By Alan Zhao, Co-founder & Head of Marketing at Warmly
TLDR: We used AI marketing tools to go from less than $1M in pipeline in February 2026 to over $3.2M in March. Two people on marketing. Under $30K in B2B demand generation spend. Half the sales team we had a year ago. This is the full AI marketing automation playbook: every tool, every tactic, every number.
How We 3x'd Pipeline With AI Marketing Automation (The Starting Point)
February was rough. Sub-$1M pipeline. The market is getting tougher. Claude Code is making buyers think they can build everything themselves. Techie startups are harder to sell to than ever.
A year ago, in February 2025, we were spending $60K+ per month on B2B demand generation. We had a bigger sales team. We had a dedicated GTM engineer. We had a larger marketing team. And we were generating around $2.2M-$2.5M in pipeline.
Fast forward: in February 2026, we spent about $10K on demand gen with 10 salespeople and generated less than $1M in pipeline. In March, we spent under $30K with 11 salespeople and generated $3.2M. Our marketing team is me and Lina, our marketing manager. That's it. Two people.
I'm finding that everything a GTM engineer used to do, and more, can now be done by one person with the right AI marketing tools connected together. This is what agentic marketing looks like in practice.
Here's exactly how.
Step 1: Rebuild the Website as Your AI Marketing Platform Knowledge Base
This is where everything starts. Not with ads. Not with outbound. With your website.
Why? Because AEO (Answer Engine Optimization) is starting to scrape websites directly. ChatGPT, Perplexity, Google's AI Mode. They all prioritize pages that are close to your homepage. If a page is five clicks deep, you're telling Google it doesn't matter. So we restructured everything.
We created pages for every single thing Warmly does:
- Product pages: TAM Agent, Inbound Agent, orchestration, de-anonymization, every feature with its own dedicated page
- Integration pages: Every platform we connect with
- Versus pages: Warmly vs. Clay, vs. Qualified, vs. 6sense, vs. ZoomInfo, vs. Unify, vs. HubSpot. Each one custom-built with honest positioning
- Use case pages: Account-based marketing, signal-based orchestration, sales automation, AI marketing
- Persona pages: Rev ops, sales, marketing
- Segment pages: SMB, mid-market, enterprise
- Data layer pages: Contact database, intent signals, 220M+ contacts
- Services pages: Forward deploy motion, CSM support
- GTM brand page: Our manifesto on how we think about go-to-market
Every page answers the questions that LLMs want answered:
- What exactly does this feature do?
- Who does this serve?
- Why is this important?
- How do you get value from it?
- What's the summary?
- What are the pros and cons versus competitors?
- What makes this unique?
We added examples showing how each feature works. We added FAQ sections at the bottom of every page. We made everything explicit. No vague marketing speak.
The graphics were the only thing we couldn't automate. AI image generation still isn't good enough to showcase exactly how a product works at the standard we wanted. Our designer created those by hand. Everything else was built programmatically through Claude Code.
★ Insight ─────────────────────────────────────
When you force yourself to create all these pages, you're building a comprehensive understanding of your entire product, market positioning, and competitive landscape. Claude Code stores this context. It becomes the foundation for everything else you do in marketing. Your website becomes the nexus of your product offering, and that context gets infused into every ad, every email, every piece of content you create after.
─────────────────────────────────────────────────
We saved the entire sitemap as a Claude MD file. Each page got its own .md file. When Claude Code reads through that folder inside its 1M token context window, it understands your entire business.
Step 2: SEO, AEO, and GEO. AI Powered Marketing Starts With Being Findable
Once the pages existed, we had to make sure they were actually optimized for both traditional search and AI search engines.
We connected Claude Code to:
- Google Search Console (via service account) to see all traffic, where it's coming from, broken pages, which keywords are hot
- Google Analytics to see which pages get traffic, how long people stay, general traffic patterns
- SE Ranking for keyword tracking and competitive analysis
What Google Search Console tells you is gold. You can see exactly what people search to find your website, the impressions, the clicks, the trends over time. For us, a lot of searches are competitor-related: "ZoomInfo pricing," "Apollo alternatives." People are actively searching for this stuff, so we created pages to capture it.
