Homechevron_rightFeatureschevron_rightHyper-personalization

Feature 01 · AI Personalization

Emails that read like a human did the research.

Our writing model wasn't trained on the open internet. It was trained on 25 million real B2B outbound emails, the ones that worked, and the ones that didn't. The output isn't merge fields with a name.

Book a demo See the proof

25M+

B2B emails analyzed

6×

Reply rate vs templates

Angles per prospect

12.4%

Average reply rate

01Watch the AI write

Live, the writer types the opening line.

25M+

B2B outbound emails used to fine-tune our writing model. By reply rate, by industry, by what gets ignored.

Most "AI email writers" are GPT-4o with a prompt that says "write a cold email." The output is technically grammatical, but it has the unmistakable flavor of a model that learned email from the open internet. We built ours differently.

The model reads each prospect. Funding, hiring, product launch, award, partnership, whatever signal is freshest. It writes to that one.
6× reply rate vs templates. Same data inputs, different outcomes.
Skips weak personalization if the signal isn't there. No filler.

See the proof

AI Writer Generating

Andrea KnoxVP Sales · Stripe · LinkedIn + Crunchbase

Series B

From: maya@yourco.io · To: andrea@stripe.com

Saw your Series B

Hi Andrea,

Best,
Maya

Funding

Hiring

Product

How it works

Research, write, pick a winner. Three steps, every send.

When ReachIQ writes an email for a specific prospect, three things happen at once.

manage_search01

Crawlers research

Recent posts, company news, funding, hiring, awards, events, partnerships, tech stack, GitHub commits, Crunchbase.

tune02

Model writes

Research feeds the writing model along with vertical, role, and prospect signal. No fixed variants. The model writes to whatever matters most for this prospect.

check_circle03

Best variant picked

The model recommends the strongest opener based on signals it surfaced. Your SDR can override.

send04

Sent on cadence

Email goes out at the prospect's optimal send time, tracked per recipient. Auto-pause on reply.

The training data

What "25 million emails" actually means.

Other tools claim "AI personalization" and mean merge fields with a name. We mean a writing model that has seen what works in your specific industry, at your specific stage, for your specific buyer.

Slice 1

A fintech VP Sales doesn't open like a healthcare CIO.

Different opening lines, different proof points, different objections. The model knows the patterns from millions of sends in each segment.

Slice 2

A Series A founder doesn't respond like a Fortune 500 buyer.

Stage-of-company signals tune the writing voice. Procurement language to enterprise. Founder language to startup.

Slice 3

A 50-person CTO reads emails differently than a Stripe CTO.

Company size is its own model dimension. Our training set covers the entire spread, so the output knows the difference.

Proof, not pitch. Same prospect, two emails.

Templated personalization is merge fields with a name. We mean something different.

Template + merge fields2.1% reply

Quick question, Andrea

Hi {{first_name}},

I noticed you work at {{company}} as a {{title}}. Curious, are you looking to scale outbound at {{company}}?

We help {{industry}} teams book more meetings. Open to a quick call?

ReachIQ AI12.4% reply

Congrats on the $42M Series B

Hi Andrea,

Congrats on the Series B last week. Saw you're scaling the sales team and hiring 4 SDRs this quarter. Curious if you're building outbound in-house or evaluating partners while the team ramps.

We helped Linear hit their first 50 enterprise meetings post-Series-A in 6 weeks. Happy to share, no pitch.

The voice

What we don't do, and won't.

The model is fine-tuned to do the opposite of generic, and to refuse when the data isn't there for genuine personalization.

The bad personalization

What other AI writers still do.

If your "AI" still generates these, it's just GPT-4o with a thin prompt.

47-word "Hope you're well!"Filler opener that asks for nothing. Skip rate, ~85%.
"Your impressive work in [vague space]"The classic LLM hedge. Reads as confessional fake-flattery.
"I came across your profile"Universally read as a sign the writer didn't actually look.
"I'd love to learn more about your role"The "tell me about you" trap. SDRs send this when they have nothing to say.
Faking a personal connectionPretending to have met at a conference, mutual friend, etc.

What ours does instead

Real signal, or no email at all.

If the writer can't find a real signal worth referencing, it skips the prospect rather than fall back to a template.

References the actual signalRecent funding, role change, product launch, hire, news mention.
Reads the freshest signalFunding, hiring, product launch, award, partnership. The model picks per prospect.
Specific proof in the second paragraphA named customer in the same stage / industry / role.
One clear askEither a 15-minute call or a single resource. Never both.
No-fit skipIf no signal worth quoting exists for the prospect, the AI doesn't send.

Questions

Hyper-personalization, explained.

Aggregated and anonymized outbound campaign performance from the ReachIQ client base over the past four years, combined with public benchmark data. We never train on a specific customer's data without explicit opt-in, and we never use one customer's data to inform another's campaigns.

Most cold-email tools give you a template editor with merge fields. ReachIQ generates the writing per prospect, using a model fine-tuned on outbound-specific data. The merge-field tools are great if you already have an exceptional writer crafting templates. ReachIQ is for teams that want the per-prospect writing to happen automatically.

Yes. Every generated email lands in a review queue. Your SDR can edit, swap to a different angle, or kill it entirely. Done-for-You teams typically auto-send the top-confidence variants and review the borderline ones. Just-the-Tools teams typically review the first 500 emails per campaign, then loosen the gate.

English (US, UK), German, French, Spanish, and Portuguese (Brazilian) at production quality. The 25M-email training base is English-dominant; the other languages benefit from broader pretraining and a smaller fine-tune. Other languages are available on request.

Yes. During setup, we collect 6-12 sample emails that reflect your voice (typically from your founder or top-performing SDR) and tune the output to match. You can also write a 1-page voice brief that the model honors per sequence.

Keep exploring