← Back to Blog

Can Claude Turn a Buying-Signal List Into a 1:1 Outbound Campaign in 30 Minutes?

By Justin Parnell, Founder at Justin GPT ·

Most teams that buy intent data hit the same wall: the signal arrives, and then nothing happens, or a generic sequence goes out that has nothing to do with why the account showed up in the first place. Justin Parnell's answer is deliberately small. He did not bring an intricate agent system. He brought a plugin and a public repo you can install the same day, and he showed it taking a list of hiring signals, sorting the leads by persona, writing every email and LinkedIn message one-to-one, and loading the whole campaign into Lemlist. This recap walks through the flow, how the per-contact copy actually works, the human review step he refuses to skip, what he sees in results across clients, and why he thinks the AI SDR market is swinging back toward humans.



What is the Signal to Sequence play?

Signal to Sequence is Justin's packaged workflow for turning a buying-signal list into a personalized, multichannel outbound campaign with Claude. The stack in the demo was:

Layer Tool in the demo Swappable?
Signal and contact data SalesIntel Yes. Clay, Firecrawl, or plain Claude web scraping also work
Reasoning and copywriting Claude, running the Signal to Sequence plugin Yes. The plugin also works with OpenAI
Sending (email and LinkedIn) Lemlist, through its Claude connector Yes. HeyReach or other send tools, or HubSpot for email
Reply handling (optional) Justin's assistant in ChatGPT, posting to Slack Yes. Any assistant that can watch the inbox

Justin designed the demo around his own business. As a forward deployed engineer for hire, his best signal is a company that has just posted a job for a forward deployed engineer or a GTM engineer. Behind each of those postings he wanted the decision maker who might be hiring for the role, such as the CMO, CTO, or CPO, with their contact details, enrolled in a sequence that runs on both LinkedIn and email.

Your signal will be different. Website visitors, funding, or any other trigger your data provider tracks can feed the same flow.

How does a signal become a campaign?

Justin broke it into a straight line of steps:

  1. Gather the signal. He pulled the list from SalesIntel. He did not even click through it himself: a browser agent built and downloaded the list for him.
  2. Verify the signal. Claude reviews each signal before anything gets written. Nothing triggers an email until Claude has confirmed the signal is real and matches something you actually want to message about.
  3. Route to a persona skill. Each lead goes to the copywriting skill for their role.
  4. Write the copy. Claude reads what the job posting asks for and matches it to proof of work, in Justin's case public examples from his GitHub repo with no client-sensitive material.
  5. Build and load the sequence. Claude writes everything into Lemlist through the connector.
  6. Human review, then launch. No one receives anything until a person opens Lemlist and hits launch.

From loading the list to a fully customized campaign, Justin put it at about 30 minutes.

What are the persona skills?

The plugin leans on one copywriting skill per persona in your ICP. Justin built eight of his own with Claude. When you install the plugin, setup walks you through creating yours: it asks who your ICP is and whether you have positioning and messaging documents for each persona, then helps you turn those into skills. The public repo ships eight persona skill templates, from CMO to VP of Revenue Operations, that you can adapt.

Justin's advice for teams: have someone central build the persona skills so they reflect the positioning and messaging you want, then hand the plugin to your SDRs and BDRs.

What happens to leads that don't fit a persona?

They get held. In the live run, Justin fed Claude 273 leads. Claude held 183 of them because they did not match any of his persona skills: CTOs, VPs of engineering, CIOs, CPOs, heads of product, heads of FDE, and solutions engineering leaders. Nobody in that group received a message.

Claude also flagged the gap. Those were real buyers, it pointed out, and it suggested building new skills such as a CTO and VP of engineering persona so they could be messaged properly. That is the behavior you want. A lead without a messaging playbook waits until you write one instead of getting a generic email.

How does one sequence send unique copy to every contact?

This is the core trick, and it is simpler than it sounds. The sequence itself holds almost no copy. Each touch is a contact property merge field:

Touch Field on the contact record
Email 1 subject and body email subject 1, email body 1
Email 2 subject and body email subject 2, email body 2
LinkedIn messages DM 1, DM 2

Claude writes the words for each touch onto each contact record, and Lemlist pulls them in at send time. Every person gets their own copy, and the sequence never needs editing. To change one person's message, you change one record. The plugin knows how to do this in Lemlist, and the same pattern works in HubSpot and other email tools that support custom contact properties.

