CEOs Are Building GTM Plans in Claude. Here's the Bill Their Teams End Up Paying.

On a recent episode of growth path, Andy Culligan sat down with Janessa Drainville, Senior Demand Generation Content Specialist for Central AI at Atlassian, for a wider conversation on AI, marketing, and where human judgment still earns its keep. Drainville's side of it comes from two years inside a company shipping AI features at speed, watching what happens when those tools meet real, messy customer data.

One of the threads Andy pulled on from his own side of the business: a pattern he keeps running into with founders across the market. CEOs are opening Claude or ChatGPT, building full GTM plans, or in some cases entire CRM replacements, and handing them over to their teams with the quiet implication that headcount is now optional.

TL;DR: CEOs handing marketing teams an AI-generated GTM plan, or vibe-coding their way out of a CRM contract entirely, isn't a shortcut. It's a liability dressed up as innovation. The teams that win aren't fighting AI; they're the ones who can show the data-cleanup, the nuance, and the "here's what this actually costs" analysis that a chatbot can't produce. That's the entire case for Fractional Marketing done properly: expertise embedded, not replaced.

The "I Built It With Claude" Problem

Andy Culligan described the pattern directly from his own conversations with founders: CEOs, confident and armed with a free afternoon and an LLM, arriving with a fully-formed plan and the assumption that marketing headcount is now negotiable. His read on it: this isn't a bad strategy because Claude is bad. It's a bad strategy because you're outsourcing judgment to something built to agree with you. As he put it, nobody's asking who implements that plan, or who iterates on it when something in it doesn't work.

A generated plan is a snapshot. A GTM motion is a living thing that needs someone accountable for it, especially in categories moving as fast as AI, where Janessa Drainville says her own team refreshes positioning quarterly rather than the twice-a-year cadence most functions are used to.

The CRM That Got Cancelled, Then Un-Cancelled

The sharpest example from the episode is one Andy Culligan came across through his own network, something he read about secondhand: one CEO cancels a HubSpot contract, convinced the team can vibe-code a replacement. The UI looks great. Everyone's pleased. Then they try to replicate what HubSpot was actually doing under the hood, and the token bill alone comes back at roughly a million dollars. The CRM contract gets quietly reinstated.

It's one story, but it's the kind Andy says he's hearing versions of increasingly often. That's the gap nobody accounts for when they compare "what AI built me" to "what a properly resourced team builds." A slick front end is not a functioning back end, and patch cycles, security, and data integrity cost real money whether a human or a model wrote the code.

Takeaway: a plausible-looking output is not the same as a working system, in marketing or in engineering. The bill for finding that out is usually higher than the bill for doing it properly the first time.

Trash In, Trash Out Isn't a New Problem, It's Just a Bigger One Now

This is where Janessa Drainville's actual experience at Atlassian comes in. Long before "AI rollout" was a job title, marketers knew a messy CRM takes actual years to unwind. Her version of it for AI:

"It's like handing a human a giant box of files that have been 20 years in the making, and there's no rhyme or reason to the organization structure. They're going to do it, but it's going to take them a great deal of time. AI might do it a little bit faster, but it's still going to take a lot of compute power and a lot of token usage."

This is where a lot of the "just AI it" enthusiasm quietly dies. Every AI feature, chatbot, or copilot a company ships is only as good as the second and third-party signal feeding it, the same logic purple path applies to intent data: a signal without context is noise, and a model without clean inputs is expensive guesswork.

What This Means for GTM Teams Specifically

If your AI rollout is struggling and the instinct is to blame the model, check the data hygiene first. Usage-based pricing makes this an increasingly expensive place to be sloppy: every token spent detangling bad structure is a token not spent on the actual job.

Transparency Is the New Differentiator

Janessa Drainville flagged something that maps directly onto B2B tech's current AI-messaging problem: customers can smell tone-deaf AI positioning immediately, especially now that "our AI is the best" is what every vendor on the market is saying. Her fix isn't more polish, it's more honesty:

"Letting folks know what some of the shortcomings are really helps. People want to talk to a human. They want to have grace for people."

