The B2B Demand Gen Playbook Is Broken. AI Might Finally Fix It.
A dollar of new ARR costs 70% more than it did in 2021. The man who wrote the old playbook explains why — and what replaces it.
We’re living through the end of an era in B2B marketing.
The leads-and-MQLs playbook — content nurturing, gated assets, reverse-waterfall pipeline models — defined how a generation of marketers operated. It earned marketing its seat at the revenue table. And it’s now running out of road.
The numbers aren’t subtle. A dollar of new ARR costs 70% more to generate today than it did in 2021. Only half of teams are hitting their pipeline goals, even as budgets shrink. And the number of quality conversations per SDR has dropped 55% since 2014.
So what comes next?
That was the question at the center of my conversation this week with Jon Miller — the person who, more than almost anyone, built the playbook now under pressure. Jon co-founded Marketo and helped create the marketing automation category, then founded Engagio and built out account-based marketing before its acquisition by Demandbase, where he served as CMO. Today he’s building Phave, a stealth AI-native martech company aiming to reinvent B2B demand generation from first principles.
When the architect of the old model tells you it’s broken, it’s worth listening to why.
Here were my five biggest takeaways.
1. The Old Model Treated Marketing Like a Gumball Machine
“The traditional model treated marketing like a gumball machine — you put a dollar of campaign in, you get x number of leads out.”
The core flaw in the traditional playbook, Jon argues, is that it assumed buying is linear. Put budget in, get MQLs out, convert them down a predictable funnel. We built our pipeline models and budgets on a reverse waterfall that depended on that assumption.
But buying isn’t linear. It’s a complex, nonlinear system — six to sixteen people on a buying committee moving unpredictably, most of the journey happening anonymously, shaped by conversations and influences no attribution model will ever capture.
The data backs up the consequences: by the time a buyer contacts sales, 80% already have a preferred vendor, and they buy from that day-one shortlist 92% of the time. If you’re not on the shortlist before the process starts, no amount of demand gen will save the deal.
For Jon, that nonlinearity also means attribution is a fool’s errand — trying to pinpoint what caused a purchase is like trying to predict the weather six months out. The system is too chaotic to model that way.
Check out my full conversation with Jon on YouTube or wherever you get your podcasts.
2. We Optimized for Metrics That Trained Buyers to Avoid Us
“Almost everything we did to drive those metrics came at the expense of a good buyer experience. We trained the buyers to avoid us.”
This was the most uncomfortable admission in the conversation — and it came from the person who preached the metrics.
When the primary KPI became MQLs, it drove bad behavior: run more campaigns, send more emails, gate the content, sic an SDR on anyone who downloads an ebook. Each move generated the number. Each move also degraded the buyer experience.
Over time, that’s why buyers research anonymously and do their homework before they ever talk to sales. We optimized ourselves into irrelevance. As Jon put it: bombarding buyers and chasing every download isn’t building relationships — it’s burning them.
The hindsight is brutal because it’s so obvious. Friction designed to force a transaction was never going to build a relationship.
3. The Real Unlock Is the Shift from Rules to Reasoning
“Marketing automation platforms are all glorified rules-based systems. What AI enables that’s so magical is its ability to reason and think.”
Here’s where the AI conversation gets concrete — and why Jon believes this moment is different from every hype cycle before it.
Every legacy martech platform runs on deterministic rules: if they open the email, add a point; if they’re in financial services, send this. The problem is the real world is full of edge cases, so you write more and more rules until the system becomes brittle and unmaintainable. Anyone who’s seen a mature Marketo instance knows the mess. It’s why most companies have, at best, two or three nurture tracks.
AI changes the unit of work from a rule to reasoning. Instead of pre-scripting every path, an agent can look at everything it knows about a specific person and decide what to do now, tomorrow, and the day after. Jon’s analogy: a Spotify DJ building a personalized playlist — your journey, different from everyone else’s, learning as it goes. Send an email Thursday, no open; try Sunday, they respond; the system learns they engage on weekends.
That’s the 25-year promise of true one-to-one personalization finally meeting the technology that can deliver it.
