants guide strategy:
- High importance, low performance: priority for improvement. These attributes drive satisfaction but you're underperforming. Investing here yields the biggest return.
- High importance, high performance: strengths to maintain. These drive satisfaction and you're performing well. Don't take them for granted.
- Low importance, low performance: low priority. These don't drive satisfaction much, so poor performance here is less damaging.
- Low importance, high performance: potential over-investment. You're performing well on things that don't move the needle. Resources might be better allocated elsewhere.
Interpreting Results
A few principles guide sound interpretation:
Relative importance matters more than absolute values. The specific beta weight or R-squared contribution is less important than the ranking of attributes relative to each other.
Derived importance and stated importance often diverge. Respondents tend to rate everything as important when asked directly. Derived importance differentiates better because it's based on actual patterns in the data.
Context-dependent drivers shift. The key drivers for a luxury hotel differ from those for a budget hotel. Run KDA within segments, not just for the total sample.
Low performance inflates derived importance. If everyone rates an attribute poorly, it may show high correlation with overall satisfaction simply because it's dragging scores down. Check whether the driver is genuinely influential or just universally underperforming.
When to Use Key Driver Analysis
- Customer satisfaction programs: identifying which touchpoints or attributes have the biggest impact on overall satisfaction scores.
- NPS follow-up analysis: determining what separates promoters from detractors beyond the topline score.
- Product development: prioritizing which features to build or improve based on their impact on purchase intent.
- Employee engagement surveys: identifying which workplace factors most influence engagement and retention.
- Brand equity studies: understanding which brand attributes drive consideration, preference, or loyalty.
Common Mistakes to Avoid
- Using stated importance instead of derived importance: when you ask people "how important is price?" almost everyone says very important. Derived importance from KDA reveals the actual relationship between attribute performance and outcome, which is far more useful for prioritization.
- Ignoring multicollinearity: if "friendliness of staff" and "quality of service" are highly correlated (which they often are), standard regression can't cleanly separate their effects. Use relative importance analysis or Shapley values instead of raw beta weights.
- Running KDA on the total sample only: key drivers often differ by segment. What drives satisfaction for new customers may differ from what drives satisfaction for long-term customers. Run segment-level KDA to uncover these differences.
Quali-Fi Support
Quali-Fi's analytics include derived importance scoring that automatically identifies key drivers from satisfaction and experience surveys. The platform generates priority matrices showing each attribute's performance against its importance, highlighting the highest-impact improvement opportunities. For advanced modeling, data exports to SPSS, R, and Tableau support Shapley value and dominance analysis workflows.
Frequently Asked Questions
How many attributes should I include in a KDA?
Aim for 8-15 attributes. Fewer than 8 might miss important drivers. More than 20 creates multicollinearity problems and makes the results harder to act on. Group granular items into broader categories if your survey has many detailed attributes.
Does KDA prove causation?
No. KDA identifies statistical association, not causation. A strong derived importance score means an attribute co-varies with the outcome, not that improving it will necessarily improve the outcome. However, when combined with experimental evidence (A/B testing, controlled interventions), the case for causation strengthens considerably.
Can I run KDA with NPS as the outcome variable?
Yes, and it's one of the most common applications. Use the 0-10 NPS rating as a continuous outcome variable rather than the promoter/passive/detractor categories. This preserves the full range of variation and produces more stable driver estimates.
Related Topics
- Importance-Performance Analysis
- Gap Analysis (Research)
- Segmentation Analysis
- Data Visualization for Research
- Correspondence Analysis
- Data Collection Methods
Identify what drives satisfaction with built-in importance scoring. Start your free 14-day Quali-Fi trial, no credit card required.