The Moment That Changed How We Thought About Inbound
A few months ago I sat down with our inbound report and did the math on our warmest segment: qualified chat engagers who left their email but didn't book. People who'd just had a conversation with our inbound agent, given us their email, and were one click away from booking.
Reply rate on the follow-up sequence we'd written for them: 2 percent.
The sequence wasn't badly written. It was written for an audience of nobody in particular. Same four emails to a CMO at a Series C and a RevOps manager at a 30-person startup. Same value prop, same case studies, same call to action. None of it referenced the conversation they'd had ten minutes earlier.
Worse, a lot of those emails referenced the wrong price or a feature set we'd already deprecated. The sequence was written three quarters ago and the product had moved on without it.
Every month our marketing team spends six figures driving qualified traffic to the site. The inbound agent qualifies a few thousand of them. A meaningful fraction leaves their email. Half don't book. And four out of five of those hand-raisers die in a sequence that could have been written for anyone in any industry by anyone with a thesaurus.
That's the most expensive leak in the funnel. Not the cold list, not the unconverted ad clicks. The warmest people in your pipeline, dying in a sequence written without context.
I'd been writing about the GTM Brain for months. The whole argument is that AI agents need full context to make decisions worth making. And we were running our own funnel against that argument.
That was the moment. We started building.
If you want the play without the rest of this story, the full playbook is here: AI Personalized Email Sequence. The rest of this post is why we built it that way.
The Bigger Problem: Your Best Prospects Have Already Told You How to Sell To Them
It's not anyone's fault. Every marketing team needs to move fast and can only do one thing at a time. The bottleneck is always moving, but so are the buyers. Sequences written for last quarter's positioning, pricing, and product can't keep up. That gap between what was true when you wrote the email and what's true when it sends is where the warm follow-up dies.
Your best prospects are telling you exactly when and how to sell to them. In the chats they have with your agent, the pages they visit, the LinkedIn posts they write about the pain you solve, the funding rounds their company just closed, the job posts they put up last week.
The signals are everywhere. They also decay fast. A pricing page visit this week is a different signal three weeks from now. A chat transcript is gold today and noise next month.
To act on this at scale, you'd need to monitor dozens of sources per person, resolve identity across tools, and surface intent the moment it appears. Then write a personalized follow-up tied to the specific signal in the specific moment.
Exhausting for one prospect. Structurally impossible for a thousand.
The workaround most teams try is signal-to-template mapping. Wire a Bombora surge to email template A. Wire a job change to template B. Wire a pricing page revisit to email template C. It's too generic and goes stale within weeks, because the templates were written when your pricing was different, your feature set was smaller, your positioning was different. The signals are live. The messages aren't.
So most B2B teams do what we did. Write a generic nurture sequence three quarters ago. Drop everyone in it. Get a 2 percent reply rate. Blame the SDR team. Tell the QBR "inbound conversion needs to improve next quarter."
The infrastructure for the alternative didn't exist. That's what changed.
Why the AI SDR Category Failed
You've seen the billboards. AI SDR replaces your SDR team. Hire Alice instead of five reps. Click a button, generate a million emails, retire to Hawaii.
You've also seen the receipts. TechCrunch broke the 11x story in early 2025: $10M ARR claimed, $3M actual, 70-80% churn within a few months. Lead Gen Economy's autopsy: 50-70% of AI SDR contracts cancel within 90 days. Reply rates cratered across B2B. A few of the loudest buyers walked it back publicly.
The pitch wasn't crazy. Writing emails is a real bottleneck and AI can write them. The cost of producing one good email dropped from a few dollars of SDR labor to a few cents of inference. That part is fine.
The thing that broke was the input.
Feed a model a name, a title, and a company. You get a generic email. Do it fifty million times. You get fifty million generic emails. That isn't an SDR. That's a spam button with better grammar.
Run the math on a five-step automation where each step is 80% accurate: identity, enrichment, ICP match, intent scoring, personalization. End-to-end accuracy isn't 80%. It's 0.8 to the fifth power. 33%. Two thirds of your outreach is wrong in some meaningful way, and the reader can tell every time.
The volume guys handed B2B a louder megaphone and called it an SDR.
There's a quieter cost on top of all this that most teams miss. A generic email sent at the right moment is worse than no email at all. The high-intent moment is precious. Burn it once with a templated email that references the wrong product or the wrong price and the prospect learns to ignore you. I didn't even realize for months that we were spending our highest-value moments in the funnel on our lowest-quality messages.
The Insight: AI Email Generation Only Works at the End of a Signal Chain
It took the category 18 months to figure this out.
