AI-powered cold email personalisation that generates unique, contextual messages at scale. From 8% to 28% reply rate using research automation and dynamic generation.
TL;DR
Cold email templates don't work anymore.
"Hi {{FirstName}}, I noticed {{CompanyName}} recently {{GenericObservation}}..." is spam. Everyone knows it. Reply rates reflect it: 2-8% if you're lucky.
But actually personalising emails? That's 40 hours per 100 emails at scale. Impossible.
We built an AI system that researches each prospect, identifies genuine insights, and writes contextually relevant emails -1,000 per day.
Reply rate went from 8% (templates) to 28% (AI-personalised). Cost: £120/month vs £3,200/month for VAs doing it manually.
This is the complete framework.
We tested 5 different cold email approaches with 1,000 emails each:
| Approach | Personalisation | Reply Rate | Positive Reply Rate |
|---|---|---|---|
| Generic template | {{FirstName}} only | 2.4% | 0.8% |
| Basic template | {{FirstName}}, {{CompanyName}} | 5.2% | 2.1% |
| "Personalised" template | + {{RecentLinkedInPost}} | 8.1% | 3.4% |
| VA manual research | Full custom per person | 24.2% | 18.8% |
| AI-powered personalisation | Research + dynamic generation | 28.4% | 22.1% |
The insight: True personalisation (referencing specific, recent, relevant information) is 10x more effective than templates.
But: Manual personalisation doesn't scale. AI does.
What the AI researches:
Example research output for one prospect:
{
"name": "Sarah Chen",
"title": "VP of Marketing",
"company": "DataSync",
"recent_activity": [
"Posted on LinkedIn about struggling with content velocity (3 days ago)",
"Company raised Series A ($8M) announced 2 weeks ago",
"Hiring 3 content marketers per LinkedIn jobs"
],
"tech_stack": ["HubSpot", "WordPress", "Ahrefs"],
"pain_points": ["Content bottleneck", "Scaling team"],
"triggers": ["Recent funding", "Hiring spree", "Mentioned content challenges"]
}
How AI does this:
# Simplified research workflow
def research_prospect(linkedin_url):
# 1. Scrape LinkedIn (using Apify or similar)
linkedin_data = scrape_linkedin(linkedin_url)
# 2. Find recent posts
recent_posts = linkedin_data['recent_activity'][:5]
# 3. Analyze for pain points
pain_points = analyze_with_gpt(
f"What business challenges is this person discussing? {recent_posts}"
)
# 4. Get company data
company_data = enrich_company(linkedin_data['company'])
return {
'recent_activity': recent_posts,
'pain_points': pain_points,
'company_triggers': company_data['triggers']
}
Time per research: 30-45 seconds Cost per research: £0.04 (AI API + data enrichment)
The AI identifies what to reference:
Not: "I noticed you work in marketing" (generic)
Yes: "I saw your LinkedIn post from Tuesday about struggling to hit your 40-post/month content goal with a 2-person team"
The prompt:
Based on this research:
[Paste research JSON]
Generate 3 contextual hooks for a cold email. Each hook should:
1. Reference something specific and recent (last 30 days)
2. Connect to a genuine pain point
3. Feel like you actually read their content (because you did)
4. Be concise (1 sentence)
Example good hook:
"I saw your post about struggling to scale content from 10 to 40 posts/month without hiring -we had the same challenge last year."
Example bad hook:
"I noticed you work in marketing."
Generate 3 hooks:
Output:
1. "Saw your Tuesday post about the content bottleneck -we struggled with the same thing (2-person team, 40-post goal). Managed to 10x output without hiring. Thought you might find our approach useful."
2. "Congrats on the Series A (£8M, impressive for developer tools space). Noticed you're hiring 3 content marketers -before you scale headcount, we found a way to 10x output with same team size using AI. Worth a look?"
3. "You mentioned hitting 'content velocity ceiling' in your LinkedIn post. We hit the same wall at DataSync's revenue stage. Built a system that took us from 15 to 180 posts/month. Happy to share what worked."
AI picks best hook based on:
The prompt (with selected hook):
Write a cold outbound email using this hook:
Hook: "Saw your Tuesday post about the content bottleneck -we had the same thing (2-person team, 40-post goal). Managed to 10x output without hiring. Thought you might find our approach useful."
Recipient context:
- Name: Sarah Chen
- Title: VP of Marketing
- Company: DataSync (Series A, $8M raised)
- Pain point: Content velocity (stuck at 15-20 posts/month)
Email requirements:
- Max 120 words
- Tone: Peer-to-peer (not salesperson)
- Structure: Hook → Credibility → Soft CTA
- UK English
- Sign-off: Just first name
Product: [Product Name] - AI content generation that increased our output 10x
Write email:
Generated email:
Subject: Your content bottleneck (saw your post)
Hi Sarah,
Saw your Tuesday post about hitting the content velocity ceiling with your 2-person team. We had the exact same challenge last year -stuck at 15-20 posts/month, knew we needed 40+ but couldn't justify hiring yet.
We built an AI system that took us from 15 to 180 posts/month (same headcount). Not templates or low-quality spam -actually good content that ranks and converts.
Sounds relevant for DataSync's post-Series A growth phase?
Happy to show you how we did it (15-min call, no pitch). Or if not the right time, totally fine.
Cheers,
Max
Time to generate (including research): 90 seconds Cost: £0.06 per email Quality: Feels hand-written
We A/B tested over 10,000 cold emails:
Example template:
Subject: Quick question, {{FirstName}}
Hi {{FirstName}},
I help {{Title}}s at companies like {{CompanyName}} solve {{GenericPainPoint}}.
Would you be open to a quick chat about how we can help?
Best,
Max
Results (5,000 emails sent):
Example AI email:
Subject: DataSync content challenge (saw your post)
Hi Sarah,
Congrats on the Series A -£8M is brilliant for developer tools.
Saw you're hiring 3 content marketers. Before you scale headcount, we found a way to 10x content output with our existing team using AI (15 posts → 180 posts/month).
Might save you £180K in salaries if it works for DataSync.
15-min call to show you the system? Or if not relevant, totally fine.
Cheers,
Max
Results (5,000 emails sent):
The difference:
The problem: AI finds information but misinterprets it.
Example:
The fix: Prompt AI to frame neutrally:
The problem: Referencing too many specific details feels like stalking.
Example:
"Hi Sarah,
I saw you posted on LinkedIn Tuesday at 9:42 AM about content velocity, noticed you commented on Tom's post about SEO, and saw you changed your profile picture last week..."
Creepy. Don't do this.
The fix: Reference 1 recent thing. That's enough.
The problem: Referencing something irrelevant to your pitch.
Example:
"I saw you went hiking in the Lake District last weekend. Beautiful area!
Anyway, want to buy our SaaS product?"
Disconnect between hook and offer.
The fix: Only personalise around pain points relevant to your solution.
Want AI to research prospects and generate personalised emails automatically? Athenic handles prospect research, email generation, and send automation -scaling outbound to 1,000+ personalised emails/day. See how it works →
Related reading: