Find what drives satisfaction: key driver analysis

Knowing that customers are only moderately satisfied is not enough; managers need to know which levers to pull. Key driver analysis (KDA) estimates how much each attribute (price, quality, service, delivery and so on) contributes to an overall outcome such as satisfaction, loyalty or likelihood to recommend.

Why not just ask customers what matters?

Stated importance (“How important is price to you?”) tends to rate everything as important and favours socially acceptable answers. Derived importance instead infers importance from the data: attributes whose ratings move with overall satisfaction matter most. KDA is the standard way to derive importance.

Four ways to measure importance

• Correlation with the outcome. Simple, but every attribute gets credit for what it shares with the others.
• Standardised regression coefficients (β). They show unique effects, but when drivers are correlated (as they almost always are), β values become unstable and can even take the wrong sign.
• Relative weights (Johnson, 2000). They divide R² into shares for each driver, taking correlations among drivers into account. The shares sum to 100% and are stable.
• Shapley values (general dominance weights). They average each driver’s contribution to R² over all possible orders of entry. They give almost the same answer as relative weights and are the gold standard, but are slow with many drivers.

Reading the example

For 320 customers, seven drivers explain 61.4% of the variance in overall satisfaction (R² = .614). By relative weights, Price Value is the most important driver (22.8% of the explained variance), followed by Product Quality (21.0%) and Customer Service (19.0%); Delivery Speed (10.9%), Brand Image (10.8%), Website Ease (9.1%) and Returns Process (6.4%) matter less. The Shapley values give the same order with nearly identical shares, and the bootstrap confidence intervals show which differences are reliable: the top three overlap, so they are similarly important, but all three are clearly more important than the bottom four.

Notice how correlations alone would mislead: Product Quality has the highest correlation with satisfaction (r = .558), slightly above Price Value (.555), but once the overlap among drivers is accounted for, Price Value contributes slightly more.

Adding performance: the priority matrix

Importance alone does not say what to do. Combine it with performance, the mean rating of each attribute. Attributes with high importance and low performance are the priorities for improvement; high importance and high performance are strengths to maintain; low importance and high performance may be over-invested; and low importance and low performance are secondary. In the example, Price Value and Customer Service are important but perform below average, so they are the first place to improve. This is the same idea as the importance-performance map analysis (IPMA) used in PLS-SEM (Course 2).

Good practice

• Use the same scale for all drivers and the outcome, and ask the overall question before the attribute questions to avoid order effects.
• Choose drivers that are actionable and distinct. Two near-duplicate attributes split their importance between them.
• Check the maximum VIF (here 1.35); high values mean drivers overlap strongly, which makes β unreliable but relative weights still interpretable.
• KDA shows association, not causation. Validate key decisions with experiments where possible.

Reporting

A relative weights analysis (Johnson, 2000) of seven drivers of overall satisfaction (N = 320) explained 61.4% of the variance. Price value accounted for the largest share of explained variance (22.8%), followed by product quality (21.0%) and customer service (19.0%). Combined with below-average performance ratings, price value and customer service emerged as the priorities for improvement.