We try to publish about 7 articles a week. But only on topics we actually have authority on. Google's recent algorithm updates punish you for writing about stuff you have no business talking about, and if you post too much low-quality content, you get docked for spam. So every article has to be something our ICP would actually find useful or interesting. We put the goods up front at the top for people who are searching. That performs the best.
Our Drift shutting down post is a good example. It was timely, relevant to our space, and genuinely useful for people trying to figure out what to do next. That kind of content ranks because it deserves to.
Drift shutting down also gave us a huge outbound momentum boost. We built an entire campaign around it. LinkedIn posts about the shutdown got turned into LinkedIn thought leadership ads. We pushed email sequences on the sales side to Drift prospects about the shutdown and how Warmly can help them migrate. Same messaging went into Meta ads and Google search ads. When something that big happens in your space, you go all-in across every channel simultaneously. Blog post, social, ads, outbound sequences. That coordinated push was one of the biggest pipeline drivers in March.
What We Learned About SEO in the AI Era
I had a conversation with John Ozuysal from Houses of Growth that completely changed how I think about this. Some key takeaways:
Don't churn net new content. If you already have a solid content library (we do), creating more mediocre content actually weighs down your domain. Especially after Google's recent spam updates, companies that churn posts got wiped. Instead, optimize what you have.
Don't touch titles. The title is one of the main signals to search engines about what your page is about. Moving a keyword from the beginning to the end of a title kills your entity salience score. The only exception: updating a year ("2025" → "2026") or adding one more item to a listicle.
The top 20-25% of your page is everything. AI crawlers cite the first quarter of a page most often. Don't waste that space with storytelling intros. If the H2 asks "What are the best website visitor identification tools?" start the answer with "The best website visitor identification tools are..." Not "Are you wondering who visits your website? Imagine if you could..." Go straight to the answer.
Add TLDRs after your intro. Three bullet points summarizing the key answer. Crawlers love this.
Short intros, NLP-friendly writing. If there's a question, answer it immediately. Keep relevant entities together. This is the foundation of writing that search engines and AI models can actually parse.
Freshness matters more than ever. AI search prioritizes fresh content. Some of our most competitive articles get updated monthly. New information, new infographics, new data points. Not just changing the publish date. That's a recipe for getting penalized.
Use Google's AI Mode to understand the buyer journey. Search your target terms in AI Mode. Look at the follow-up questions it suggests. Those become FAQ sections in your existing articles. You're covering the entire buyer journey without creating new posts.
The Mention Strategy (This Is the Highest ROI Play)
For AEO and GEO visibility, creating content isn't the fastest path. Getting mentioned is.
Here's the play: Search your target prompt in ChatGPT or Google AI Mode. Look at the sources on the right side. Those are the articles feeding the AI's response. Reach out to those publishers. Ask to get mentioned. Pay $200 if you have to.
But be specific:
- Ask for top 3 placement. AI crawlers prioritize the first part of the page. Position 7 barely gets cited.
- Write your own description. Don't let them describe you. Control the narrative. "Warmly is an agentic GTM platform that..." Include your use cases, pricing, ICP. Otherwise AI might tell prospects something wrong about you and you'll lose deals before you even know it.
- Vary your anchor text. Don't use "website visitor identification software" every time. Mix it up: "website visitor tracking," "visitor identification app," "de-anonymization platform." Same anchor text repeated looks like spam.
- Check existing mentions. Go read every page where you're mentioned. Is the pricing current? Are the use cases accurate? Fix anything that's wrong.
Mentions give you backlinks too. And you can request they link to a specific page, like your most important blog post, to strengthen it directly.
Internal Linking
Your most important pages should be one click from your homepage. If it's six clicks deep, you're telling Google it's not important. We put competitor comparison pages in the footer for exactly this reason. They get link juice from the homepage.
Also: find your pages with the most traffic and backlinks. Link from those pages to your bottom-of-funnel content. You're passing authority without buying a single backlink.
Step 3: AI Marketing Automation for Google Ads
We spent about $4,000-$5,000 on Google Ads last month. Not a huge budget. But it's incredibly efficient when you use AI marketing tools with full product context.