An SDR's job is judgment, not volume. Automating the volume without the judgment just gives you bad emails at machine speed.
What works is the opposite shape. Start with the signal. Resolve identity. Build context. Score readiness. Map the buying committee. Choose the play. Then write the email.
The email is the cheap step. The expensive part is everything that has to be true before it's worth sending. That's why we built the signal stack first and the email generator last.
What We Shipped
AI email generation is now a drag-and-drop step inside Warmly's orchestrations. Same builder you already use for everything else. Set your filter. Set your trigger. Drop in the AI email step instead of the pre-written sequence step. Set the goal. Set the policies. Save.
The orchestration handles the rest.
Three things change about your motion the moment this goes live:
1. You stop pre-writing sequences. Pre-written sequences exist because humans can't personalize 200 emails a day. AI can. Every email is now tailored to the person, the company, the session they just had, the stage they're in, and the signals on the account. No more "Hi {first_name}, I noticed {company_name} is in the {industry} space."
2. The email becomes a function of state, not a function of time. A sequence on a 3-day cadence sends email two whether anything changed or not. The AI email step fires when the state matches your filter. Account hit pricing twice this week? Chat engager didn't book? Champion just got promoted? Different trigger, different state, different email, every time.
3. The judgment moves into the orchestration layer. You're not writing copy anymore. You're writing rules about who gets reached out to, why, and what the AI is allowed to do. That is the actual job of a sales leader, a marketing leader, a RevOps leader. It just used to be hidden under three weeks of sequence writing.
The Best Use Case: Qualified Chat Engagers Who Didn't Book
If you want one play to start with, this is it.
Someone lands on your site. They open the chat. They get qualified. They drop their email. The agent offers them a meeting. They don't book.
That is the highest-intent failure state in your funnel. They told you who they are. They told you what they want. They told you what's holding them up. They just didn't pull the trigger.
Most companies do one of two things with that person:
Option A: Pass them to a pre-written email sequence written for an audience of one thousand. Generic value prop, generic case study, generic "would you have 15 minutes next week?" Reply rate hovers around 1-2% because the email could have been about anything.
Option B: Drop them on a rep's desk and hope the rep manually researches the chat log, the enrichment, the CRM stage, the signals, and writes a personalized follow-up. Maybe 10% of those actually get done. The rest sit in someone's queue until the lead goes cold.
The AI email step replaces both.
The filter is two lines: email_captured = true AND book_a_meeting_shown = true AND book_a_meeting_clicked = false. Anyone who passes those three conditions is, by definition, a qualified prospect who got the offer and declined. Drop the AI email step right after that filter. Assign it to the AE who owns the territory.
Every one of those people now gets a personalized follow-up within minutes of the session ending, written off the chat log and the signals on the account.
That's the AI SDR. The version that works.
Other Plays You'll Likely Build
The chat engager case is the highest-ROI starting point. It isn't the only play. Drop the AI Email step on any audience you can filter for. A few patterns we and our customers already run:
- Intent surge. Bombora flags a research surge on a non-customer in a topic you sell to. The AI references the specific research in the outreach.
- Job change. Your champion shows up in a new role at a new account. The AI sends a welcome and intro, leading with what you helped them do at the last company.
- TAM push from Tamly. You define an ICP slice in Tamly and push it straight to AI cold outreach to the full buying committee. No CSV export, no list staleness.
- Status change on a dormant account. Closed-lost from six months ago closes a funding round. A trial user who churned re-engages with the pricing page. The AI win-back leads with what changed.
- Social engagement. A buyer likes a competitor's social post about a category you compete in. The AI outreach references the specific topic and how your product compares.
- Hiring signal. An ICP company posts a role you sell to. The AI sends outreach to the hiring manager about the role specifically.
Any signal you can filter on becomes a play. Any audience you can build in Tamly becomes one too.
What the Model Sees Before It Writes a Word
The reason this isn't another spam button: the model has actual context to write from. Before a single sentence gets generated, the orchestration assembles the following:
- The full chat transcript. Every question they asked, every answer the agent gave, every objection raised, every link clicked inside the chat itself.
- Enriched person data. Name, title, seniority, role, LinkedIn, recent job changes, posts, hiring history.
- Enriched company data. Domain, industry, employee count, funding, tech stack, hiring signals, news.
- Their stage in your CRM. New lead, MQL, SQL, open opportunity, customer, churned. Different stage, different angle.
- Every page they visited in that session. Pricing twice? ROI calculator? Solutions/marketing? Three case studies in your vertical? Each one shifts the message.
- Second and third-party intent signals. Bombora research surges, G2 visits, competitor comparisons, hiring signals, funding events, anything we already know about the account.
