AI SDRs: Anatomy of the Hype Cycle

The promise, and the receipts

The pitch was irresistible: a tireless rep that researches, writes, and books while you sleep, at a fraction of a salary. By 2026 the receipts are in, and they are sobering.

50–70%

cancellation rate on managed AI-SDR contracts

Leadgen Economy, 2025–26

90 days

window in which most pilots die

Industry analysis

~2%

of deployments that stick

Industry analysis

Fully autonomous AI SDRs have not replaced human teams at any meaningful scale. The tools once sold as replacements are now, almost universally, repositioned as copilots. That repositioning is the tell.

The decay nobody priced in

The early numbers were real. In 2023 and 2024, AI-drafted cold emails saw a 2 to 3 times reply lift over templates. Then buyers learned the patterns. By 2026, AI-only outbound performs at or below the templated baseline it once beat, and reply rates on AI-SDR campaigns decay 60 percent or more within 18 months as recipients pattern-match the structure.

AI-drafted outbound, reply performance over time

2023–242–3× lift

2025fading

2026≤ baseline

Where AI genuinely earns its place

None of this means AI has no role. It means the role is narrower and more useful than the pitch. AI is strong at the repetitive front half of the job — gathering account context, summarising signals, drafting a grounded first version — and weak at everything that needs judgment.

What to automate, what to keep human

AI does well	Keep with the operator
Research and account context	Which accounts get a non-standard play
Drafting a first version	Reading a delicate reply correctly
Summarising signals at scale	Brand- and legally-sensitive messages
Logging and admin	Knowing when a deal needs a human voice

The hybrid pattern that works

Every successful 2026 deployment keeps a human in the loop. The pattern is consistent: the agent drafts, an operator approves, and only then does it send. Thresholds are set by risk — standard messages to fitting accounts can auto-send; anything novel or sensitive queues for review.

The goal is to remove the typing, not the thinking. The moment you remove the thinking, the numbers tell on you within a quarter.

Salesgluon — operating principle

How to pilot one without the kill curve

01
Scope it to one narrow segment
Not your whole motion. One ICP slice where you can judge output quality by eye.
02
Sandbox the sending
A separate domain, never your primary. If it degrades, you spent a sandbox, not your reputation.
03
Write the kill criterion first
Decide before you start what reply quality and spam rate would end the pilot. The 90-day kill curve catches teams who never set one.
04
Measure on meetings, not activity
AI will manufacture volume happily. Hold it to qualified meetings and guardrail metrics — spam rate, reply sentiment, bounce rate — or it trades your reputation for a busy dashboard.

Sources

Rest of the play · 04

Want the rest of this one?

You have read the opening of “AI SDRs: Anatomy of the Hype Cycle”. Tell me where to send it and I will email you the full play, plus every new one I publish. Operator notes, no spam.

I send it by hand · No spam · Unsubscribe anytime