Because Claude Code already knows our entire product, our competitors, our versus pages, and our positioning, it generates high-converting ad copy naturally. This is AI for marketing at its most practical. No need for image creatives here. It's all text-based.
We connected Claude Code to the Google Ads API via a service account. It can:
- Create campaigns, ad groups, and keywords
- Analyze which keywords drive the most clicks at the best cost per click
- Kill underperforming campaigns
- Add new keywords based on Search Console data
- Project ad spend for the month
- Suggest optimizations daily
We also connected Google Tag Manager so Claude Code can fix conversion tracking automatically. It creates the script tags for Google, Meta, and LinkedIn conversion tracking. Then it uses Playwright MCP to simulate a real user clicking through the site, filling out forms, to verify that conversion events fire correctly.
This whole loop. Analyze performance → optimize campaigns → verify tracking. Runs every day. Automatically.
Step 4: AI-Powered Account Based Marketing. Build Your Target List
This is where the agentic marketing system really kicks in.
Sales and marketing sit down together and build a target account list. This is where AI for marketing meets account based marketing. About 5,000 companies that:
- Use tech stacks that indicate they're a good fit
- Are competitors' customers (Drift, Qualified, 6sense users)
- Get a lead score and rating
- Get territory-mapped to specific AEs and SDRs
For each of those 5,000 companies, we identify roughly 5 people in the buying committee. That's 25,000 contacts.
We use Warmly's TAM Agent (our own beta product) to:
- Generate the buying committee contacts
- Get business email, LinkedIn profile, and phone number
- Waterfall through multiple enrichment vendors (Clearbit, Apollo, People Data Labs, and others) to get the most up-to-date information
We have 220 million contacts in our database. The enrichment is real. This isn't just company-level data. It's person-level, with verified contact info.
Step 5: AI Marketing Tools for Multi-Channel ABM
We run two types of ad audiences simultaneously.
ABM Audiences (Signal-Driven)
Our TAM Agent continuously ingests 150+ signals. Website visits, 10-K/10-Q filings, job changes, job openings, Bombora research intent, person-level site visits. When a company exhibits a signal, the agent:
- Finds the buying committee
- Checks if any member has high intent surge
- Pushes them into ad audiences via API (LinkedIn, YouTube, Meta)
- Adds the highest-intent contacts to email sequences (via Outreach) and LinkedIn sequences (via HeyReach)
The decision logic matters here. Highest intent + best fit = email + LinkedIn + ads. High fit but low intent = ads only. Not a good fit = nothing. Don't waste resources.
We push contacts directly into ad audiences via each platform's API. For LinkedIn, match rates stay above 90% because we're uploading verified email + company data. For Meta, it's email + first name + last name + location (less precise, hence lookalike audiences).
Exclusion lists are just as important as inclusion lists. Current customers get excluded immediately. Active deals get excluded. Bad titles (interns, students) get excluded from LinkedIn targeting. We constantly prune. One of the biggest time investments was getting the exclusion layer right.
These audiences get refreshed automatically. Every time a new signal comes in and qualifies a company, the contact gets added to the appropriate audience. Every time someone becomes a customer or active deal, they get removed.
Evergreen Audiences (Always-On)
Separately, we run:
- LinkedIn: Demographic targeting. Right company size, right job title, right role, right country. LinkedIn has the best B2B targeting filters.
- Meta: Lookalike audiences built from our best customers. Upload your customer list, let Meta find similar people.
These run continuously as top-of-funnel and mid-funnel awareness. The ABM layer handles bottom-of-funnel precision.
Ad Creatives That Actually Work
For YouTube: Horizontal video (16:9) is required, so we put our best performing, highest quality videos in there. We upload contact audiences directly via the Google Ads API. Match rates are lower than LinkedIn, but it's still another touchpoint. Every impression across another channel compounds.
For Meta: Image ads perform best. Our designer created the same graphic illustrations used on our website, retooled into ad formats. Square (1:1) dimensions work across both Instagram feeds and can auto-adjust to 9:16 for Stories. One creative, multiple placements.
For LinkedIn: Thought leadership ads. We post on LinkedIn every day. We know which posts perform well. The winners (high engagement, lots of comments) get recycled into promoted thought leadership ads. They have built-in social proof and we already know they resonate.
