How a GTM Agency Runs on Claude Code

You can be up and running in 10 minutes. After a couple of hours experimenting you'll be ahead of 80% of people using it. The only prerequisite is installing it locally — two lines of script.

CLAUDE.md — your master file

This is the memory file Claude Code reads before every task. It stores all of your agency's knowledge, skills, workflows, and context.

Set it up first, then update it constantly. Every new process you build goes in here, so future tasks can reference it automatically.

Plan mode — use it before you build

Before building anything, enable plan mode and describe the task. Claude Code returns several solution options. Pick the most optimal one, then tell it to execute.

This avoids wasted builds and mistakes on complex tasks.

Time investment

Budget a minimum of 1 hour per day per person building in Claude Code. Every process you automate compounds — each one makes the next faster to build.

You don't need to watch tutorials. Ask Claude Code directly what to do and how to build it; it will come up with the execution plan and run it.

The entire onboarding process — from client joining Slack to a fully built onboarding form — runs automatically. Here's the exact sequence.

Step 1: client joins Slack, research starts automatically

As soon as the client enters the Slack channel, a Claude Code agent creates a client folder on GitHub where all code for that client will be hosted.

Three agents then run in parallel:

• Industry analysis
• Competitor landscape research
• Future market trends

If sales call transcripts exist pre-signing, those also feed in as a fourth source.

Step 2: research feeds a custom TypeForm

Claude Code transforms the research into a curated onboarding form in TypeForm.

The form covers the responsible person for the project, sending infrastructure details, DNC list, offer details, and ICP — but the ICP questions are customised based on the deep research done in Step 1, not generic questions.

A Slack agent posts the TypeForm link directly to the client in the channel.

Step 3: TypeForm response triggers three outputs

1. A golden ICP document — one source of truth for that client's ICP, fed to every subsequent agent.
2. A full GTM strategy including campaigns and offers, built by a separate Claude agent reading the ICP document.
3. Campaign copy — 80–90% done without manual input, pulled from an offer bible of best-performing campaigns.

Building email sending infrastructure manually takes around 40 minutes per client. With Claude Code connected to Salesforge's API, it takes 5 minutes — and most of that is pressing go.

The inputs

Feed Claude Code the sending account names, signatures, and profile pictures. Confirm the details. Press go.

What happens automatically

• Sending accounts are created
• Accounts are automatically pushed to the warm-up tool
• Inboxes begin warming without any manual steps

Time saved per client: roughly 35 minutes.

API documentation quality is the bottleneck

This only works when the tool has detailed, clean API documentation. If the API docs are incomplete, Claude Code hits a wall and cannot continue.

When evaluating any infrastructure tool for automation, check the API documentation quality first. If it's not good, the automation is not possible regardless of how you prompt.

Once onboarding is automated and the offer bible is built, campaigns can go live within days of a client signing. The human role becomes quality control and creative input, not production.

The offer bible

A database of your agency's best-performing campaigns and offers, stored in Claude.

When a new client's TypeForm comes in, Claude reads the ICP document plus the offer bible and generates campaign copy that's 80–90% done automatically. The output covers multiple angles, value propositions, and decision makers. Human review adds creativity and brand-specific nuance before launch.

Automating campaign iteration

Pull all campaign data via API from your infrastructure tools — reply rates per message, interest rate, meetings booked. Feed that data back into Claude automatically, and use it as input for the next round of campaign creation.

This creates a self-improving loop: more volume → more data → better campaigns → more volume.

Automated reporting for clients

Automate data dumps to clients showing campaigns live, messages sent, interest rate, and meetings booked.

This gives clients predictability: if 1,000 emails generated 10 positive replies and 4 meetings, doubling send volume means roughly 8 meetings. Transparent reporting also removes the need to manually pull and format stats.

Real data from a client with a TAM of ~2,500 companies. Same message, same value proposition, three channels tested. The gap between them is not small.