What does the multichannel sequence look like?

Justin's sequence in Lemlist was a standard LinkedIn-first flow:

  • Send a LinkedIn connection request immediately, then wait three days.
  • If the person accepts within seven days, wait a couple of days, send a LinkedIn message, wait a couple more days, and send another. If there is still no reply, switch to email.
  • If the connection request is not accepted within a week, skip straight to email.

Where does the human review step fit?

Justin is firm that this is human in the loop, not human out of the loop, and the live run showed why. After Claude loaded the first pass, he went into Lemlist and read through the copy. It did not match the job requirements closely enough. So he told Claude to go back, read the job postings, and tie the copy to them.

Claude stopped, rewrote five sample leads, then redid the copy for everyone else and reloaded it. He never edited a message by hand. He reviewed, gave feedback, and let Claude redo the work. In the end, 71 contacts sat in Lemlist ready to launch.

If you do want to tweak a single message, Lemlist lets you click into any lead, hit personalize, and edit it inline.

What does good copy look like?

The messages Justin showed tie a specific line from the job posting to something he has built. One LinkedIn note referenced a GTM engineer posting that put Salesforce CPQ and quote-to-order next to AI-native playbooks, said he builds that plumbing on contract, and requested a connection. The follow-up got more specific about where the role sat in RevOps, then offered to send a short write-up of what he would ship first, with a public worked example.

That offer is the point. Justin calls it permissionless value: give the prospect something useful before asking for anything. In his case, that means GitHub links. He has found that prospects often look at the repo, or have Claude or ChatGPT review it, before ever replying. By the time someone responds, they have already seen his work and are a much more qualified lead.

His view on the alternative: a demo request as the first touch works worse. Offering something free with no commitment gets a much better response.

What happens after someone replies?

This part is optional, but it closes the loop. Justin's AI assistant in ChatGPT, which he named Ollie, checks for email and LinkedIn replies every hour. When someone responds, Ollie pings him in Slack, including on his phone, with a drafted reply. He approves it, asks for changes, or tells Ollie he will handle that one himself.

Does signal-based outbound actually work?

Justin was clear that he is not running this campaign himself right now, because he does not have capacity for new clients. He built it for the demo. But he has data from clients who use the flow, and he shared it in broad strokes:

Channel What Justin sees across clients
Cold email Above industry standard, but still cold email. He has not seen anything much better than industry standard
LinkedIn Roughly 150% of industry standard, across many verticals and personas

His read is that email as a channel is weak now. People block it, filter it, and ignore it. He would rather teams stop over-engineering email and put that effort into LinkedIn and owned content. The gains come from doing the work of matching the message to the signal and to the persona.

How do you stay safe on LinkedIn?

LinkedIn is cracking down on automation, and Justin's take is that most of the crackdown hits people who over-automate and send programmatic-looking messages at an annoying volume. Lemlist handles throttling for you, limiting connection requests and messages per day based on your account type. As a rule of thumb, he considers around 20 messages in 24 hours safe.

The catch is that LinkedIn is the better channel, so people want to scale it, and that is exactly where the limits bite. Scale comes from how many accounts you can connect and what each one allows. When a reply needs a message beyond the daily cap, Justin sends it himself rather than waiting on the automation.

Can you track results without leaving Claude?

Yes. The plugin includes a slash command that builds a campaign tracker as a Claude artifact. It queries Lemlist through the MCP connector, and you refresh it daily to regenerate the dashboard. You can break results down by rep or by BDR, add calling data (Lemlist does calling too, or you can connect a dialer such as Aircall), and share the artifact with your team. The one in the demo used synthetic data.

In this setup, Lemlist is a dumb pipe: it sends, records the data, and acts as the source of truth. Claude does all the thinking, pushing copy in and pulling data out.

