Study 2026
Customer satisfaction in Dealerships: What customers really want

In an industry where metrics, studies, and forecasts are part of everyday life, one perspective is often surprisingly overlooked: that of the end customer. Their experience at the dealership is decisive for brand perception, loyalty, and referrals—and thus for long-term economic success.

Our study fills this important gap.

Outcome: well-founded insights into actual service and support quality at key touchpoints—both the Point of Sale and the Point of Service.

Our study draws on 19,000+ verified customer reviews from 474 dealership groups (Jan–Dec 2025) across sales and service. To reduce channel bias, we combined spontaneous online reviews with call‑center prompts, email surveys, and phone feedback. Before analysis, we rigorously de‑duplicated entries, normalized language, and removed all personally identifiable information.

We used a mixed‑methods design: a quantitative baseline from star ratings and visit reasons to enable fair comparisons, plus a qualitative, AI‑assisted analysis of free‑text to uncover topics, sentiment, and patterns at scale. At the core is the rpc Automotive Retail Codebook — an automotive‑specific taxonomy built in two steps: data‑driven topic discovery via transformer‑based semantic clustering at sentence level, followed by MECE refinement in expert workshops. A human‑in‑the‑loop process validated the taxonomy and added underrepresented yet journey‑critical themes. The codebook is modular and versioned for year‑over‑year comparability.

Each verbatim underwent a dual‑pass LLM annotation: topic assignment against the codebook and sentiment on a 0–5 scale. We used domain‑calibrated prompts, inter‑rater checks against expert coders, spot‑audits on edge cases, and drift monitoring to ensure quality. Bias controls covered channel‑mix sensitivity, class‑imbalance handling, and sentiment calibration against human labels and automotive‑tuned lexicons. Clustering parameters were stress‑tested for stability with holdout validation.

We then synthesized results into intuitive metrics: topic share of voice as an attention proxy and a −100 to +100 Net Sentiment Score for direction and intensity. Plotting relevance versus sentiment yields a two‑axis topics matrix distinguishing showpieces, pressure points, hygiene factors, and potential powder kegs. Finally, we contrasted top‑ and bottom‑performing groups in sales and service to isolate the operational differentiators.

Our pipelines are privacy‑by‑design, exclude PII, and aggregate at group level. Codebook, prompts, and model configurations are versioned for transparency. Crucially, domain experts review and interpret the AI outputs—turning raw voices into actionable priorities detailed in the full report.