A deep read of Aaron Shepherd's I Sent 1,500,000 Cold Emails Last Month, Here's What Works in 2026 (17:59). Shepherd runs GrowthFlare, an outbound agency that sent 1.5M cold emails across its clients in a single month and reports thousands of booked calls from that volume.
Why this analysis exists
This is one of the source layers behind the BuildOS cold-email-engagement-first-outreach skill. It supplies the infrastructure + offer + voice layer — the math of how much sending capacity is required for a given goal, the offer-design pattern that earns replies from strangers, and the script discipline that keeps emails from sounding like marketing.
Core thesis
The most defensible argument against the "send only to a perfect list" school is that a small list cannot produce statistical signal. If you send 50 emails and book one meeting, you cannot tell whether the offer worked, the subject line worked, the timing worked, or you got lucky. The next campaign starts from zero. Volume is what generates the data that lets you replicate success.
This works only if three things are true at once:
- The infrastructure can ship the volume without nuking deliverability.
- There is a front-end offer — a no-commitment "test drive" version of whatever you actually sell — that strangers will accept.
- The scripts sound like emails a human typed from their phone, not marketing automation.
Volume + quality on the back end. Pre-call forms and questionnaires sort the good leads from the noise after the reply, not before the send.
Pillar 1: Volume is king (and filter on the back end)
Shepherd's first move is rejecting the "perfect message to the perfect person" school. Not because personalization is bad, but because a small list produces ambiguous results.
"Sure, you can send to a small list of perfect-fit leads, but whether you get good or bad results you'll have no idea what does or doesn't work, since you don't really have enough volume or data to statistically validate it. Which means you have no way of replicating your success the next time you start a campaign."
To send tens of thousands of emails without blowing up deliverability, two things have to be true:
The infrastructure has to actually exist. To reach roughly 100,000 prospects a month, Shepherd uses a 70/30 split: 72 domains on Google, 7 on Outlook, with 5 sending accounts per domain, for 79 domains and 837 sending accounts total. The split exists because (a) different providers underperform at different times and (b) if one ecosystem ever gets restricted, the other carries the campaign. Domains are .info bought in bulk on Porkbun ($3 each). Inboxes come from a reseller like Premium Inboxes ($3/inbox vs ~$7 direct), which also handles SPF/DKIM/DMARC and loading the inboxes into the sequencer.
The leads have to be segmented before send. Apollo is the source-of-record list builder (job title, location, employee headcount, keywords). The list URL goes into a scraper like Ample Leads, which returns a verified CSV. Lists are segmented by vertical so each campaign maps to one persona — that is how the data becomes legible after the campaign ships.
The quality concern — "won't I just book unqualified calls?" — is solved on the back end with a Calendly form gated by pre-call questions: name, business email, monthly revenue, deal size, what they're most interested in. Anyone who answers wrong gets cancelled before the call. Each pre-call question is another layer of friction that filters self-selecting unqualified prospects out before they waste sales time.
Pillar 2: The front-end offer
Volume only matters if anyone wants what you're selling at the moment they receive the email. Most B2B cold email fails here. The main service is too big a yes from a stranger.
The fix is the front-end offer: a smaller, no-commitment, ideally free version of what you do, designed to feel like a test drive of the main service. The car dealership analogy is exact — you cannot get someone to commit to a $40K purchase from a billboard, but you can get them to walk in for a 15-minute drive.
Shepherd's example, for a publishing-service business:
| Core offer (too big for cold) | Front-end offer (size of yes a stranger will give) |
|---|---|
| "Turn your experience into a client-generating book in 60 days, without writing a single word." | "Free strategic book positioning audit — teardown of your existing personal brand messaging and how a book could be structured to attract your ideal clients." |
| "Done-for-you book that unlocks speaking gigs, podcasts, and high-ticket clients in 90 days." | "Book topic validation call — 15 minutes, validates your book idea against what's working in your market." |
| "Seven-figure-founder book funnel teardown — break down a similar founder's funnel, show how their book performed, and what to copy." |
The shape of a working front-end offer:
- Free or near-free. Removes the cost objection entirely.
