Cold email + AI demo: a repeatable playbook that gets 8% reply rate
The exact sequence, sample size, and hit rate of the cold-email-to-live-AI-demo motion that moved rag.art from zero to qualified pipeline in six weeks.
Cold email is the worst channel. It's also the fastest B2B channel available to a solo founder with zero ad budget, and the only one where conversion is fully in your control. For rag.art, the playbook that worked wasn't clever — it was mechanical. Here's what we ran, what we hit, and where the leverage was.
The numbers
- 850 emails sent over 6 weeks across 8 verticals in Spain.
- 11.7% open rate (weak — subject-line still needs work).
- 8.1% reply rate of opens (strong for cold — median is 2–3%).
- 3.4% of replies booked a call.
- Conversion-to-demo-to-qualified-lead: 42%.
The one move that tripled reply rate
The single intervention that moved reply rate from 2.5% to 8%: we stopped sending cold emails. We sent warm demos. Each message linked to a chatbot we had pre-built for the prospect's own website. The subject line named their domain. The body had one sentence, one link. The bot on the other side of the link was already trained on their pages.
Nothing beats the visceral 'wait, this is my stuff' moment. Prospects played with the bot for 2–10 minutes before replying. The replies were rarely 'no'. They were 'interesting, what would this cost?' or 'how did you build it?'
The automation
We generate ~50 pre-built demos overnight on a schedule. The pipeline: fetch target-list websites, ingest each, create a demo bot at <slug>.demo.rag.art, send the email at 9:15 local time with that URL.
The cost per demo is under €0.10 — embeddings + a small model for suggested questions. The value of the 'holy shit' reaction is in another economic universe entirely.
The template
Subject: a chatbot for {{domain}}I built this playing with your site: https://{{slug}}.demo.rag.artTry asking it '{{the question your reception gets ten times a day}}'.No ask attached. Just wanted to show you what's possible now.— Carlos
What doesn't work
- Merge tags beyond {{domain}}. The more personalised the template looks, the less personal it feels.
- Case studies in the first touch. They convert a warm lead, not a cold one.
- Sequences of 5+ emails. Two touches, then stop. If they didn't bite twice, they won't bite at seven.
- Sending before 9:00 local. Pre-coffee replies are 3× more negative.
The compliance note
Spain has the LSSI on top of GDPR. B2B cold email is permissible under LSSI Art. 21 if the message is commercial, identifies the sender clearly, and includes an opt-out. That's a one-line signature footer. We write about it separately — see the 'GDPR, LSSI and the cold email landmine in Spain' post.
The only thing worth optimising
After the playbook is dialed in, the single highest-leverage lever is the bot the email points to. Improve ingest quality, improve suggested questions, improve response latency. Every 200ms shaved off first-token-out increased our book-a-call rate measurably.
The channel isn't magic. The demo is the product. Cold email is just the delivery truck.