Overview
Split traffic across rules/models and pick winners on KPIs like CSAT, cost, and containment.
Problem
Routing and prompt changes ship without evidence; hard to know what actually helps.
Solution
Chat Core randomizes traffic across routes/models, tracks outcomes, and reports significance in InsightLake. Promote winners with one click.
How it works
Define experiments in the Admin UI; choose split, KPIs, and stop rules; export traces and outcomes for audit and reproducibility.
Who is this for
Expected outcomes
- Evidence-based routing & model selection
- Lower per-contact cost without hurting CSAT
Key metrics
Cost per resolved contact
Baseline
2.1 USD
Target
1.4 USD
Containment (winner vs control)
Baseline
0 pp
Target
6 pp
Gallery
Downloads & templates
Case studies
Retail CX trims LLM spend 23%
A/B between models cut cost while improving containment by 4pp.
Security impact
- Aggregated experiment metrics & traces · PII: none (aggregated)
Compliance
- SOC2 (monitoring & change control)