- Specific deliverable. Not "let's chat" — an audit, a teardown, a topic validation, a sample.
- Custom to the recipient. It would not be the same artifact for any two recipients.
- Opens the loop to the core offer. What you produce in the front-end offer is the natural lead-in to "and here's how we'd do this for real."
Generate front-end offers by feeding ChatGPT the service, target buyer, pain points, and positioning, and asking for three core offer variants and three front-end offer variants. Iterate the prompt until the front-end offers are free, specific, and obviously useful.
This is the part most cold email fails on. The scripts are fine; the offer is too big. "Worth a chat" is not an offer. "Free positioning audit" is an offer.
Pillar 3: Scripts that sound human
The 2026 inbox has more pattern-matching ability than ever. Over-personalized emails — "hi John, congrats on your new role at [Company], saw your post about [topic]" — read as obvious automation. Recipients have been conditioned to delete them.
The counter is to write so casually that the email reads like it came from a phone:
- Subject lines: 2–3 words, lowercase, no punctuation. "quick question," "book positioning," "linkedin idea." The subject should build curiosity, not summarize the email.
- No greeting. Not "Hi John," not "Hello," not "Happy Friday." Open with the first name on its own line, then go straight into the offer.
- Single-line body. One sentence with the front-end offer and a low-friction ask. Use casual contractions and even abbreviations to signal a human wrote it in real time.
- Low-commitment CTA. "Would that be worth sharing more?" not "would that be worth a quick chat?" "Sharing more" is a smaller yes than "a chat."
Example opener Shepherd ships:
First name,
If I could help you map out the exact book you should write to turn your LinkedIn into high-ticket client meetings — would that be worth sharing more?
For follow-ups, the same casual register applies — don't write a fresh pitch. Abbreviate the first message, or add one line of context that points back to it:
First name — if I could show you the book positioning that helps you build awareness that actually converts into pipeline, would that be worth sharing more?
What to avoid in follow-ups:
- "Hey, just nudging you." (Reads as automation.)
- "Probably bad timing." (Tells them they ignored you — guilt-trip energy.)
- "Notice you're building authority on content. Curious if you've ever thought about…" (Theatric prepositioning. Get to the offer.)
- Vague references like "that thing" without enough context that the reader knows what you mean.
Shepherd builds a custom Claude project trained on his best-performing scripts to handle "about 50% of the email copywriting process." The AI is not generating from blank — it has seen what wins for his offer category and is constrained to that style.
What is not in this playbook (worth noting)
- No mention of one-to-one personalization. Shepherd is explicit that over-personalization is a 2022 tactic that now hurts; recipients have learned to recognize it.
- No mention of multi-channel outreach. Phone, LinkedIn, voicemail don't appear in this video. The system is pure email at scale.
- No mention of touch counts. Shepherd shows follow-up samples but doesn't prescribe a specific 4-touch or 2-touch cadence — that gap is filled by other source layers in the BuildOS skill.
Where Shepherd's system diverges from the rest of cold-email-internet
- Volume is the data engine, not vanity. Most "send more" advice treats volume as the prize. Shepherd treats it as instrumentation — how you find out what works.
- Quality lives on the back end. Pre-call forms, not pre-send research, are how unqualified leads get sorted out.
- The offer matters more than the message. Shepherd is direct: "Nothing in a cold email strategy is more important than your offer." Most cold-email training is body-craft training. He says it's offer-craft training.
- Casual beats polished. Lowercase subjects, no greeting, one-sentence body. The opposite of what most agencies teach.
What this layer contributes to the BuildOS skill
The Shepherd analysis supplies the infrastructure math, the offer mechanic, and the casual-voice register of the skill. It pairs with Connor Murray's body craft and cadence (in many ways their voices disagree — Murray is more formal, Shepherd is more casual) and with Austin Schneider's engagement-first / micro-targeting math at Instantly. The skill should let the agent ask which register the campaign requires and route to the appropriate voice.