How We Use Claude Code to Run Outbound for 40+ B2B SaaS Companies

We use Claude Code as the orchestration layer underneath every outbound campaign we run for clients. It ingests demo call transcripts, builds competitor displacement maps, pulls and enriches contact lists, generates copy, and syncs everything to Google Docs for client review. The emails it produces are downstream of a system that compounds with every campaign.

Below is the seven-step build, with details from a real client deployment, plus the operational pieces that let us run the same architecture across 40+ B2B SaaS accounts.

What makes Claude Code work for outbound operations

For outbound operations, the part that matters is the agent loop. Claude Code receives input, reasons about what to do, selects a tool, executes, observes the result, reasons again, and repeats until the task is done. That loop is what lets us chain together "ingest 569 call transcripts, build a competitor map, pull contact lists, generate copy, sync to Google Docs" without a human babysitting each transition.

Claude Code also connects to external tools through MCP. We use that pattern to wire Claude into Apify for list building, Google Docs for copy review, Instantly for cold email, and HeyReach for LinkedIn outreach. Once those connections are in place, the same agent reasoning loop can move across the full client stack.

Some teams say Claude Code can replace parts of workflows previously handled in tools like Clay and speed up implementation. We use it alongside Clay rather than as a replacement, since Clay's enrichment workflows remain central to how we build hyper-personalized lists at scale.

The 7-step system we built

Our most publicly documented build is the outbound engine we created for a B2B SaaS founder. Each step below maps to a specific job Claude Code handles inside the workflow.

Step 1: Competitive intelligence from call transcripts

We ingested demo call transcripts into Claude Code via API integration with the client's call recording platform. The goal was not summarization. It was competitive signal extraction: which tools kept coming up, what language frustrated prospects used, what objections stalled deals, and what made someone stay on the call.

Step 2: Competitor displacement map

Claude analyzed all transcripts and built a competitor map ranking nine competitors across two dimensions: how often they appeared in conversations and how dissatisfied prospects sounded based on exact language patterns. The output is pattern recognition across hundreds of conversations, scored against criteria a human reviewer can audit.

Step 3: Structured knowledge base

Every transcript, intake form, strategy session, and past campaign result got indexed into a structured knowledge base. The base updates as new information is added, and it serves as the central intelligence layer that downstream copy, targeting, and campaign briefs reference.

This is where the competitive advantage lives. Claude Code, Clay, Apify, Instantly, and HeyReach are infrastructure. The knowledge base, fed by proprietary client data no competitor has access to, is what compounds.

Step 4: Automated list building with Apify

We connected Apify directly into Claude Code via MCP. The workflow used Apify to pull decision-maker data and push contacts to a webhook, eliminating manual list building from the loop.

Step 5: Clay enrichment and segmentation

The webhook delivered all contacts directly into Clay. Contacts were enriched, email addresses were double-validated across multiple data sources, and records were segmented by firmographic signals and intent data before anyone entered a sequence.

Step 6: Copy generation and Google Docs sync

Claude Code generated campaign copy based on what the knowledge base surfaced, with messaging built around the exact language target prospects use when they describe dissatisfaction with a competitor. The client could review and edit versions in real time in Google Docs.

Step 7: Multi-channel launch and self-improving feedback loop

Campaigns launched through Instantly for email and HeyReach for LinkedIn. Replies and responses feed back into the knowledge base, so messaging refines over time and the campaign brief updates automatically.

One client's take on the result: "Most agencies finish a campaign and start the next one from scratch. The Understory guys built us a GTM system that gets better every run."

Campaign 2 starts smarter than Campaign 1. Campaign 3 starts smarter than Campaign 2, which is the entire point of building the knowledge base in Step 3.

The AI reply agent

On top of the campaign system, we built an AI reply agent that responds to every positive reply within minutes. It is loaded with a knowledge base containing the specific offer, value propositions, and answers to the most common questions. It reads the reply, understands what is being asked, and responds in a way that moves the prospect toward booking. The knowledge base is updated through review and refresh processes as offers or messaging evolve.

We run this internally too. On one recent day, 5 out of 6 leads the agent followed up with booked meetings. That is a single data point, but it shows what is possible when a reply agent has product knowledge embedded rather than generic auto-response logic.

How we run this across 40+ clients

A few infrastructure details matter when you are running this architecture across a large client roster.

