Everyone is lying about AI

It's Time For a Reality Check: How Three B2B CMOs Are (Actually) Using AI Right Now

With all due respect…

Most of the use cases you see on LinkedIn about AI in marketing ARE NOT REAL.

Sorry I don't know why I am yelling this early but I'm not happy about this.

Sure, maybe if you're a solo consultant and you write a Substack on AI and your entire job is to write about cool little workflows and ways you can use AI to "one-shot" the website, or crawl every TikTok video about gardening trends and the auto-generate a bunch of blog articles… that's cool. That's useful I'm sure. But my focus here is on B2B marketing use cases and the reality for a CMO running marketing at a $100M ARR life sciences company based in Cleveland is much different than what you see online.

You have a tech stack that took three years and six figures plus to wire up. You have an infosec team that needs to review every new tool. You have data governance policies, vendor approval processes, and procurement cycles that take months. You have customers in regulated industries who care a lot about where their data goes and who's touching it. You can't just rip everything out and rebuild because someone on X said agents are the future, so the new VP of Marketing is Claude.

Let's be clear though: it's not that the CMOs I talk to are anti-AI. They're just trying to figure out how to actually move forward inside the constraints of a real company with a lot of dependencies and nuance.

We recently hosted a session where a group of CMOs showed what they're really doing with AI right now. Not the demo. Not the hype. The stuff that's working inside the business, with real teams, on real workflows and I thought it was very grounding. Took my anxiety down a bit.

We had Tara Corey (CMO, Optimizely) and Julia Maguire (Director of AI Adoption, Optimizely), Lily Bond (CMO, Three Play Media), and Pejman Roshan (CMO, Menlo Security) on. The full session is up now on my YouTube channel if you want to follow along.

Here are the three plays that stood out:

1. Lily Bond (CMO at Three Play Media) replaced two BDR headcount with one HubSpot agent.

Three Play Media promoted two of their three inbound BDRs to AE roles this year. Instead of backfilling, Lily decided to test something: could a single BDR plus an AI agent handle the same inbound volume as three people?

She built the first version in 30 minutes using HubSpot's native AI agents which she said have improved dramatically of late. The agent pulls in the contact's CRM history, every page they've viewed, every form they've filled, every webinar they've attended, plus company research, and drafts a personalized response. Sequence-based or adaptive, depending on the prospect's behavior.

The numbers after launch: Response rate went from 18% to 46%. Meeting booked rate went from 5% to 35%. Their one remaining BDR now spends all day on calls instead of follow-up emails.

Lily tried this same play a year ago with Cassidy AI and it didn't work. The tools weren't ready. She came back to it when HubSpot's agents got good (she suspects they upgraded the underlying model). The lesson is that "we tried AI and it didn't work" is a moment in time, not a permanent answer. Second, she didn't try to boil the ocean. She picked one specific workflow with a clear forcing function (two open headcount) and built the smallest version that could work.

2. Tara Corey (CMO at Optimizely) focused on bringing AI to the tools everyone is already using today.

Most AI workflows fail because they ask marketers to leave their tools, open a chatbot, copy-paste something, and bring the output back. Tara's whole approach at Optimizely is the opposite: bring the AI to wherever the marketing work is already happening.

Her content team writes every deliverable inside Optimizely's content marketing platform. So that's where the agents live. A writer finishes a blog post and hands it off to two agents inside the same view: an E-E-A-T checker (scans AI content) for unsupported claims, outdated references, missing internal links, and shallow citations. And a brand voice checker, which only leaves a comment if there's something to actually flag. (Her rule: "If it's not gonna give you something helpful, it should just stay out of your way.")

Then there's the post-publish piece, which I loved. Optimizely's GA4 instance is wired into the platform, so an agent scans every article 28 days after it goes live, checks performance against benchmark, and if it underperformed, auto-creates a ticket and assigns it back to the content marketer with recommended fixes. Smart. No more "set it and forget it." No more begging the one analytics person on the team to pull a report. The work just shows up in your queue.

If you've ever tried to roll out an AI workflow and watched it die because nobody wanted to switch tools, this is the model. Meet your team where they already are.

3. Pejman Roshan (CMO, Menlo Security) says the secret is to measure twice, cut once.

Menlo Security's team was where a lot of teams are: lots of excitement, lots of people playing with chatbots, but stuck in what Pejman calls pilot purgatory. Nothing was actually moving the business. Just a whole lot of little side projects and "cool" but not useful stuff.

So before building anything, his team mapped the workflows. Where were people spending the biggest chunks of time on repeatable work? That's where they'd hone in. Not the shiny stuff. Not the thing I said earlier about someone mentioning it on X or LinkedIn. But the boring, expensive stuff. (This was the point where everyone in the chat asked for during the session; timestamped to that section here.)

Two examples of what that approach produced:

The first was brand voice. Menlo has 18+ content creators across product marketing and content marketing, and their brand guide is 40,000 words long. Reviewing every piece of content against that document manually was eating hours, sometimes days, especially around launches. So his team built a tool that scores any URL against the brand guide across voice characteristics, audience language, and archetype alignment. It flags the lines that drift, suggests tracked changes, or just rewrites the whole thing. Multi-level manual reviews became minutes.

The second was even more practical. Whenever Menlo updates major messaging or launches new positioning, they have hundreds of backlog documents that need to be reviewed and updated. Trying to do that in a chatbot doesn't work because of document limits. So they loaded everything into NotebookLM via URL, wrote one prompt, and got back a Google Sheet mapping every document that needed updating with the option to auto-update them. Weeks of work, gone.

Pejman's framing on AI to his team is one I'm stealing: this isn't about cost savings. It's about finally having time to do the things you've been meaning to do for two years. The campaigns you couldn't staff. The audits you kept pushing. That's how you get a team bought in.

Beyond the hype, the CMOs winning with AI right now are the ones who picked one workflow, built the smallest version that could work, and proved it before scaling. Not the ones who bought five tools and announced a transformation initiative. WE REPLACED THE WHOLE MARKETING TEAM WITH CLAUDE! No you didn't.

Pick the one workflow that's costing your team the most hours right now. Build the v1. Ship it. Then go find the next one.

That's the reality. It just doesn't play as well in a viral LinkedIn post…

– Dave

P.S. What's the one AI workflow you've built that's actually working inside your team? Hit reply, I want to hear about it. Did you relate to what I'm writing here?

If you want to go a level deeper on AI, here's a video we did with Corey Haines all about Claude Code for B2B marketers.

‍