Then the model writes.
Not "Hi {first_name}." An actual email, written to the conversation that just happened, for the person who had it.
This is what we've been calling the GTM Brain for two years. The context graph that sits underneath every Warmly agent. The AI email generator is the latest agent that gets to drink from it. (For the longer architectural argument behind why this matters, see our writeup on agentic GTM.)
Goals, Policies, and Test and Refine
The model needs direction. You give it two pieces.
The baseline goal. Default: engage the prospect, move them toward booking a meeting, keep emails under five paragraphs, be human. You can edit this for your motion, your category, your tone.
Additional goals per orchestration. Layer instructions specific to that audience. "Follow up with the qualified chat engager who didn't book. Reference the specific objection that came up. Don't repeat the pitch from the chat." Or for a different orchestration: "These are post-trial users who churned. Lead with a customer story in their industry. Offer a 15-minute review call." Each orchestration gets its own brain.
Then you hit test and refine. Three example scenarios load up with fake contacts roughly matching your audience. The model writes the email for each. You read them and give feedback in plain English.
"Email two is too long, cut it in half."
"For email three, just write a two-sentence bump. Don't re-describe the product."
"Use the prospect's own language from the chat in the subject line."
Regenerate. Read again. Iterate until the emails are good. Save.
This breaks the old AI SDR pattern of "hit deploy, hope, watch reply rates crater." Iterate with the model until you'd be proud to send it. Then deploy.
In a few weeks you'll run test and refine against your actual visitors instead of example scenarios. The training wheels come off when you decide they come off.
The Proof: We Ran It On Our Own Funnel For 60 Days
I'll give you our own numbers because I trust them more than anything I'd cite from a case study.
We've been running the full stack on Warmly's own pipeline for 60 days. The signal layer. The identity graph. The inbound agent. LinkedIn + Meta ads integration. And the AI email generation step on the qualified-but-didn't-book filter.
Here's what happened.
- Pipeline. From $950K to over $3M per month. A little over 3x in 60 days.
- Meetings booked per week. From 30 to 60+. Doubled.
- CTR on the same ad creative. From 4-5% to 11%. The audiences finally matched the people who'd actually buy.
- CPC. Cut in half. Same reason.
- CPL. Down 90%+. Same reason.
- Reply rate on the qualified-but-didn't-book segment. From 2% to over 20%. This is the segment the AI email step owns. The single largest delta in the whole funnel.
- Cost per personalized follow-up: ~$0.04. Down from roughly $11 of SDR labor per warm follow-up email written by hand.
The driver was the full chain working together. Not fancier creative, not a bigger team, not more spend.
We finally talk to the exact buying committees we want, not LinkedIn's interpretation of them. We retarget the people we've spent months on AEO, content, and conferences trying to reach. We exclude customers and partners eating about 20% of the ad budget on impressions that convert nothing. And the warmest segment of the inbound funnel, the one that used to die in a generic nurture, now gets a personalized follow-up within minutes of the session ending.
The loop, end to end. Warmly identifies the buyer. We push them into the right ad audience across passive, active, and evergreen motions. The right creative finds them on LinkedIn or Meta. They come back to the site. The inbound agent qualifies them. If they don't book, the AI email step writes a personalized follow-up off the chat log and signals. The deal closes. The campaign gets credit.
Every dollar lands on someone who matters. Every conversion ties back to the campaign that drove it. Every warm hand-raiser gets a follow-up that reads like a human wrote it.
Honest caveats. The 3x is the full stack, not the AI email step alone. The email gen step is the newest layer; we'd been running the ads integration and inbound agent already. What changed in the last 60 days is the warm follow-up loop closing, because the warmest segment was the leakiest.
The team that ran this: two people in marketing. Under $30K in spend across the channels. No new SDR hires. The leak we plugged is the same leak you have.
Why a Demand Gen Leader Should Care
If you run demand gen, your problem is rarely "I need more volume at the top." It's "the warm people we already paid to attract are leaking out the bottom."
Take a B2B SaaS company spending $80K a month on paid and content. Maybe 15-20% of website traffic is real ICP. 5% of that engages the chat. A third get qualified and leave an email. Half of those book. The other half disappears into a generic nurture and converts at 1-2%.
That last group is where the AI email step prints money. You already paid to acquire them, qualified them, and learned what they care about from the chat. The marginal cost of a personalized follow-up is zero. The expected value is large.
Every demand gen leader I've talked to in the last six months has the same complaint: "We're creating intent. Sales isn't converting it fast enough." This is the version of that conversation where you stop blaming sales and automate the handoff that was always going to break under human volume.