For LinkedIn DMs: This one surprised us. Joe, our Head of RevOps, was manually sending LinkedIn messages offering a free AirPod to ICP contacts who'd book a meeting. It was booking 76 meetings a month. We scaled it into LinkedIn Sponsored Message ads from our CEO with the same offer. For bottom-of-funnel contacts who are 100% ICP fit, it's incredibly cost-effective.
Results: Click-through rates consistently above 10%. Cost per click between $1-$2. Cost per lead (meeting booked) under $200.
All of this. Campaign creation, audience uploads, creative rotation, exclusion list management. Can be done programmatically via API through Claude Code. It'll create the campaigns, create the ad groups, upload the PNG images, generate the copy. You just review and approve.
Step 6: AI SDR and Outbound at Scale With the TAM Agent
Our TAM Agent is the core of the AI outbound sales motion. It's not just a sequencing tool. It's a full agentic marketing system with:
- A knowledge base: Everything about our product, positioning, competitive landscape (built from the website work in Step 1)
- A policy layer: Rules about who gets what type of outreach, when, and through which channel
- Trust gates: An approval layer for emails before they go out. Humans review. The agent doesn't go rogue
The agent decides what action to take for every contact based on signal strength, fit score, and channel capacity constraints:
- High intent + high fit: Email sequence (via Outreach) + LinkedIn sequence (via HeyReach) + ads
- Low intent + high fit: Ads only (for now). Email and LinkedIn when intent spikes
- Low fit: Nothing. Don't waste the budget
We have limited sends. LinkedIn messages are the scarcest resource. Emails are more abundant but still finite. Ads can reach everyone. The AI SDR optimizes allocation automatically across every channel.
We also went beyond signal based marketing. We mapped out entire TAMs of 100,000+ accounts. Every Drift user, every Qualified user, every 6sense user. Built buying committees for all of them. Auto-generated personalized emails. Pushed them through the system.
Sometimes you don't wait for signals. You force pipeline through.
Step 7: AI for Marketing Content. Post Every Single Day
We have a mandate: the entire GTM team posts on LinkedIn every day.
- Co-founders post daily
- AEs and SDRs post regularly
- Leadership team posts daily
- Even engineering posts (our Head of Engineering got 100K+ views on a post about how we 3x'd engineering velocity)
Every week has a theme. One week it's the Inbound Agent launch. Next week it's the Sendoso integration. The whole team coordinates messaging around that theme so our ICP hears it from multiple angles.
Content types that work:
- Educational how-to content (highest engagement)
- Thought leadership about the GTM space (builds authority)
- Video content (strong hook in first 3 seconds, text overlay, less about the company and more about a relevant trend)
- Funny/cultural content (shows personality)
- Product demos and releases (coordinated launches)
What doesn't work: self-promotional photos. "We're excited to announce." Generic corporate content.
Coordinated Launches
For big releases, we activate our network. About 100 influencers and friends drop comments in the first 10 minutes. Comments in the first hour are the most powerful signal to LinkedIn's algorithm. Get over 100 comments and the post takes off organically.
We do one or two of these big coordinated launches per month. The most successful posts become. You guessed it. Thought leadership ads that get recycled into our always-on campaign.
Step 8: AI for Marketing Product Launches on Autopilot
This is where the full system comes together.
When engineering ships a new feature, the PR and release notes are written by agents. They get posted to a Slack channel with:
- The use case
- What the feature includes
- How it works
- Who it's for
- How to implement and onboard
Our Head of Product creates a Loom walkthrough video.
From there, Claude Code takes over:
- Reads the Slack post via Slack MCP
- Generates a Playbooks page using our Webflow API token and existing template
- Uploads the video to Wistia via Wistia API, gets back an embed link
- Embeds the video on the Webflow Playbooks page
- Generates a Customer.io email with the video thumbnail, link to the Playbooks page, and proper UTM parameters
- Sends to our list of 15-20K users who have used Warmly at some point (free or paid)
The UTM parameters include a special w_email= parameter that passes the recipient's email. When they click through to our site, Warmly de-anonymizes them instantly. That data feeds back into the entire system.
One Slack post → website page + video hosting + email newsletter. Automatically.
