Updated September 28, 2026 | 11 min read
Updated September 28, 2026 | 11 min read
Ready to launch?
Convert strategy into pipeline
Launch your first outbound campaign with verified leads, AI sequences, and multichannel outreach from one platform.
I Built Outbound Sequences from 244K+ Campaigns Using Claude Skills (Here's the Architecture)
Most outbound sequences are dead on arrival. The channel mix is wrong, the step count is a guess, and the timing is copy-pasted from whatever worked two quarters ago. I’ve watched teams pour hours into cold email copy only to push it through a five-step, email-only sequence with uniform three-day delays. Replies stay flat, and nobody thinks to look at the structure underneath.
I wanted to fix that. In this article, I’ll walk you through exactly how we use the Outbound Campaign Architect, a Claude Skill built on empirical lemlist data from 244K+ campaigns and 249M+ emails, to design outbound sequences that actually match how your ICP engages. You’ll see the full workflow: how the skill works, how to install it, how to customize by seniority, and how to connect the architecture to live campaigns.
What is an outbound campaign architect
An outbound campaign architect is a system that designs the structure of your outbound sequence before you write a single message. That means deciding which channels to use, how many steps to send, how much time to leave between each one, and what job each touchpoint does. Most teams skip this step entirely and jump straight to writing emails, which is a bit like picking paint colors before you’ve drawn up the blueprint.
The Outbound Campaign Architect is a downloadable .md file you load into Claude. Once loaded, it gives Claude specialized knowledge for a specific task (GitHub - anthropics/skills: Public repository for Agent Skills · GitHub), in this case the knowledge of a senior outbound strategist, minus the years it would take to build that experience yourself.
A couple of definitions worth pinning down:
- Outbound campaign architecture: the deliberate structure of steps, channels, timing, and message types, built around a specific ICP
- Claude Skill: a reusable instruction file (structured Markdown) that gives Claude specialized knowledge for one task, rather than a generic prompt you retype every time. Skills are folders of instructions, scripts, and resources that Claude loads dynamically to improve performance on specialized tasks. (GitHub - anthropics/skills: Public repository for Agent Skills · GitHub)
You can learn more about how Claude Skills work in this step-by-step guide.
Why most outbound sequences fail before the first send
Here’s the pattern I see over and over: a team writes a strong cold email, sends it into a five-step, email-only sequence with uniform three-day delays, and wonders why replies stay flat. Structure causes the failure here, not copy.
Reusing the same sequence for every persona, guessing at step count, defaulting to whatever worked last quarter: these are structural decisions made without data. When the architecture is grounded in benchmark data instead of habit, reply rates tend to move because the sequence actually matches how that audience engages.
How teams design sequences today (and where it breaks)
Before I get into the Claude Skill, it’s worth being honest about the other ways teams handle this.
Spreadsheets and RevOps playbooks. The most common approach. A RevOps lead or sales manager builds a sequence template in a Google Sheet or Notion doc, often based on what a VP of Sales ran at a previous company. It’s manual, it’s slow to update, and it almost never accounts for seniority-level differences. But it works for small teams that only target one persona.
Cadence builders in Outreach or Salesloft. Both platforms let you build multichannel sequences with rules and branching logic. If your team already runs on one of these, the cadence builder is the natural place to design structure. The gap: neither tool benchmarks your architecture against aggregate campaign data. You’re still designing from instinct unless you run your own A/B tests.
Generic AI prompting. You can ask ChatGPT or Claude (without a skill) to design a sequence. You’ll get something reasonable. The problem is that the output is based on general training data, not on what actually performs across real outbound campaigns. The skill closes that gap by embedding empirical data from 244K+ campaigns directly into the instructions Claude follows.
The Outbound Campaign Architect doesn’t replace these methods for every team. If you’re already running structured A/B tests inside Outreach and have a RevOps function analyzing the results, you may not need it. But if you’re designing sequences from a blank page, or if you’re a founder or early-stage team without a dedicated RevOps hire, the skill gets you to a data-informed structure in minutes.
Core components of a data-driven outbound sequence
Every outbound sequence, whether you design it by hand or hand it to Claude, is built from the same four blocks. Get one wrong and the whole sequence underperforms, even if the copy is sharp.
Component | What it controls | Why it matters |
|---|---|---|
Channel selection | Where you reach the prospect | Matches the ICP’s actual behavior, not your preference |
Step count | Total number of touches | Too few and you give up early; too many and you cause fatigue |
Timing and delays | Spacing between steps | Uniform delays tend to underperform variable, channel-aware spacing |
Message type | The purpose of each step | Each step needs a distinct job, not a rephrased version of step one |
Channel mix probably gets the most attention, and for good reason: multichannel sequences (email plus LinkedIn, for example) consistently outperform single-channel ones. But step count and timing matter just as much, and they’re the pieces teams tend to guess at rather than test.
How to choose your channel mix by ICP and seniority
Channel choice depends on who’s on the other end, not which tools happen to be sitting in your stack. A VP and an individual contributor don’t behave the same way, so the sequence shouldn’t either.
- C-suite and VP targets: Shorter sequences, weighted toward LinkedIn and calls. Cold email alone rarely breaks through at this level, since inboxes are gatekept or simply ignored.
- Directors and managers: A balanced multichannel mix works well here. These prospects read email, but a LinkedIn touch or a well-timed call often reinforces the message.
- Individual contributors: Email-led sequences with a few more steps. ICs tend to be more reachable by email, though they often need extra touches since they’re rarely the final decision-maker.