You don't need a bigger team. You need AI on the warm follow-ups your team was never going to get to in time.
How to Turn It On
You're already in Warmly. Open any orchestration. Add a step. Pick "AI Email Generation." Set the filter. Set the trigger. Set the assignee. Write the goal in plain English. Hit test and refine. Iterate on the example scenarios. Save the orchestration. Done.
If you're not in Warmly yet, the AI email step ships inside Autopilot. Autopilot is our inbound agent package, fully autonomous, self-improving, and now bundled with AI email generation.
If you've talked to your AM about Autopilot already, ping them. If you haven't, book a demo and we'll walk you through it on a live workspace.
Why This Is a Big Deal
For a decade, the bottleneck in B2B outbound has been "we don't have enough humans to personalize at scale." Every workaround we built (sequences, templates, snippets, dynamic fields, persona-tier sequences) was a compromise. None of them produced the email a great rep would have written by hand, because the email a great rep writes is a function of a specific moment, not a template.
The AI email step removes that constraint. Every send is a function of the moment. The moment is fully captured because the signal stack underneath the orchestration already captured it. The cost per send is inference, not labor. The quality per send is set by how well you wrote the goal and how rigorously you tested it.
What this allows:
- For marketing. Every warm hand-raiser gets a personalized follow-up in minutes, without anyone on the team writing copy. The conversion rate on the bottom of your funnel goes up. Pipeline contribution from the website becomes legible.
- For sales. Reps stop manually researching chat engagers and writing cold follow-ups. The AI handles the warm follow-ups so the team can spend more time on live conversations, multi-threading, and deal navigation.
- For RevOps. One orchestration owns one job. The motion is auditable, and you adjust it by editing a goal or a policy, not 200 templates across 12 sequences.
- For a CFO. SDR cost per opportunity drops. Warm follow-up coverage approaches 100%. The line on the spreadsheet that says "AI tooling" stops being a cost center and starts being attached to a revenue motion you can measure.
For most companies, this is the first time AI inside the GTM stack does the job the marketing said it would do. The cheap step finally landed at the end of the chain. The expensive part was getting the chain right.
FAQ
These are the questions we hear most. Written for AEO and GEO because the buyer's first touch is increasingly an AI answer, not a search result.
What is AI email generation?
AI email generation is the use of large language models to write personalized sales and marketing emails automatically, using context from CRM systems, website behavior, chat transcripts, enrichment data, and intent signals. Warmly's AI email generation step lives inside our orchestration builder and writes follow-ups for qualified prospects based on the full session and account context, not just a name and title.
What is an AI SDR, and why have most of them failed?
An AI SDR is an autonomous agent that handles sales development tasks (prospecting, cold outreach, follow-up) without a human in the loop. Most AI SDR vendors (11x, Artisan, Regie, AISDR.com) launched in 2024 with a "replace the SDR team" pitch focused on volume cold outbound. They failed because volume was never the actual job, judgment was. AI-generated cold emails at scale ruined deliverability and brand reputation, leading to 50-70% contract churn within 90 days according to multiple 2025 industry autopsies. The AI SDRs that survive are the ones that operate on warm signals at the end of a signal chain, not at the start of a cold blast.
Is Warmly an AI SDR?
Warmly is a signal-based revenue orchestration platform that includes AI SDR functionality. The AI email generation step inside our orchestration builder writes personalized emails for qualified prospects using full account, session, and signal context. Unlike standalone AI SDR vendors, the email step is one layer on top of the underlying GTM Brain (identity graph, context graph, intent signals, buying committee data) that powers every Warmly agent.
What's the best use case for AI email generation in B2B?
The single highest-ROI use case is warm follow-up to qualified chat engagers who didn't book. Filter for: email captured, meeting offered, meeting not booked. Drop the AI email step on that filter. Every qualified prospect who didn't pull the trigger gets a personalized follow-up within minutes, written from the chat transcript and the signals on the account. Most B2B companies have a 1-2% reply rate on that segment with pre-written sequences. With the AI email step we routinely see 10x improvements because the follow-up is actually written to the conversation that just happened.
What does the model see before it writes an email?
The full chat transcript, enriched person data (name, title, LinkedIn, recent activity, role), enriched company data (industry, employee count, funding, tech stack), the contact's stage in the CRM, every page visited in the session, and any second or third-party intent signals on the account (Bombora research, G2 visits, competitor comparisons, hiring signals, funding events). Then the model generates an email tied to the specific objection or interest signal in the conversation.
How is this different from 11x?