Step 9: The Daily AI Marketing Analytics Loop
Every day, we ask Claude Code to analyze:
- Google Ads performance
- Meta Ads performance
- LinkedIn Ads performance
- YouTube Ads performance
- Google Search Console rankings
- Google Analytics traffic patterns
- Warmly session data (de-anonymized visitors)
- HubSpot CRM pipeline and conversion data
- Blog post SEO health
AI marketing analytics means looking at everything together. Not in silos. It finds:
- Where the biggest spend gaps are
- Where we're wasting money
- Which campaigns to kill
- Which keywords to add
- What the session-to-meeting conversion rate looks like
- Where people are getting stuck in the funnel
- What's driving actual pipeline vs. vanity metrics
Then it fixes things. Programmatically. Kill an ad, add keywords, upload a new creative, update a blog post's FAQ section. No manual work. Just review the changes.
Step 10: De-anonymization Closes the B2B Demand Generation Loop
Warmly's core product is the glue. Someone visits our website from any channel (ads, newsletter, ChatGPT referral, organic search) and we know who they are. Not just the company. The person.
If they're ICP and they haven't been added to an ad audience, they get added automatically. They get a follow-up email. A LinkedIn message. And ads across YouTube, LinkedIn, and Meta.
The system knows if they came from Google, from a newsletter, from an AI chatbot referral. It knows which pages they visited, how long they stayed, whether there's a surge of visits from their company. It filters out bot traffic. It overlays demographic data on activity data.
That intelligence feeds back into every decision the TAM Agent makes.
The Results
|
Feb 2025 |
Feb 2026 |
Mar 2026 |
| Pipeline |
~$2.2M |
<$1M |
$3.2M |
| Demand gen spend |
$60K+ |
~$10K |
<$30K |
| Marketing team |
3-4 people |
2 people |
2 people |
| Sales headcount |
20+ |
10 |
11 |
We've never been more efficient. Half the spend. Half the team. 3x the output.
The breakdown:
- Ads (LinkedIn, Meta, YouTube, Google): Always-on awareness + precision ABM targeting
- Outbound (email via Outreach, LinkedIn via HeyReach): Signal-driven + forced TAM outreach
- Inbound (website, SEO, AEO, content): Optimized for both human and AI search
- Product marketing (automated Playbooks + email): Every release drives re-engagement
- De-anonymization: Closes the loop on every channel
What AI Marketing Tools Do You Need to Do This?
Here's the actual stack for building an agentic marketing OS:
- Claude Code with Wispr Flow (voice-to-code), WozCode, and MCP connections
- API access to: Google Ads, Google Analytics, Google Search Console, Google Tag Manager, LinkedIn Ads, Meta Ads, Webflow, Wistia, Customer.io, HubSpot, Outreach, HeyReach
- A Claude folder on your desktop that stores your market understanding, website content, competitive analysis, and positioning docs. Claude reads through this folder to understand your entire business
- Warmly for de-anonymization, signals, orchestration, and contact data
- A designer for ad creatives and product illustrations (this is the one thing AI still can't do well enough)
- A team willing to post daily on LinkedIn
Save your API credentials and endpoint references in Claude MD files. That way Claude Code knows how to access every tool through agentic search. It'll figure out which APIs to call based on the problem you're trying to solve.
Is AI Marketing Automation Actually Easy?
No. The initial setup takes real effort. Connecting all the APIs, building the exclusion logic, getting ad creative right, training the TAM Agent's policy layer.
And not everything is automated. The designer still makes graphics by hand. The team still has to post content daily. Trust gates mean humans review outbound before it sends. John still has to audit our SEO because his eyes are better than mine for that stuff.
But the system compounds. Every day it runs, it gets smarter. More signals. Better targeting. Tighter exclusions. More context about what's working.
We used to need a dedicated GTM engineer to wire all this together. Now one person can do it. We used to spend $60K+ a month on demand gen. Now we spend under $30K. We used to have double the sales team. Now the agents handle the volume that humans used to.
We went from sub-$1M to $3.2M in 30 days. Not because we found some magic trick. Because we connected everything together and stopped leaving pipeline on the table.
The AI marketing tools exist. The APIs are there. The question is whether you're willing to wire it all up and let it run.
Last Updated: April 2026