- Founders and startup leaders: Similar attention scarcity to the C-suite, but often more responsive on LinkedIn or direct channels. Short sequences with fast follow-up tend to perform best.
You might be wondering why seniority matters this much. It comes down to how each role filters incoming outreach, and a sequence built for a VP’s inbox habits will likely fall flat with an IC, and vice versa.
How the Outbound Campaign Architect Claude Skill works
The skill takes your inputs and returns a full sequence design grounded in empirical lemlist data from 244K+ campaigns and 249M+ emails, rather than a template pulled from memory. You describe your ICP, list your available channels, and state your goal. Claude does the structural thinking.
What it needs from you:
- ICP description: role, industry, company size, seniority level
- Available channels: email, LinkedIn, calls, SMS, WhatsApp
- Campaign goal: booking meetings, re-engaging cold leads, driving sign-ups, expanding into a new market
What it hands back: a step-by-step architecture, complete with the channel for each step, the message type, the delay before the next touch, and the reasoning behind each choice. That last part matters, since a plan you understand is one you can actually adjust later.
How to install and run the Outbound Campaign Architect in Claude
No coding involved here, just a file and a conversation.
- Download the skill file. Head to the GitHub repository and grab the .md file.
- Load it into Claude Desktop. Start a new conversation and attach the file as context. Claude reads it and picks up the specialized knowledge immediately.
- Describe your ICP and goals. A plain-language prompt works fine here; no rigid format required.
- Review the sequence. Claude returns the full architecture. From there, you can ask it to swap a channel, add a step, or shorten a delay, and it’ll adjust without starting over.
Customize outbound sequences for different seniority levels
Most campaigns target more than one persona inside the same account (think VP, director, and an SDR manager, all at the same company). The skill handles this by producing a new step count, channel order, and timing for each seniority input.
That’s worth pausing on. A different seniority input can change the step count, the channel order, and the timing, all at once. If the channel mix diverges enough between personas, running separate campaigns tends to keep things cleaner than trying to branch one sequence three different ways.
Connect campaign architecture to copywriting and launch
The Architect gives you the blueprint, but two things still need to happen: writing the messages and getting the sequence live.
For copy, a separate Claude Skill takes the architecture as input and writes the email and LinkedIn messages for each step. lemlist publishes these copywriting skills in the same GitHub repository, so the handoff from structure to copy stays inside one workflow.
For launch, lemlist MCP connects Claude directly to the lemlist platform using the Model Context Protocol, an open standard for connecting AI applications to external systems. From the same conversation, you can search leads, enrich contacts, build the sequence, and launch across channels, so the architecture you just designed becomes a running campaign without switching tools.
If you’d rather skip the Claude workflow entirely, lemAgent does the same idea-to-launch flow natively inside lemlist. lemAgent turns your campaign goal into a ready-to-send multichannel sequence. Describe your target, set your tone, and launch. (lemAgent: Build and Launch Campaigns in One Conversation)
Tip: If you’re testing the Architect for the first time, run it on a small segment (under 50 contacts) before applying the structure to your full list. We’ve found internally that smaller test lists give you a cleaner read on whether the sequence is working before you commit your full audience to it.
Optimize live outbound campaigns with benchmark data
Once a campaign is running, the same logic that built it can help you fix it. If replies drop off sharply between two specific steps, that’s usually a structural issue, not a copy issue, and worth investigating before you touch the messaging.
One caveat I’ve learned the hard way: reply drop-offs aren’t always structural. Sometimes the problem is deliverability. If your emails aren’t reaching the inbox in the first place, no amount of sequence redesign will help. Before you adjust the architecture, check your sending health in the Deliverability Hub to rule out bounce spikes, spam placement, or domain reputation issues.
Feed the campaign data back into Claude with the skill loaded, and ask it to recommend specific changes: swap the channel at step three, shorten the gap between steps one and two, cut a step that isn’t pulling weight, or add a LinkedIn touch where email alone isn’t getting traction. Small, targeted adjustments like these tend to move reply rates more than a full rewrite of the copy.
Over to you
Outbound design has historically leaned on intuition, which explains why so many sequences look identical regardless of who they’re targeting. Grounding the structure in real campaign data shifts your time toward strategy and personalization, instead of guessing at step count.
lemlist brings the rest of the workflow into the same place: lead generation, contact enrichment, multichannel execution, and deliverability protection. The architecture you design in Claude carries straight through into a live campaign. Rated 4.6/5 on G2 with 2,000+ reviews, lemlist is where 10,000+ sales teams run their outbound.
FAQs about the Outbound Campaign Architect Claude Skill
What data does the skill use to design sequences? Benchmark data from real outbound campaigns run through lemlist, covering reply rates, step performance, and channel effectiveness across different ICPs and seniority levels. The companion outbound-analyst skill benchmarks against data from 244K+ campaigns and 249M+ emails.
Can I use it without a lemlist account? Yes. The skill is a free file that runs inside Claude Desktop. You’ll need lemlist, or another outbound tool, only when it’s time to launch.
Does it write the actual email and LinkedIn copy? No, it designs the structure only. A separate copywriting skill takes that structure and produces the messages for each step.
How is it different from lemlist MCP? The skill designs the sequence inside Claude. lemlist MCP executes it, handling lead search, enrichment, launch, and optimization from the same conversation.
How is it different from lemAgent? lemAgent handles the full flow natively inside lemlist, including lead finding, sequence building, and message writing in one conversation. The Claude Skill is for teams that want to work inside Claude Desktop and have more control over the architecture step before handing off to lemlist for execution.
Share this post