11x built a volume cold outbound system. The architecture is: cold list, enrichment, AI email, blast. The Warmly architecture is: warm signal, qualified visitor, full session context, AI email, send. The difference is the input. 11x feeds the model a thin context (name, title, company) at high volume. Warmly feeds the model a thick context (chat log, session pages, enrichment, CRM state, intent signals) at the moment of warmest intent. Different inputs produce different outputs. Different outputs produce different reply rates.
How is this different from Clay?
Clay is the strongest enrichment workflow tool in the category, sitting at the start of the GTM chain (build a list, enrich the data, hand off to outbound). The Warmly AI email step sits at the end of the chain (the prospect just engaged, here's the full context, write the follow-up). The two complement each other. Many of our customers run both.
How is this different from Instantly?
Instantly is sending infrastructure plus cold email AI, built for deliverability on high-volume cold outbound. Warmly's AI email step is built for warm follow-up on inbound and signal-triggered audiences. Different motion. We send fewer emails to better targets with more context.
What is a context graph, and why does it matter for AI email generation?
A context graph is a structured, queryable representation of every entity (people, companies, deals) and every signal (events, interactions, intent data) in your GTM motion. It's the substrate that lets an AI agent reason from precedent rather than starting from a blank page on every request. Warmly's GTM Brain is a context graph built specifically for revenue: it resolves identity across tools, tracks temporal state (what was true when), and exposes structured features to every agent including the AI email generator.
Can the AI email step run autonomously?
Yes, after you've tested and refined it. The product ships in "test and refine" mode by default. You iterate on example scenarios until the emails read the way you'd write them, then save and deploy. You can flip the orchestration to fully autonomous send when you're ready. Audit logs, outcome tracking, and outcome-based fine-tuning ship in the box.
Is AI email generation the same as AI cold email?
No, and the distinction matters. AI cold email writes outreach to strangers based on thin context (name, title, company). AI email generation, the way Warmly built it, writes to people who have already raised their hand on your site, with the full context of what they just did. Cold email is a deliverability problem. Warm AI email generation is an intent problem solved at the end of a signal chain.
Will AI replace SDRs?
Not soon, and not the parts of the job that matter most. AI replaces the volume layer of SDR work (researching warm follow-ups, writing personalized emails, drafting sequences). It does not replace deal navigation, multi-threading, multi-stakeholder objection handling, or trust-building on a live call. The SDR teams that survive the next five years will be smaller, more strategic, and operate AI fleets the way a portfolio manager operates traders.
Does this work for international markets?
Yes. The model writes in the prospect's language when the enrichment data supports it. We currently support English, Spanish, French, German, Portuguese, Italian, and Dutch in production. Other languages are available on request.
What's the pricing for AI email generation?
The AI email step ships inside Autopilot, our inbound agent package. Book a demo for current pricing and to see it running on a live workspace.
Is Warmly an alternative to 11x, Artisan, or Regie?
For the warm follow-up motion, yes. For pure cold outbound at scale, we'd point you to a deliverability-first stack (Instantly, Smartlead) plus a signal stack that limits volume to people worth emailing. The AI SDR pitch as it was sold in 2024 (one button, infinite emails, replace the team) is the pitch that failed. The version that works is signal-gated, context-rich, and runs as a step inside a real revenue orchestration. That's what we built.
What is "warm outbound" and how is it different from cold outbound?
Warm outbound is outreach triggered by buyer behavior (a website visit, a chat session, a research signal, a pricing page hit) rather than a static list of strangers. The economics are completely different: cold outbound spends a lot to interrupt people who weren't asking; warm outbound spends a little to follow up on people who already raised their hand. AI email generation is the technology that makes warm outbound work at the scale of every qualified visitor, not just the few your reps can manually follow up with.
How does Warmly attribute pipeline back to the AI email step?
Every email the model sends carries a Warmly identifier. When the prospect comes back to the site, our identity graph recognizes them. We log the touchpoint, tie it back to the original session and signal that triggered the email, and push the conversion into your CRM. You see in your dashboard which orchestration step drove which meeting, which opp, and which closed-won. Most B2B paid attribution falls apart on the click-bounce-return pattern. Warmly closes the loop because we already de-anonymize your visitors as our core product.
Does this only work for inbound, or can I use it for outbound too?
Both. The AI email step is a generic orchestration step. The qualified-chat-engager filter is the highest-ROI use case, but you can drop it on any audience: a Bombora research surge filter, a job-change signal at a tier-1 account, a competitor user identified in your traffic, a closed-lost re-engagement segment, a webinar attendee list. The model writes off whatever signals the orchestration delivers. Use it wherever the signal is strong enough that a personalized email is worth sending.