Visa Agentic Commerce Benchmark
Mechanism and classification
Simulated customer conversation, hidden requirements and buy/no-buy decision → passed or failed test scenario. This is a forecast or ecosystem reference, not evidence of a realised business model.
Editorial scope
The status describes the documented stage of the model; it is not a claim of market penetration or commercial success. Provider metrics remain provider claims unless independently confirmed.
Evidence
- Source 1 · investor.visa.com ↗Published: 05.11.2026
- Source 2 · independence must be assessed separately · benchmarks.finance ↗Publiziert / aktualisiert: 05.11.2026
Complete research dossier
The complete editorial record — including the case-specific mechanism, chronology, limitations and cross-references — is maintained in German. Use the DE switch to open the exact counterpart.
Mechanism and limits
Visa tested 22 AI models across 15 private shopping scenarios. A run passed only when the model identified all stated and implied requirements and made the correct purchase decision.
Visa reports a 42.8% failure rate, but the models, prompts and scenario set are not public and the results have not been independently replicated. Checkout, payment and order placement were outside the test.
Dates and evidence scope
- Recorded event / reference date
- 2026-11-05
- Atlas first publication
- Not separately established
- Source review in this audit
- Not separately established
The editorial update date is not a claim that every source or current product feature was rechecked. Unknown dates remain open. Source publication dates are listed with the evidence above.
Structured evidence and geography register · Unclassified source roles and unassigned origin/market fields are explicitly marked.