- A startup backed by $74M and a $350M valuation collapsed in 2025 because its AI SDR sent confident, wrong emails at scale.
The failure mode that killed 11x is quietly killing deployments at companies you'd recognize. 🧵
The problem was never volume. It's verification. - Single-pass LLM systems hallucinate 12-18% of prospect-specific claims. At 1,000 emails a day, that's 120-180 wrong messages in executive inboxes — each one tagging your brand as the company that didn't check.
The grammar is always perfect — that's what makes it worse. - An AI email cites a "recent APAC expansion" from a 2019 article, or says the prospect runs Salesforce when their job ad says HubSpot. The tools feeding them cover ~4,500 public companies — for everyone else, it guesses.
Then the deliverability wall hits. - In Nov 2025 Gmail stopped routing non-compliant mail to spam and began rejecting it at the SMTP level. Your emails don't land in spam — they never arrive. Google's RETVec even flags AI-generated patterns across reworded messages.
- Push your spam rate past 0.3% and your domain reputation takes the hit.
Recovery runs 6-12 weeks of throttled sending — during which invoices, password resets and deal confirmations from the same domain get strangled too. The blast radius is your whole company. - This is the 11x story.
TechCrunch found it showing ZoomInfo and Airtable logos for customers it didn't have. ZoomInfo's actual 1-month trial: it performed "significantly worse" than human SDRs. Ex-staff reported 70-80% churn in early cohorts. $74M couldn't fix it. - And it isn't only reputational. It's legal.
Under apparent authority, an AI agent can bind your company. One promising "guaranteed 100% uptime" or hallucinating "we're FedRAMP authorized" is real exposure. Just 7% of enterprises have agentic-specific governance (Deloitte, 2026). - The volume bet doesn't even pay off.
AI SDRs win higher initial reply rates, but convert 15% of meetings to qualified opps vs 25% for humans. Human reps generate 2.6x more revenue. The tools churn 50-70% a year because the math breaks by day 90. - What works isn't "more AI" — it's verification built in, not bolted on.
Signal-personalized outreach converts 5x better than generic blasts (1-3% → 15-25% reply rates), but only when every claim is checked against source data. Human-in-the-loop pays 317% ROI. - Question for anyone running AI outbound: where does your domain reputation sit 90 days after you turned the volume up? And do you actually verify each claim against source data before send — or trust the model's grammar? #AISDR
- We wrote up the full architecture — hallucination guards, governance, domain-reputation protection — for teams building or recovering AI outreach: https://veriprajna.com/solutions/ai-sales-intelligence