4. AI Doesn’t Kill Martech — It Moves the Value to Decisioning and Context
“MarTech will survive as infrastructure. It’s just going to be decisioning, context, and governance. It’ll look pretty different than the past.”
Will martech even survive if AI becomes the interface where work gets done? Jon’s answer is more nuanced than “everything becomes a chatbot.”
He frames the future stack in three layers: data, decisioning, and delivery. The data layer gets more standardized over time (no marketer alive today says their data is great). The delivery layer fragments and becomes composable — new apps, new channels, vibe-coded tools all the time. But the value concentrates in the decisioning layer, which has two inseparable halves: the reasoning agents and the context that governs them.
That context is the part most marketers haven’t built yet. Jon breaks it into four kinds, borrowing language from AI tools:
A “marketing.md” file — your naming conventions, UTM rules, what “financial services” actually maps to in your messy CRM. The stuff a marketing ops person keeps in their head, written down for agents to use.
Marketing memory — the system learning what works and what doesn’t as campaigns run.
Hard rules — non-negotiables like consent and frequency caps, where violations carry real fines.
Voice and tone — how your brand actually sounds.
The takeaway for leaders: getting your context right now is what will make AI agents useful later. That’s a project you can start before you’ve picked a single new tool.
5. The Best Demand Gen Won’t Save a Weak Brand — Just Look at Anthropic
“When you have that kind of pull, you can have relatively simple demand machinery and it’ll work. The star has to be brand and product-market fit.”
It’s tempting to look at a company like Anthropic — adding billions in revenue with a tiny growth team — and conclude AI has made demand gen trivial. Jon thinks that reads the lesson backwards.
Anthropic isn’t growing because of a demand-gen machine. It’s growing because of extraordinary product-market fit and brand. The lesson isn’t “AI replaces demand gen” — it’s “when you have real pull, simple demand machinery works.” And for the 99% of companies without Anthropic-level luxury (i.e., all of us), that means brand and product-market fit have to be the star, not the afterthought. Even if your title is Head of Demand Gen, if you’re not thinking about brand and the 95% of buyers who aren’t yet in-market, you’re on a losing path.
This is also why, in a world where AI makes content production nearly free, the source of content matters more than the substance. Jon’s “holy trinity” for standing out in the age of AI slop: relationships (founder-led brand, partners, niche influencers), experiences (a great event can’t be summarized by an AI agent), and craftsmanship (visible evidence of real human work becomes a trust signal). His own LinkedIn cocktail videos underperform on raw impressions — yet they’re what everyone wants to talk to him about. The dashboard misses the brand impact entirely.
Closing Thoughts
My biggest takeaway from this conversation with Jon is this:
AI isn’t a faster version of the old playbook. It’s permission to throw the broken parts out.
The leads-and-MQLs machine was a product of its constraints — rules-based tools that forced us to treat nonlinear, deeply human buying as if it were a linear funnel. Those constraints are lifting. The shift from rules to reasoning makes true one-to-one finally possible, which means the marketers who win won’t be the ones who run the old plays faster. They’ll be the ones who rebuild from first principles: brand and product-market fit as the foundation, reasoning agents on top of well-governed context, and a relentless focus on the buyer experience the old model trained us to ignore.
Or, as Jon reminded me about the hardest part of this moment: change is hard, the goalposts move every six months, and staying grounded comes down to the same trinity that differentiates great brands — real relationships, your health, and finding joy in the work.
I’m co-hosting a webinar next month with the Contentstack team who will unveil the Agentic Enterprise Report — including findings from 700 enterprise AI leaders on the new CMO playbook. If you’re driving (or inheriting) an AI initiative in your organization, it’s a planning-season resource you don’t want to miss.
This is actually hopeful. One of the big flaws of b2b marketing was that followed the path of b2c. Theres definitely overlap but sales cycles are different and you have to navigate the different needs of gatekeeper, stakeholders, and decision maker. What those groups care about is often very different? What we forgot is that B2b is even more human and relational driven than b2c