This is how the evaluation happens.

The Commerce Eligibility Index measures AI readiness

Flowchart illustrating the process of training a neural network with five steps connected in sequence, with two pink-highlighted steps.

CEI is not a survey, a consultant's opinion, or a search ranking proxy. It is a computed score — built from real catalog data, run through a repeatable analytical pipeline, and delivered with product-level evidence you can verify.

The pipeline

  • Every engagement runs the same sequence. Your catalog data arrives as an export or permissioned crawl. The engine runs ten analysis modes across your SKUs — completeness checks, semantic embedding, retrieval simulation, claim verification, sentiment analysis, and live AI discoverability testing. Sixteen signals are computed. Five pillar scores are produced. The CEI score is calculated as a weighted geometric mean of all sixteen.

  • Geometric mean, not additive. A strong Authority score cannot compensate for broken Foundation data. Every pillar has to work.

What the engine actually does

  • Detects terminology drift across your catalog using sentence-transformer embeddings — so "galvanized," "galv.," and "zinc-coated" are recognized as the same thing your catalog is inconsistently naming

  • Generates hundreds of realistic shopping queries and runs them against your catalog to measure retrieval quality and constraint satisfaction

  • Extracts specific product claims and cross-validates them using NLI (Natural Language Inference) to detect fabrication and distortion before AI systems do

  • Queries Claude, ChatGPT, and Gemini with real shopping scenarios to measure whether your brand actually appears — and whether what they say about your products is accurate

  • Analyzes on-site and third-party review sentiment using a transformer model trained on 150,000+ product reviews

The engine is powered by Bubo AI™

The CEI pipeline is built on Bubo AI™ — a production-grade machine learning platform designed to power analytical products in real operating environments. Bubo AI handles model selection, uncertainty quantification, and drift detection, ensuring every score reflects current catalog reality rather than a snapshot that quietly goes stale. Learn more about Bubo AI →

What you get

Every finding links back to product-level evidence. No signal is reported without a traceable source. Confidence indicators disclose what was measurable and what wasn't — you never receive a falsely precise score.

The output is a board-ready diagnostic, not a traffic report.