Documentation correction, 2026-10-15: ABsmartly publishes product documentation for its MCP server. The overview describes experiment-management operations; the setup guide documents hosted endpoints and local execution, and the authentication guide documents API-key and OAuth options. This is a documentation check, not a hands-on workflow or reliability test. The earlier statement below that MCP is not documented is superseded.
Review status: MCP documentation checked on 2026-10-15. A source fetch or enrichment date is not a full factual review. Human editorial review of the older assessment remains pending.
Earlier assessment retained for editorial review (other claims below have not been reverified in this correction):
ABsmartly deliberately doesn't market AI. The positioning is built by developers, for developers, around trustworthy A/B testing. The Health Check Panel is called an experimentation copilot, but it's rule-based anomaly detection (sample ratio mismatch, audience mismatch, variable conflicts), not generative AI.
The group sequential testing engine cuts test cycles by 20-80% against a fixed horizon, twice as fast as standard sequential testing. Warehouse Native mode runs analysis inside the customer's own warehouse: BigQuery, Snowflake, ClickHouse, Redshift or Databricks. A hybrid deployment mode is planned.
An MCP server repo exists (absmartly/mcp-server) but isn't documented in the product docs yet. The LaunchPad browser extension lets non-technical users launch tests.
Pricing is event-based, starting at €60,000/year for 50M events a month, with unlimited seats, experiments and goals.