AI SDRs That Work: Real Data from 21,000 Messages
SaaStr shares real AI SDR data: 7.5% response rates, multivariate testing at scale, and why these products didn't work until Q2 2024.
Why AI SDRs Hit an Inflection Point in 2024
This session is a masterclass in AI product honesty. Jason Lemkin (SaaStr founder) and Jasper (Artisan CEO) share real data from SaaStr's own use of Artisan's AI SDR, "Ava." The key revelation: these products didn't work until Q2 2024.
Lemkin sets the context bluntly: "If you came into this skeptical, fair enough because these products sucked until Q2 of this year." He cites Gamma (founded 2020, didn't work until Q2 2024), Replit (10 years old, not good until February 2024), and Lovable. The inflection point was the "Claude 4 moment" - when hallucinations faded and LLMs got reliable enough for production use.
The data is refreshingly specific: 21,000 messages sent over ~5 months, 7.5% overall response rate, 4.5% positive response rate. The response rate actually improved over time (from 3.7% to 4.5%) through autonomous multivariate testing. They test 10+ variants of pain points, solutions, CTAs, and proof points simultaneously - something humans simply can't do at scale.
Practical insights abound: warm outbound performs 2-3x better than cold. LinkedIn worked "too well" - they turned it off because founders got too many connection requests. Two lowercase words in subject lines outperform everything else. Sunday afternoon is the best time to email founders (they're catching up with no meetings).
The honest discussion of buyer personas is valuable: early-stage companies have founders or heads of sales owning it; enterprise has RevOps/Sales Ops orchestrating. One fintech customer runs AI emails on behalf of thousands of AEs - Sales Ops manages it, AEs just get leads delivered.
4 Insights From SaaStr's AI SDR Experiment
- AI SDR products hit an inflection point in Q2 2024 - before that, they "sucked" due to hallucinations and unreliability
- Warm outbound (existing contacts) performs 2-3x better than pure cold; most companies have "hundreds of thousands" of unnurtured CRM contacts
- Multivariate testing at scale (10+ variants across 4 variables) creates continuous improvement humans can't match
- Two-week domain warmup is essential - spin up secondary domains, max 20 emails/day per address, 2 addresses per domain
