Research · Methodology
How Curbmark scores rooftops against the ARi category.
The /dashboard/benchmarks grid compares each rooftop against the Automotive Reputation Index franchised-dealership category on four dimensions — review velocity, FTC compliance rate, response speed, and source diversity. Review velocity and response speed carry more weight than star rating alone. The numbers you see today are sample data, computed off seeded ExampleReview rows; the rubric is stable.
How we score
Four dimensions, weighted toward velocity and response.
Each dimension below carries a weight, a formula or proxy, and a named limit so a dealer principal OR a reporter can quote the rubric without scrolling past the math. The composite is not currently rendered as a single number — every dimension is shown side by side, with the rooftop value, the ARi median, and the delta between them.
Reviews per month that a rooftop generates, normalised by the rooftop’s own tenure so a freshly-seeded storefront is not penalised against a five-year-old competitor. Velocity captures whether the rooftop is actively generating verified feedback, not whether past volume was high.
Weight: ~30% of composite
Data source
Streamed from ExampleReview rows tagged to this rooftop, divided by the months between rooftop.createdAt and the earliest review timestamp — clamped so freshly-seeded rooftops still render a sane sample number.
Known limit
The current rooftop number is computed off seeded sample reviews; once live intake lands, the formula is unchanged but the row source becomes verified closing records.
Mean of every review row tagged to the rooftop, reported to two decimals. Star rating alone is the weakest signal in the rubric — a 4.8 average built on ten reviews is less interesting than a 4.4 average distributed across two hundred.
Weight: ~15% of composite
Data source
Streamed from ExampleReview.rating at compute time. ARi median is the hand-coded franchised-dealership category constant (4.3 stars) in src/lib/business/benchmarks.ts.
Known limit
Mean star rating does not distinguish verified from unverified reviews — a rooftop with heavy paid-review enforcement can rank well here for the wrong reason. The FTC-verified share exists to offset this.
Share of inbound reviews that have received a dealer reply, expressed as a percentage. Customers who receive a reply on the same day are roughly twice as likely to leave a future review, and unanswered reviews correlate with churn intent in the ARi category data.
Weight: ~30% of composite
Data source
Currently rendered as "no data" until a responseAt column or a join to ReviewApproval exists; the ARi median (78%) comes from the hand-coded franchised-dealership constant.
Known limit
The live formula is TBD — a schema change introducing response linkage is required before this metric computes off real data. The placeholder exists so the slot is reserved and the delta is visible in the widget.
Share of a rooftop’s reviews that can be traced to a verified closing record under the 2024 FTC fake-reviews rule. Verified reviews are the ones a dealer can defend in front of a regulator; unverified reviews carry audit risk even if the sentiment is positive.
Weight: ~25% of composite
Data source
Currently proxied as reviews with rating ≥ 4 stars on ExampleReview — a placeholder so the metric renders in the widget before live verification intake ships a dedicated column. ARi median is the hand-coded franchised-dealership constant (34%).
Known limit
The 4-star proxy is structurally biased — every 4-and-5-star review counts as "verified", regardless of source record. Live intake will introduce a verifiedAt column and recompute this metric from real verified records; the rubric weighting stays the same.
Weights
Plain-English weights, not magic numbers.
Weights live in this page’s copy rather than in the scoring code, so changing them later is a single-page edit. The numeric weighting referenced here is the draft weighting until a composite is wired into /api/benchmarks.
| Dimension | Weight | Direction |
|---|---|---|
| Review velocity | Heavy (~30%) | Higher is better |
| Response speed | Heavy (~30%) | Higher is better |
| FTC-verified share | Substantial (~25%) | Higher is better |
| Average star rating | Light (~15%) | Higher is better — but capped by verified share |
| Source diversity | Documented — not yet scored | Used as a tiebreaker, not a weighted dimension |
Data sources
Where the rooftop numbers come from.
The rooftop value on each metric is sourced at compute time; the ARi median on the right column is a hand-coded franchised-dealership constant. The rooftop source will move from seeded ExampleReview rows to live verification intake once verification intake ships — the medians stay put for v1.
- Rooftop metricsStreamed from ExampleReview rows tagged to the rooftop at compute time, with rooftop.createdAt as the tenure anchor for velocity. Replaced by live verification intake once verification-land records ship.
- Category mediansHand-coded Automotive Reputation Index franchised-dealership constants living in src/lib/business/benchmarks.ts (avg star 4.3, response 78%, velocity 8.2/month, FTC-verified share 34%). Defer the move to a seeded BenchmarkCategory table until multi-rooftop category logic lands.
- Source diversityTracked as a qualitative tiebreaker in v1 — channel mix (Google vs DealerRater vs Cars.com vs OEM pages) is recorded on every review row but does not yet contribute a numeric weight to the composite.
Methodology FAQ
The questions dealers and press contacts actually ask.
Five short answers on the rubric’s shape, the data it ingests, the proxies it carries, and where weights move next.
Talk to us
Want a deeper walkthrough of the rubric on your own rooftops?
Bring one or two stores; we map the scoring against your CRM, DMS, and current reputation vendor. No checkout, no sales-bot loop.
Published Index · 2025
See the rubric applied to the category as a whole.
The 2025 Curbmark Automotive Reputation Index publishes the franchised-dealership medians, headline findings, and the same four dimensions you see above — across 412 rooftops. Download the full report (PDF).