Readme

I’m Cole D’Ambra, Head of Growth and Demand Gen at Plain. Before that: Apollo.io, Ashby, Autodesk, HackerOne. For a decade I’ve treated demand gen as a building discipline. Ship systems that turn attention into pipeline, get judged on the revenue.

I think of that time in three eras. The question that separates them isn’t the tools. It’s who writes the playbook.

DG 1.0 (2010–2018): The MQL era

Marketing’s job was generating MQLs and arguing with sales about whether they were any good.

PMM produced personas (Sales Leader Sally, CFO Christopher). Demand gen built campaigns around them. We measured success in MQLs, let sales tell us they were trash, and wrote SLAs about response time so we could fight about it more efficiently.

The industry said it took eight touches to get a meeting. We built twelve-step Outreach sequences.

It worked, sort of. Attention was cheap, buyers tolerated form gates, and they still expected an AE to educate them. The playbook came from Marketo and Engagio, and your job was to run it well.

Built in it: the machinery. Lifecycle and automation at Autodesk Construction Cloud, SMB and mid-market demand gen at HackerOne. I learned the clockwork well enough to know exactly where it leaks.

DG 2.0 (2019–2024): The signal era

Chris Walker started recording State of Demand Gen and said out loud what everyone suspected: last-touch attribution was a lie, the funnel wasn’t a funnel, and “how did you hear about us?” was telling a different story than the HubSpot report.

The dark funnel became gospel. And it was correct. Buyers stopped wanting four AEs to explain the category. They asked peers. They read LinkedIn. They did their own research.

6sense, Clay, Common Room, Demandbase, Terminus, G2, Bombora: a whole category showed up to chase signals instead of leads. Outbound stopped being volume and started being signal. CRM migrations, headcount triggers, stack changes, and, say it with me, website deanonymization.

ABM stopped being “ten target accounts, a Salesforce report, and a dinner.” Forrester counted 27 buyer interactions per purchase in 2021, up from 17 two years earlier. The teams that read the room shifted. The ones still shoveling MQLs got slowly bypassed.

But the playbook still came from a vendor. 6sense told you what a signal was. Bombora told you who was in-market. You rented the intelligence and ran the motion.

Built in it: the signal layer, before I had a name for it. First marketing hire at Ashby, out of stealth, $1M to $7M ARR. The custom outbound engine keyed on two signals we could see from the outside: which ATS an account already ran, and what they were hiring for. Every email ended with a plant I’d picked out for them and a photo of me holding my cat, Jimmy, who chews on it. 8% reply rate, ~$10M pipeline in the first six months.

Then Apollo, $50M to $140M, where I built product-led sales: an SDR signs up, we invite their teammates (8% conversion, ~2,500 extra signups a month). Three ICs activate, we reach their manager with the team’s real usage. A buyer builds sequences, an AE reviews them on video. About a thousand demos a quarter.

The signal playbook commoditized the moment everyone had the same data and the same outbound tools.

Apollo gives you unlimited contacts for $99 a month. Bombora comes free with your oil change. Every team has a Clay table, an AI SDR, a website pixel, and the “industry-leading proprietary reverse-IP lookup database,” which means every prospect gets the same “noticed you’re hiring SDRs” email from six vendors on Tuesday at 10AM.

The inbox noticed. Average cold email reply rates fell from about 5% in 2024 to under 1.5% in 2026, and the teams still clearing double digits are the ones who stopped automating the message and started automating the research.

The floor is covered in instruments. AI SDRs sending mail nobody reads. Dashboards nobody trusts. Intent data stacked on intent data until the signal becomes its own genre of noise.

The $40K walled garden

Here’s the pattern I see most. A team pays $40K a year for an all-in-one signal platform. It doesn’t work. They say, “we didn’t set it up right.”

It isn’t a setup problem. It’s a walled garden of someone else’s signals, someone else’s contact data, an AI writer you can’t tune, and deliverability you don’t control. It doesn’t fit how you sell because it wasn’t built for how you sell. It was built for how ten thousand companies sell on average, and nobody sells on average.

