We sent 275 cold emails with an AI agent. The telemetry is humbling.

Description and use

We sent 275 cold emails via an AI agent, achieving 91.8 % delivery but zero human replies. An em‑dash encoding bug caused hard bounces, fixed with linting; a self‑hosted pixel tracks opens; a kill‑gate cuts the lane if replies stay low.

Problem

Low reply rate from a cold email campaign run by an AI agent.

Outcome

Delivered 91.8% of 182 verifiable emails; 0 human replies; identified encoding issue; added open tracking; established reply‑rate kill‑gate.

Workflow

• 1. Vet business domains and compile one role address per business. 2. Enforce CAN‑SPAM compliance: role sender, postal address, List‑Unsubscribe. 3. Apply hard send‑caps in the sending script (per‑day and per‑month ceilings). 4. Send 275 emails (first touch only). 5. Read provider event store for last five days; re‑verify receipt for 182 emails. 6. Count delivered (167), permanent failures (15), hard bounces (4), replies (0), opt‑outs (0). 7. Identify em‑dash encoding regression causing 550 errors; fix with body linting and suppression. 8. Deploy self‑hosted aggregate‑only tracking pixel for open rates. 9. Set a kill‑gate on 2026‑11‑04 if replies remain below threshold. 10. Publish telemetry and plan for lane cut if needed.

Evidence

275 countable sends to business domains we researched and vetted by public criteria (no scraped personal emails, no free‑mail, one role address per business, a real suppression list).

Source

delivered (receipt confirmed) 167 — 91.8% (167/182)

Source

replies from real humans 0

Source

Some of the earliest sends were permanently rejected with 550-class errors because of an encoding regression of ours: an em‑dash in the body tripped a receiving filter.

Source