Reproducible business proof

Synthetic lead-intake routing demonstration

A focused, inspectable routing test using synthetic inquiries only. It measures classification and review routing, not reply quality, customer-system updates, scheduling, or production savings.

Method

Deterministic keyword rules with confidence below 0.75 routed to human review.

100 synthetic inquiries across 5 categories.

Result

95.0% routing accuracy on the labeled synthetic set. 44 of 100 routes require human review.

Failure examples

  • synthetic-001: expected sales_lead, routed manual_review
  • synthetic-018: expected spam, routed manual_review
  • synthetic-052: expected support, routed manual_review
  • synthetic-069: expected sensitive_review, routed manual_review
  • synthetic-086: expected sales_lead, routed manual_review

Limits

  • Synthetic inquiries do not establish production accuracy.
  • No customer data or customer attribution is used.
  • Time savings are not claimed because handling time was not measured.
  • Sensitive and ambiguous cases remain under human review.

Inspect the generated JSON report.