Meanwhile, the buyer moved

While marketing was arguing about which platform to rent, the buyer changed how they buy.

They start in a language model. Take a peek at the pricing page. They see the CEO post on LinkedIn or catch a video ad. Then they validate: the typical B2B decision now runs through 13 people inside the company and 9 outside it, and what actually triggers a conversation with a vendor is more often an industry expert or a peer than anything the AI said. Maybe they see a billboard. Then, when they can, they trial before they buy.

LLM, then people, then product. If we’re lucky, some paid media made it to their eyes along the way. That is the journey now. No one channel is king, and no vendor’s playbook was written for it.

What’s working

A few teams stopped thinking in funnels and started thinking in systems: programmable, instrumented end to end. They look less like marketing departments and more like product surfaces. They’re small. They ship. They measure how fast they learn.

That’s the era I’m building in now.

The next era doesn’t replace marketing with AI. It replaces the operating model.

For fifteen years the model was: buy the platform, run its playbook, report its dashboard. Marketo told you how to nurture. 6sense told you what a signal was. The new model is three moves, and you make all of them yourself with primitives to move faster, on your own data, for your own buyer.

1. Learn from everything

Every sales call, every outbound reply, every page view, every “how did you hear about us?”, every citation in an LLM answer is data you already own. Most teams let it sit in six tools that don’t talk. The first move is putting it in one place and treating it as the company’s memory.

At Plain that’s a set of tables that hold call transcripts, company data, citation tracking, and the full path a buyer takes through the site. When a company reads a comparison page after asking Claude about a competitor, we know which page, which search brought them there, and which link they clicked on the way in. The next message they get from us is about that, not about us.

2. Run experiments on all of it

Tracking used to be the expensive part, so teams ran one test a quarter and called it a program. AI builds the tracking now. That means everything gets a baseline and a measurement window, including the things nobody used to measure.

AEO is the worked example. When I joined Plain, none of the models had heard of us, and the honest read was that our buyer had already started asking them. So I made a bet that the models were a channel, not a fad, and the only way to find out was to treat it as an experiment: give every content change a baseline, watch what the models cite over the following month, and change the plan based on what actually moved. The strategy came from studying thousands of pages the models were choosing, not from what I assumed they’d want.

Three quarters in, it’s about half of Plain’s inbound pipeline and the majority of self-serve revenue, and the buyers who arrive that way come in already told we’re the answer. Hundreds of prompts tracked daily, technical fixes running themselves through an MCP, and a program that gets a little smarter every cycle. That’s what “learning velocity” looks like when you write it down.

3. Choreograph the channels

The buyer’s journey is LLM, then people, then product. An expensive wildcard. So that’s the score.

Be in the answer when they ask the model. Be in the room when they check with their peers. Be on the timeline they scroll in the morning. Be the easiest trial to start when they want proof. In practice: a buyer reads a comparison in Gemini, sees a question in a Slack group, gets an email that already knows what they’re evaluating, and meets us at a dinner where three of their peers are already customers. AEO, signal-based outbound, paid, content, events, all playing the same score, in the order the buyer actually moves.

None of it is an accident, and none of it came out of a vendor’s playbook. The goal isn’t awareness or even trust. It’s a buyer who’s excited to talk to you before you’ve said a word.

The rule underneath

Buy the primitives. Build everything that touches your buyer.

Data providers, sequencers, ad platforms, models: rent them, they’re commodities. The signal logic, the message, the timing, the measurement, the choreography: those are yours, because they’re the only parts that know your buyer, and they’re the parts a $40K platform can’t sell you.

The team that ships it

Smaller, more technical, shipping systems instead of campaigns. A handful of operators who can hold the whole loop in their head, research to render to outcome. Pipeline becomes a lagging indicator. Learning velocity per FTE becomes the leading one.

The teams that will win are already doing some version of this. The teams that will lose are calling AEO a fad while doing all their own research in ChatGPT.

If you’re building this way, say hi.