10 to 15 qualified calls every week.

The average founder's cold email experience looks like this: they spend an afternoon writing five email templates, buy a list from Apollo, connect their main domain to a sending tool, and launch a campaign. A week later, their open rate is 12%, reply rate is 0.3%, and their domain has started showing up in spam filters. They conclude that cold email does not work.

Cold email does work. What does not work is skipping the infrastructure and personalization layers that separate effective outbound from noise. The failures are almost always structural, not strategic.

Ideal Customer Profile targeting is where campaigns win or lose before the first email is sent. Most founders define their ICP broadly - "B2B SaaS companies with 10–200 employees" - and then wonder why reply rates are low. The broader your ICP definition, the less relevant your copy can be, and relevance is the single most important variable in cold email conversion.

Your ICP definition needs to go three levels deep: the firmographic layer, the behavioral layer, and the trigger layer. Each layer narrows your list and sharpens your copy simultaneously.

Deep Loom uses a three-source list-building approach: Apollo for firmographic filtering, LinkedIn Sales Navigator for behavioral signals, and intent data providers like Bombora or 6sense for in-market signals. Each source produces a different quality of lead, and the best campaigns blend all three.

Before loading any contact into a campaign, run them through email verification (NeverBounce or ZeroBounce) and a manual spot-check of the top 20 contacts on the list. If the top 20 do not look like your best customers, the rest of the list will not perform. Quality control before launch is far cheaper than a damaged domain reputation after it.

Infrastructure is the least exciting part of cold email and the most important. Every experienced outbound operator has a story about burning a domain because they skipped the warm-up or ignored authentication settings. Do not be that story.

The infrastructure layer has four components: domain setup, authentication configuration, inbox warming, and sending volume management. All four must be in place before you send a single prospecting email.

Register separate sending domains for outbound - never use your primary domain. Buy three to five domain variations of your brand (e.g., usedeeploom.com, getdeeploom.com, deeploomhq.com) and distribute your sending volume across them. This protects your primary domain and gives you redundancy if one sending domain develops reputation issues.

Set up one or two inboxes per domain. More than two inboxes per domain on a new sending domain raises flags. Each inbox should have a real name, professional headshot as the profile image, and a consistent signature. These are real people sending emails from your team - they just happen to be outbound-only identities.

The copy gap between a 1% reply rate and a 8%+ reply rate is almost entirely explained by personalization. Generic cold email reads like spam because it is spam - the same message sent to thousands of people with a first name variable swapped in. AI-powered personalization changes the economics of writing individual, relevant messages at scale.

The goal is not to fake personalization. It is to use AI to surface real, specific signals about each prospect and incorporate those signals into the opening line, the pain point reference, and the call to action. Every email should read like it was written specifically for that one person - because functionally, it was.

Signal Collection: For each contact, collect five data points before writing: recent LinkedIn post topic, company website headline, a recent news mention if available, their job title and likely pain point, and their company's current growth stage. Clay is the best tool for automating this research at scale - it can pull all five signals in seconds per contact.

Prompt Engineering: Feed those five signals into a Claude prompt that generates a personalized opening line, a relevant pain point frame, and a call-to-action variation. The prompt should include instructions for tone (concise, direct, not salesy), word count (opening line under 25 words), and what to avoid (jargon, compliments that feel hollow, questions that require effort to answer).

Human Review Layer: Do not send AI-generated copy without a human review pass. Review a random 10% sample of every batch before it goes out. AI will occasionally produce hallucinated details or awkward phrasings that would immediately signal automation. Catching one bad email per hundred is worth the review time.

Sequence structure is where most operators overthink. They write seven-email sequences with elaborate storytelling arcs. The data does not support that approach. What works is a shorter, sharper sequence where each email has one job and earns its place.

The Deep Loom sequence structure is five emails over 21 days. Each email uses a different angle - social proof, direct ask, problem agitation, case study, final attempt - so that a prospect who did not respond to email one may respond to email three because that particular angle landed better for their situation.

Email 1 has one job: get a reply. Not a meeting - a reply. The best openers acknowledge something specific about the prospect, state the one thing you do and the result it produces, and end with one low-friction question. Three sentences is often enough.

Avoid: long introductions, company backgrounds, lists of features, social proof in email one (save it for email two), and multiple calls-to-action. The single question at the end should be answerable in one sentence. "Is this something you are currently looking at?" is better than "Would you be open to a 30-minute discovery call this week or next?"

Most cold email campaigns end when a contact finishes the initial sequence without replying. That is a significant waste of invested infrastructure and research. The follow-up layer - a separate set of automations that activate after the primary sequence ends - can recover 20–35% of contacts who went silent.

The follow-up layer runs on a longer cadence and uses entirely different content. The primary sequence focused on your offer. The follow-up layer focuses on value delivery - sharing resources, insights, and case studies that are genuinely useful regardless of whether the contact ever buys. This shifts the relationship dynamic from "salesperson chasing a lead" to "expert staying in orbit."

Cold email systems that do not get measured do not improve. The metrics layer is not about vanity numbers - it is about identifying the specific components that are underperforming and applying targeted fixes. Every metric points to a specific lever.

Every week, run a 30-minute review of the prior week's data. Identify the lowest performing metric, hypothesize one root cause, and make one change. One change per week, tracked against the same metric the following week. This discipline compounded over 12 weeks produces dramatically better results than making multiple simultaneous changes that obscure causation.

Run A/B tests on subject lines every two weeks - subject line is the single highest- leverage variable for open rate. Test one element at a time: opener style, CTA phrasing, email length, or angle. After three months of iteration, you will have a documented playbook of what works for your specific ICP - a competitive asset no one can replicate because it is built from your proprietary data.