B2B Outreach at Founder Scale: 236 Personalised Pitches, Every One Human-Sent

The problem

SEAR's retail growth runs on wholesale relationships — surf shops, swim retailers, ENT clinics, surf schools, sports governing bodies. Reaching them means researched, personalised outreach at a volume no founder has evenings for. Generic wholesale blasts get deleted, and deserve to.

The previous manual process

Research a shop for twenty minutes, write an email from scratch, quality depending entirely on my energy that evening. Follow-ups lived in my head. Realistically: a handful of contacts per week.

The AI-assisted workflow

The design principle first, because it's the whole system: Claude drafts and analyses; I send and decide. Every outbound email — cold pitch, follow-up, everything — is created as a Gmail draft. Nothing sends itself. That was a control decision, not a technical limitation.

The process, per segment:

1. Research and segment. Web research builds a prospect list per target segment, with named contacts where findable. We ended up with eleven categories across the campaign.
2. Segment-specific framing, not templates. Each segment gets genuinely different problem framing: surf schools hear "your students can still hear coaching" (the acoustic filter matters there); ENT clinics hear about patient compliance; events get a sponsorship pitch, not a wholesale one. Same product, five different commercial arguments — that's what the research step buys.
3. Batch drafting. Claude generates the drafts in batches directly into Gmail, then hands me a summary table with send-priority notes.
4. Day-7 follow-ups with a resume queue. Non-responders get short follow-up drafts — three to five lines, designed to re-open a conversation rather than re-pitch — threaded into the original conversation. The generation runs from a JSON work queue saved to disk, so when a long run gets interrupted (Claude usage limits are real), it resumes where it stopped instead of starting over.
5. Warm-check before any cold pitch. Before drafting to a lead, Claude searches the inbox history for the domain. This once turned a "cold prospect" into a paying customer from two years earlier whose reorder request had gone unanswered — so instead of a cold pitch, they got a threaded, apologetic re-engagement. A warm thread beats a cold template every time.

Human review required

Total, by design. I read every draft before sending — and rewrite some. Rules encoded into the workflow: no discounts offered without my approval, no mention of unreleased products, personalisation must reference the prospect's actual sport, region or role.

Outcome

Verified from the campaign tracker (April 2026 snapshot): 236 prospects contacted across 11 categories, a ~7% human reply rate, 4 confirmed wholesale orders and 4 sample requests. The most responsive segments were multi-sport watersports retailers (14%) — roughly triple the response rate of specialist surf shops (4%), which we'd assumed would be our best audience. Follow-up drafts were generated for all 219 non-responders. Revenue attribution: [not measured].

What failed or remained difficult

Long batch runs hit usage limits twice — hence the resume-queue pattern. Deriving company names from email domains produced errors the first time and needed a verification pass (derived fields always do). And the honest structural note: a 93% silence rate is what cold B2B outreach looks like, automation or not. The system's win is that following up with 219 people is now feasible; it doesn't make anyone reply.

What I would change next time

Log my own half of the loop — sends, manual follow-ups, outcomes — as rigorously as the automated half. The asymmetry made later measurement harder than it should have been.

Reusable lesson

Personalisation is a research problem before it's a writing problem, and trust is an architecture decision: the drafts-only rule costs a few minutes a day and means no AI-written word ever reaches a prospect without a human deciding it should.

Resources

The method is extracted into a generic, parameterised playbook in the SEAR Plugs optimisation repo.

• b2b-outreach-playbook — the segment-first research approach and the message spine (hook → purpose-built solution → differentiation → traction → specific ask → low-friction CTA); swap in your own segments and fact sheet, keep the drafts-only rule.
• Pairs with the inbox-as-CRM tracker (see that workflow) for measuring who replied.