A customer satisfaction index (CSI) combines satisfaction scores across several attributes of the customer relationship into one number, usually on a 0 to 100 scale. Where a single survey question rates one moment, the index describes the whole relationship: product, support, onboarding, pricing, and whatever else the company decides to measure.
The common construction is:
CSI = sum of (attribute satisfaction score x attribute weight) ÷ sum of weights, scaled to 0 to 100
With equal weights, that reduces to the average attribute score converted to a 100-point scale. The formula is simple. The definitions behind it are the hard part, and they are where most CSI programs fail.
CSI is not CSAT, and neither is ACSI
Three terms get mixed together:
- CSAT rates a single interaction: a support ticket, a purchase, an
onboarding call. Qualtrics defines it as the percentage of respondents who pick the top satisfaction ratings for a specific experience.
- CSI aggregates satisfaction across the relationship. It is an index built
from several questions or attributes, tracked over time.
- ACSI, the American Customer Satisfaction Index, is a specific national
benchmark. It uses customer interviews as input to an econometric model developed at the University of Michigan and covers whole industries, not one vendor's account list.
A company can borrow ACSI-style questions (overall satisfaction, expectation gap, distance from ideal) for its own index. It cannot call its internal score an ACSI score, and it should not compare its internal number against published ACSI industry figures collected under a different method.
A worked example
A SaaS company surveys its customer base quarterly on four attributes, each on a 1 to 10 scale, with equal weights:
- product capability: 8.2;
- support responsiveness: 7.6;
- onboarding experience: 6.9;
- pricing fairness: 7.1.
The calculation is:
(8.2 + 7.6 + 6.9 + 7.1) ÷ 4 = 7.45, scaled x 10 = CSI of 74.5
If the company decides support matters twice as much as pricing, the weighted result changes. That is legitimate, but the weights must be documented and then left alone. Changing weights between quarters manufactures movement.
Definition pitfalls
- comparing an internal CSI against published ACSI benchmarks built on a
different model;
- changing attributes, weights, scale, or question wording mid-series without
restarting the baseline;
- surveying only active, engaged users and reading the result as the whole
base;
- averaging away segments: one enterprise cohort at 60 can hide inside a
blended 78;
- treating a relationship index as a substitute for transactional CSAT, which
catches specific broken moments the index smooths over;
- reporting the score without response rate, sample size, and segment cuts.
Where CSI fits in the revenue system
CSI belongs in the Measurement pillar of the Tenbound Pipeline Architecture Standard: a lagging relationship signal read alongside retention outcomes, not a target to manage directly.
A falling index is a prompt to investigate, not a conclusion. Trace the moving attribute to operational evidence: ticket resolution times, onboarding duration, adoption depth, renewal-risk notes. Pair the index with net dollar retention to check whether stated satisfaction and revenue behavior agree. When they disagree, trust the revenue behavior and fix the survey.