For companies with a tighter security posture, Justin builds the same thing in Google Cloud. The Lemlist API key lives in Secret Manager, a nightly Cloud Run job pulls the data into a table, and that table powers a dashboard people sign into with their Google login, controlled through IAM.

Can you run this through Clay instead of Lemlist?

Partly. Clay does not let Claude write directly into Clay tables from your console. It wants you to use a Claygent inside Clay. You could have a browser agent paste the copy into the Clay table for you, though Justin was not sure whether that violates Clay's terms.

The better fit is Clay at the front of the workflow. Set up a signal table in Clay, and when a signal fires (a job posting, a website visitor, any of the many signals Clay offers), Claude uses the Clay MCP tools to enrich the lead with email and LinkedIn, then pushes it straight into the Lemlist campaign. You can schedule that to run daily so new signals flow in automatically.

The data provider is swappable in the other direction too. Justin used only a narrow slice of SalesIntel for the demo. It also tracks hiring trends and champion moves, and it has its own drag-and-drop workflows for reaching the buying committee, so a team that wants one platform can run the outbound there.

Why put the human back in the AI SDR loop?

This was Justin's soapbox moment. In what he calls AI SDR 1.0, teams wanted to over-automate, hand AI the entire outbound motion, and remove the human completely. The results were poor, and he sees the big AI SDR companies dialing that back. In AI SDR 2.0, the human comes back in where it makes sense.

His argument: if the signals are gathered, the copy is written, and the tooling is set up, about 80 to 90% of the work is done. An SDR can at least check that the copy matches the signal and handle the follow-up. Humans need to own the outcome, teams need training to do it, and he does not think people enjoy dealing with AI all the time anyway. An SDR quarterbacking this flow every day can cover a lot of ground.

There is an engineering case too. Justin has built fully automated versions using Anthropic's managed agents, giving an agent the same tools and letting it run everything. It works, but it takes far more configuration and setup. For this job, he does not think that extra build pays off. Augmenting a person with a lightweight plugin costs less and, in his experience, gets better results.

Key Takeaways for GTM Teams

  • Verify before you send. Let Claude check that each signal is real and relevant before any copy gets written.
  • Write one skill per persona. Leads without a matching persona should wait, not get a generic message. Treat the held list as a to-do list for new skills.
  • Put the copy on the contact, not in the sequence. Merge fields for every touch let one sequence deliver one-to-one copy.
  • Review in the sending tool, then give feedback to Claude. Read a sample, say what is off, and let Claude rewrite the batch.
  • Lead with free value. A useful asset beats a demo request as the first touch.
  • Weight LinkedIn over cold email, and stay inside the daily limits.
  • Keep a human quarterbacking it. The tooling does 80 to 90% of the work. The rep owns the last mile.

How do you set this up yourself?

Get the build: Justin's Signal to Sequence repo has the full build guide, every prompt, CSV templates and eight persona copywriting skills to turn a buying-signal list into a 1:1 outbound campaign with Claude.

Justin's suggested path is to rehearse first. After installing, run setup, then try the demo, build sequence, build load, and review status steps with fake leads or contacts from your connected CRM before touching a live list. The repo's guide adds two practical guardrails: nothing sends until the launch prompt, which only fires when you type LAUNCH, and you should buy separate sending domains on day one, because a new mailbox needs about 40 days of warm-up plus 14 days of stabilization before its first cold send. If you plan to commit real contact lists or filled-in instructions, create a private copy from the template rather than a public fork.

Want to Attend the Next Session?

GTM Clawllective is a community where go-to-market operators show how they are actually using AI in the field, live and unscripted. The live discussion stays off the record, and every recap includes the speaker's demo video so you can watch the build yourself. Come see the next one live and join the community.


About the Speaker

Justin Parnell is the founder of Justin GPT, the business he started a year or two ago. He works as a forward deployed engineer, and his engagements put him alongside CMOs, CTOs, and CPOs. For this session he packaged the full play as a plugin and a public repo so anyone watching could set it up the same day.

Connect with Justin on LinkedIn, get the Signal to Sequence repo on GitHub, or learn more at JustinGPT.ai.

Watch the session replay: Signal-Based Outbound with Claude