The pain is real. SDR teams hit quota at a record-low 60% in the Bridge Group's 2025 survey, with 40% annual attrition and 3 months of ramp. A product that replaced that work would sell. In 2024 the AI SDR startups promised exactly that, raised on it, and in 2025 the promise met reality. See AI SDR layer.
The opening
The case that this market is open rests on demand and money. The case against rests on the record.
The 11x story. 11x raised a $24M Series A led by Benchmark and a $50M Series B led by a16z in 2024. In March 2025 TechCrunch reported it had publicly claimed about $10M ARR while an employee source put real revenue nearer $3M once break-clause contracts were removed. Former staff said early cohorts lost 70–80% of customers. 11x disputed this, citing 79% retention for recent cohorts. ZoomInfo said it was never a customer despite its logo appearing on the site. The CEO stepped down to non-executive chairman in May 2025. See 11x.
Artisan. It raised $25M in April 2025 on "Stop Hiring Humans" billboards, with about 250 customers and ~$5M ARR, while its CEO conceded early churn and hallucination problems and that "not every company should be using AI SDR". LinkedIn banned it for about two weeks in December 2025 over use of LinkedIn's name and scraped broker data. See Artisan.
Churn anecdotes, with weak sourcing:
- A data vendor cites "50–70% of teams buying AI SDRs churn within three months", attributed to Reddit consensus, and quotes a user calling 11x a "literal disaster" over CRM contamination and emailing existing customers.
- The same 50–70% figure is attributed elsewhere to UserGems, but the UserGems post does not contain it.
- Gartner, as cited by a vendor, expects over 40% of agentic AI projects to be cancelled by end-2027.
The circulating churn statistics are poorly sourced: the 50–70% figure has no primary source, and a "gross retention below 50%" claim attributed to TechCrunch does not appear in its article. The evidence that early AI-SDR cohorts churned heavily rests on TechCrunch's reporting and practitioner anecdotes, not on a dataset.
Performance claims conflict. 11x claims a 9.7% reply rate on its homepage. A secondary synthesis puts AI SDR replies at 0.5–1.5% after week six (not primary). Bridge Group found AI SDR adoption among surveyed companies at 1% in early 2025.
Who pays and how much
| Product | Price (as of Oct 2026) | Source |
|---|---|---|
| AiSDR | $900/mo billed quarterly; ~$0.75 per message on Explore | MarketBetter (competitor) |
| Reply.io Jason | $500–1,000/mo Starter; $1,500–3,000 Growth | Reply.io |
| Salesforge Agent Frank | $499/mo quarterly; $0.25/contact above 2,000 | Salesforge |
| 11x, Artisan | No public prices | 11x, Artisan |
| HubSpot Prospecting Agent | $1.00 per lead recommended (Pro and Enterprise) | HubSpot |
| Typical platform fees | $12k–36k/yr | Valley (vendor analysis) |
Credit economics. The underlying cost is tiny. Personalising one email costs roughly half a cent on Claude Sonnet at $2/$10 per million tokens (my estimate, ~1,500 tokens in, 150 out). Vendors charge $0.25 per contact to $0.75 per message. Instantly charges 5 credits per AI reply "regardless of whether you send, edit, or skip it". The margin is in opaque metering, which is exactly what churned buyers learned to distrust. Outcome pricing is the counter-move: HubSpot's $1 per lead and Artisan's pay-per-response pilot.
The AI SDR market is put at "$5.81B in 2026" by a market-research firm via a vendor blog, and "41% of enterprise B2B teams" are said to run one. Neither has a primary source in this research. No current ARR, funding or headcount was found for 11x or Artisan in 2026.
Why incumbents have not closed it
They have crowded it. Instantly markets an AI SDR that "can replace a human SDR". Salesforce's outbound agent Hunter is in pilot with Apollo as flagship data partner and GA planned for November 2026. Outreach ships Agent Studio and a Reply Agent with AI credit use up 12x. HubSpot prices its agent per lead. When the CRM and the leading sequencer both bundle the agent, a standalone has no gap to fill. A vendor also reports 11x and Artisan "reverted to hybrid models", which is unverified, but it is the direction the survivors' marketing suggests.
What you would build first
Not this. Build the parts that survive scrutiny: research, draft-for-approval, reply classification and routing, inside a sequencer at sequencer prices (Signal-triggered sequencer, The reply desk, AI reply agent).
How the leaders would respond
They already have, by turning the AI SDR into a feature of products customers already pay for, and by shifting to outcome pricing.
Scores, argued
Pain: 3. SDR economics are genuinely bad and buyers want relief. But the pain is about cost, not a broken workflow, and buyers burned in 2025 now discount the solution.
Gap: 1. Funded startups, every major sequencer, the CRMs and Salesforce all sell or are shipping it.
Size: 3. Budgets are large per account ($12k–36k a year), but churn and distrust shrink the realisable pool. The market-size claims are unsourced.
Moat: 1. The model layer is shared, the data belongs to Apollo, ZoomInfo and Clay, and the distribution belongs to the CRMs.
Speed: 2. A demo takes weeks. Proving that the agent books real meetings without damaging the customer's brand takes quarters, and buyers now demand that proof first.
Safety: 1. Autonomy moves you from conduit to initiator: an agent that chooses recipients and writes and sends without approval is the clearest route to platform liability (Platform liability: what the sequencer itself risks). Hallucinations land in prospects' inboxes under the customer's name. LinkedIn bans vendors with a broker-data supply chain. In Germany an AI-sent email needs the same prior consent as any other (Germany).
What would kill it
For a new entrant it is already effectively closed. What would reopen it is a measured, public, independent result showing autonomous agents beating human-approved workflows on meetings held per euro, without brand damage. No such result exists in the record.
What this means for an entrant
- Ship the parts, not the persona. Draft, approve, send, classify, route. Those are Signal-triggered sequencer and The reply desk, at sequencer prices.
- Steal the pricing idea. Outcome-based pricing (per held meeting) is the one thing the AI SDR wave got right (Pricing strategy).
- Keep a human approval step on by default and log who approved what. It is your legal defence and your sales pitch to churned AI-SDR buyers.
- Be the engine, not the agent. AI SDR builders need deliverable, governed sending (Sending engine as an API). That serves the wave without taking its risk.
- Publish your metric definitions. After 11x, a vendor whose numbers are defined in public is a vendor buyers believe (Trust as positioning).