B2B SaaS · AI SDR
Series B SaaS blocked 178 hallucinated discount claims in 30 days
Published May 1, 2026
Outcome
178 blocked claims · 0 reached an inbox · 1 saved enterprise deal
The problem
The customer ran an AI SDR generating outbound emails to mid-market prospects. Sales leadership discovered the model had invented unauthorized discount language in a follow-up to a Fortune 500 buyer the week before the case study began.
The risk
A single fabricated discount in an enterprise inbox triggered a 4-week procurement freeze and required legal review. The customer estimated downside exposure at $140k per shipped incident.
Implementation
Replaced the Resend send call with a POST to /v1/gate/outbound. Configured a workspace policy with a regex set for unauthorized discount phrasing, a denylist for competitor names, and a tone-check rule for guarantee language. Wired RevOps email notifications and Slack as the approval channel.
What changed
- 178 hallucinated claims blocked over 30 days (avg 5.9 per day).
- 92% of risky payloads resolved by a single reviewer in under 60 seconds via Slack approval.
- One enterprise deal pipeline kept alive after a near-miss claim was intercepted.