If you’ve read any criticism of NPS or traditional satisfaction measurement, you’ve likely encountered the phrase “driver analysis” or “driver modelling” as the alternative. But the term is rarely explained. This post does that.
A customer loyalty driver model is a statistical framework that identifies the factors influencing customer loyalty in a specific relationship context, measures the relative strength of each factor and maps the pathways through which they operate. The output isn’t a score. It’s a structure.
Why structure matters
Traditional customer measurement treats loyalty as a single outcome to be tracked. Driver modelling treats it as something to be explained. The difference is significant.
When you know your NPS is 34 you know where you stand relative to last quarter. When you know that business relationship quality has a direct effect of 0.41 on loyalty in your customer base and is currently scoring 48 out of 100, you know what to do about it. One number describes a position. The other describes a mechanism.
That shift from description to explanation is what makes driver modelling useful as a management tool rather than just a measurement exercise.
The components of a driver model
A loyalty driver model has three structural elements.
The drivers themselves are the factors that influence loyalty. In B2B these typically include perceived value, product and service performance, relationship quality, supplier image and business relationship integration. The specific drivers and their definitions are calibrated to the context of each engagement — the relationship structure in a SaaS business is different from that in a logistics company.
The pathways describe how the drivers relate to each other and to the loyalty outcome. Some drivers influence loyalty directly. Others work through intermediate constructs like satisfaction or trust. Mapping these pathways is what distinguishes a structural model from a simple correlation analysis.
The path coefficients are the numerical weights assigned to each relationship in the model. A path coefficient of 0.40 between a driver and loyalty means that driver has a strong direct influence on whether customers stay. A coefficient of 0.08 means it has almost none. These weights are what generate the priority matrix.
How the model is built
The statistical method underlying the CLPS driver model is Partial Least Squares Structural Equation Modelling — PLS-SEM. It’s a technique well suited to the realities of B2B customer research: relatively small sample sizes, complex multi-dimensional constructs and the need to model latent variables like trust or perceived value that cannot be directly observed.
The model is estimated from survey data collected across the customer base. Each driver is measured through multiple survey items rather than a single question, which makes the measurement more reliable and the model more stable. The survey is designed around the model structure from the start — not retrofitted to an existing questionnaire.
What the output looks like
The primary output is the priority matrix. It plots each driver on two axes: effect on loyalty on one axis and current performance score on the other. The resulting quadrants tell a commercial team exactly where to focus.
High effect, low performance is the priority quadrant. These are the drivers that move loyalty significantly and where customers are rating you poorly. Investment here has the highest expected return.
High effect, high performance identifies the drivers you need to protect. These are working and they matter — do not assume they will stay that way without continued investment.
Low effect, high performance is where over-investment often sits. You are delivering well on something that does not significantly influence whether customers stay. That budget may be better deployed elsewhere.
Low effect, low performance is where you deprioritise. Improving performance here will not move the loyalty needle in any meaningful way.
Why it works in B2B specifically
B2B customer relationships are structurally different from consumer ones. Purchase decisions involve multiple stakeholders. Relationships span years. Switching costs are real. The factors that drive loyalty in this context — integration depth, strategic alignment, personal relationship quality — are not well captured by single-question metrics.
Driver modelling is built for this complexity. It can handle multiple constructs, model indirect effects and produce actionable output from the relatively small sample sizes that are typical in B2B customer bases. It gives you a map of a complex reality rather than a simplified score that papers over it.
The starting point for everything else
The driver model is the foundation of the CLPS engagement. The Customer Health Report, the CLV and economics modelling, the action roadmap — all of it flows from the model output. Without understanding which drivers matter most in your specific context, the financial projections are assumptions and the action plan is instinct.
That’s why the model isn’t just a deliverable. It’s the analytical core that makes everything else defensible.