CLAUDE.md for session memory

Each client has a CLAUDE.md file that retains ICP definition, team context, processes, and operating rules across every Claude Code session. When Claude starts a new task, it already knows who the client is, what their ICP looks like, and how prior campaigns performed.

Parallel sessions for research and signal detection

We use Claude Code for account research, signal detection, and daily market briefs. This is how the system scales without proportional headcount increases.

MCP across the full stack

MCP is not limited to Apify and Google Docs. It is how we connect Claude to the rest of the client stack, including LinkedIn Ads and Instantly, with related automation involving n8n, Google Ads, and HeyReach. Internally, our team queries Claude about campaign performance directly through MCP connections.

Cost matters here too. Reported Claude Code costs vary by plan and usage intensity. For what it replaces in manual GTM engineering hours and overlapping tool subscriptions, the cost asymmetry favors clients who run high campaign volume across multiple ICPs.

What this approach actually requires

Some honest framing for growth leaders evaluating Claude Code as the orchestration layer for outbound.

Claude Code operates in a terminal and IDE environment, so it requires familiarity with command-line interfaces and basic scripting. If your team does not have someone comfortable in that environment, you will need help standing up the workflows.

Iterative refinement is required for sophisticated workflows. The system above did not appear overnight. It was built, tested, broken, and rebuilt across multiple client deployments before it stabilized.

We also do not have public quantitative reply rates or pipeline dollar values for these campaigns yet. People have asked. The workflow mechanics are documented, the qualitative outcomes are strong, and the specific numbers are not public. We would rather say that than fabricate them. That is why the knowledge base in Step 3 is the center of our system, not Claude Code itself.

Why this matters right now

Buyer behavior is shifting. Forrester predicts that 61% of B2B purchase influencers will use or plan to use a private generative AI engine to support purchasing decisions. Sellers using AI report performance gains in outreach, though specific lift varies by use case.

The implication for outbound is straightforward. Prospects are getting more sophisticated about evaluating outreach, and generic volume plays produce diminishing returns. The teams that build proprietary intelligence layers, fed by their own data and compounding with every campaign, pull ahead. The teams still starting each campaign from scratch keep wondering why reply rates are declining.

Run coordinated outbound with Understory

Understory builds and runs Claude Code-powered outbound systems for B2B SaaS companies that want coordinated execution across paid media, Clay-powered outbound, and creative.

Our GTM engineering work has helped RemoFirst replace their internal SDR team entirely, scaled Rivial Security's paid media from $20K to $70K in monthly spend, and built Yofi an outbound system that generated more qualified leads than their sales team could absorb.

Book a call to see how the 7-step system would run for your ICP and competitive landscape.

Frequently asked questions

Do you need engineering resources to use Claude Code for outbound?

You need someone comfortable with command-line interfaces and basic scripting. Claude Code operates in a terminal and IDE environment, so the reasoning loop and MCP connections require technical setup before they pay off. For most growth teams, that means a technical hire, a marketing engineer, or an agency partner handles the infrastructure work.

How is the knowledge base in Step 3 different from a CRM?

A CRM stores structured records on contacts, accounts, and deals. The knowledge base stores everything else Claude needs to do its job: call transcripts, intake forms, strategy session notes, past campaign briefs, competitive language patterns, and reasoning behind previous targeting decisions. Claude references the knowledge base during copy generation and campaign briefing, which is how each new campaign starts smarter than the last.

How long does it take to deploy this system for a new client?

Deployment timing varies based on data availability and tool access. The first phase is the heaviest lift: intake, knowledge base setup, and competitive analysis from call transcripts. It requires ingesting historical recordings, past campaign data, and ICP definitions. Once the knowledge base is live, list building, copy generation, and campaign launch move quickly. The reply agent and self-improving feedback loop come online after the first campaign produces enough response data to learn from.

What kinds of B2B SaaS companies see the most value from this system?

The system works best for B2B SaaS companies with $20K+ ACVs, sophisticated buyer journeys, and enough historical demo data to feed a competitive intelligence layer. Companies with fewer than 50 demo calls in their history get less value from Step 1, though the rest of the pipeline still works. Companies that already coordinate paid media, outbound, and creative across multiple vendors tend to get the strongest compounding effect because the knowledge base unifies messaging across all three channels.