Reply rates by channel

LinkedIn's reply rate is roughly 10x higher than email for the same message to the same decision maker. Multichannel campaigns outperform both email-only and LinkedIn-only.

Cold calling (US market)

Phone gets a roughly 3% pick-up rate in the US. But when prospects do pick up, meeting conversion is higher because objections can be handled live.

Cold calling works best layered on top of email and LinkedIn, not as a replacement.

Treat every touchpoint as a first

Don't assume the prospect saw your email before your LinkedIn message. Every channel touch should carry the full value proposition as if it's the first contact.

• For smaller, more targeted lists (1–2k prospects): multichannel
• For very large TAMs where you need fast volume (e.g. 10k+ ecom brands): email-only is still viable

These three playbooks work independently of the client — the trigger is the same, the output is the same, only the company details change.

Event outreach playbook

1. Identify the top events your ICP attends.
2. Find the attendee list, either from the event's public exhibitor page or by scraping behind a login using cookie capture (Claude Code can guide this step-by-step).
3. Run social listening on LinkedIn for anyone who posts about attending the event.
4. Trigger a message referencing the event: "I saw you're attending X — I am too. Would you want to connect for a short chat about [problem]?"
5. Send via LinkedIn from your team's accounts, spread across five people to reach hundreds per week without volume risk.

Competitor follower playbook

1. Identify the top 5 competitors of your client.
2. Find their LinkedIn audience — scrape only the job titles that match the ICP.
3. Research why the client's solution is better than the competitor those people are currently using.
4. Auto-generate copy that highlights the specific differentiation and offers a conversation.
5. Launch as a targeted sequence to that competitor's user base.

Outbound-to-inbound webinar playbook

1. Identify 2,000 ICP prospects who could benefit from a webinar topic your client can run.
2. Send the registration link directly via LinkedIn outreach: "I'm hosting a webinar on [topic]. I think you'd find it useful. Here's the link."
3. Track who attends for 30–40%+ of the duration — those are the highest-intent prospects.
4. Follow up with that group directly as warm leads.

On March 11, Clay introduced a two-tier credit system. If your agency runs multiple clients on one workspace, this directly affects your cost structure.

The old system

• Credits charged only for enrichments — finding emails, phone numbers, running web research
• Third-party API integrations (Serper, Google Maps, OpenAI with your own keys) used zero Clay credits
• Agencies could run 10–20 clients on one Pro workspace

The new system

There are now two credit types: data credits (emails, phones, data lookups) and action credits.

Action credits are charged for every action run inside Clay — including third-party API calls using your own keys. So if you run a Serper search or a custom OpenAI call, you now pay Clay an action credit on top of the third-party tool cost.

Data costs went down 50–90%. But enrichment, scoring, research, and third-party actions cost significantly more. 40k action credits on the Pro plan is roughly one good Clay table, so running 10 clients means 10x the cost.

How to adapt

OpenClaw is a free, locally-run AI agent that can access your CRM, calendar, Slack, and other tools to automate repetitive tasks autonomously. One agency built a full prospecting bot on it that delivers fresh lead lists in 10 minutes.

The prospecting bot use case

Built inside Slack, so anyone in the company can request a new prospect list directly in Slack.

• Inputs: a company-level list plus the job titles you want to find
• What it does: searches local databases, LinkedIn, and tools like Prospeo to find the right decision makers and their contact details
• Output: a fresh, verified lead list in roughly 10 minutes

One use case: feeding 70 cold callers with fresh phone number lists daily, without manual list-building work.

Security rules before you use it

Do not:

• Run OpenClaw on your main work laptop
• Give it access to sensitive or legally protected data
• Give it broad access to everything — it can get expensive fast if unconstrained

Do:

• Run it on a separate, dedicated device
• Install only the apps it needs for the specific task
• Give it access only to data you're allowed to use
• Use completely private infrastructure for the device