That's a positioning discipline, not a copywriting trick, and it's one a lot of B2B tech companies are getting wrong in their AEO and GEO content right now: leading with capability claims instead of leading with credibility.

How Marketers Prove They're Still the Point

Janessa Drainville's advice for anyone handed a plug-and-play plan by someone senior: ask clarifying questions that only someone with domain expertise would know to ask. Not to be difficult, to demonstrate the gap between "a plan" and "the right plan for this specific audience, this specific sales cycle, this specific market."

She draws on a useful distinction from marketing analyst Christopher Penn here: AI is fundamentally lateral, A leads to B leads to C. Humans, and marketers specifically, think associatively across disciplines, how would a product manager approach this, how would a philosopher, how would a salesperson stress-test this claim. That cross-pollination is the actual differentiator, not "soft skills" as a vague catch-all, but a demonstrable, harder-to-automate way of thinking.

The Practical Version of This

When a CEO hands over an AI-generated plan, don't reject it outright. Come back with the missing layer: audience nuance, historical data on what's already been tried, and a cost-and-effort analysis of what full AI-led execution would actually require. That reframes the conversation from "can we skip marketing" to "here's what we keep, here's what falls away if we do."

Where This Lands for Fractional GTM

This is, functionally, the entire argument for doing Fractional Marketing properly instead of treating it as "marketing, but cheaper." The value was never headcount for headcount's sake. It's embedded expertise that catches a bad plan before it becomes an expensive lesson, whether that plan came from a slide deck, a board directive, or a chatbot.

At purple path, this is the same reasoning behind why we treat MQLs as a legacy metric and intent as a progression rather than a moment. It's also why the Fractional Marketing model exists in the first place: not to replace strategic thinking with cheaper execution, but to make sure someone with the pattern-recognition to say "this will cost you a million dollars in tokens" is in the room before the CRM gets cancelled.

Summary of Key Takeaways

• An AI-generated GTM plan is a snapshot, not a strategy. Someone still has to implement it, own it, and iterate on it.
• "Looks good" and "works" are different bars. A pretty UI hides backend, security, and data costs that show up later, often at a scale nobody budgeted for.
• Messy data is still the number one reason AI rollouts underperform. Fix the inputs before blaming the model.
• Transparency about AI's current limits builds more trust than overselling its capabilities.
• Marketers prove their value by demonstrating cross-disciplinary, associative thinking, the thing AI still can't replicate.

Frequently Asked Questions

1. My CEO wants to build our GTM plan using AI. How do I respond without sounding defensive?

Don't reject the plan, add to it. Come back with audience nuance, historical performance data, and a cost breakdown of full execution. This reframes the conversation around trade-offs instead of confrontation.

2. Why do AI tools struggle so much with messy CRM data?

AI can only work with the structure it's given. A disorganised CRM or years of untagged records forces the model to do expensive, time-consuming detangling before it can do anything useful, and usage-based pricing makes that inefficiency costly.

3. Is it actually cheaper to vibe-code a replacement for tools like a CRM?

Rarely, once you account for the full lifecycle. The upfront build might look cheap, but ongoing patching, security, and scaling costs on a custom-built system can exceed the price of the SaaS subscription it replaced.

4. How does this connect to Fractional Marketing specifically?

Fractional Marketing's value isn't reduced headcount, it's embedded expertise that catches bad decisions, like an AI-generated plan with no audience nuance, before they become expensive mistakes.

5. What's the single biggest AI positioning mistake B2B tech companies make?

Leading with capability claims instead of credibility. Being upfront about what an AI feature doesn't do yet builds more trust than polished messaging that oversells it.

Ready to build a GTM plan that survives contact with reality?

If your team is caught between "the CEO built a plan in Claude" and "the CRM data is too messy for any of it to work," that's exactly the gap purple path operates in. We embed the expertise, the playbooks, and the judgment calls that AI actually needs.

Talk to purple path about your GTM stack

‍