The marketing org got weird, in the best way possible

On any given day, my team has agents reviewing every piece of content we publish, checking for search relevance, brand guidelines, and EEAT. Our SDRs are working with AI surfacing intel on prospects before a human ever picks up the phone. Reporting that used to eat half a day now takes three minutes.

That's not a slide in a roadmap deck — it's just a regular Tuesday.

And still, if you asked me where we are in our AI journey, I wouldn't say "scaling fast." I wouldn't even say "figuring it out." I'd say "messy middle."

I hear that phrase constantly now, in conversation after conversation with other marketing leaders. Not on a scale of one to ten. Just: where are you? And the answer that keeps coming back (from people running teams far more AI-sophisticated than mine) is the same one, over and over. Messy middle.

I love that, actually — as marketing leaders, we're not going to get to any real solutions if we're not honest about where we are and what we're actually doing, versus what we're posting about.

It also stops me for a second every time I hear it, because I know that place from somewhere else, too. And I talk on it a lot; about what it costs you when you're operating at the edge of your capacity for too long, performing confidence you don't quite feel.

The messy middle of AI adoption and the messy middle of burnout have more in common than we usually admit. Both involve absorbing more than is sustainable. Both involve saying yes to everything while the work that actually matters keeps getting pushed down the list. And in both cases, the first step out is the same: naming where you actually, really, truly are.

Let's talk about what's actually on LinkedIn

Open LinkedIn right now. Within three scrolls, I guarantee you'll find something like this:

"I just spent four hours in Cursor and automated my entire workflow."

"We replaced three headcount with one AI agent."

"AI changed my life. Here's the 47-step playbook."

I'm not saying none of it is real. But what I am saying is that there's a gap between what AI sounds like in our feeds and what it actually looks like inside most marketing organizations right now. And closing that gap is where the real work is.

Here's what the data actually says:

For a lot of teams, AI hasn't taken work off their plates. It's added work onto them. New tools. New workflows. New questions nobody has good answers for yet. Is this output good? Did it hallucinate this? Who owns governance? 

There's a term for it now, and it's called AI burnout. Not from AI doing too much, but from humans doing more to accommodate AI that still hasn't delivered on the promise.

I know what that feels like — not in theory, but from the inside. Transparently, I had my own burnout experience earlier this year; a moment when I realized I was physically present in a room but not actually there. Running on fumes and calling it leadership. The AI version of that is quieter, but it's the same pattern: absorbing every new tool, every new initiative, saying yes to all of it, while the craft — the actual thinking work — keeps getting deferred. My team has felt it too. It's real, and it's worth naming before we get to any solutions.

Like, there are so many conversations that I've had that don't hit LinkedIn. The marketing leader who told me she spent six months convincing her CFO to fund an AI initiative, got the budget, got the tools, and is now sitting in front of a dashboard she can't explain. The demand gen director who built an automation that works perfectly and still can't tell whether it's helping pipeline. These aren't failure stories. They're the actual stories. And they totally deserve to be part of the conversation too.

 

The mistake almost everyone is making

So then, what's going wrong? It's not a lack of ideas — definitely not. Every member of my marketing team has a list of AI use cases as long as their arm. It's not even a lack of ambition, but something more specific.

Most teams, when they see AI, immediately ask: "What process can I automate?"

That sounds reasonable. It even sounds smart. But it leads you somewhere dangerous; it leads you to automating processes that maybe shouldn't exist in the first place.

Here's a real example from our own team:

Our campaign operations team had a form. A request form that marketers fill out to brief campaigns into the system. It was a pain point. Naturally, the idea came up: let's build an agent to help marketers fill it in faster. And on the other side, let's build an agent to check that people are filling it in correctly.

That's two agents. Both optimizing. Both burning time. Both solving for... the form.

But wait, why does the form exist?

It exists because we want a well-structured Salesforce record. A properly tagged Marketo campaign. A clean email program brief. That's the output, and that's what we actually need. We don't need the form... or at least, we need to ask that question before spending six months optimizing it.

I've started calling this the difference between AI as a photocopier and AI as a redesign. At the moment, most of us are photocopying. But the teams pulling ahead are redesigning.

Cassie Kozyrkov, Chief Decision Scientist at Google, puts it well: "The kiss of death for enterprise value is throwing AI at poorly defined problems."

That's it. That's the whole thing.

The question that separates teams that are doing AI from teams that are being transformed by it is harder and more uncomfortable: what should our processes actually look like now? What can we reimagine entirely, rather than just automate?

 

3 things CMOs are genuinely struggling with

If you're a CMO right now, you're getting pressure from every direction — the board, your CEO, your competitors. You're supposed to have a plan, a proof point, and a metric... ideally yesterday.

In almost every conversation I have with fellow marketing leaders, the same three challenges come up.

1) Delivering on unproven territory

AI-driven marketing results are still largely unproven at scale. We're building the runway while the plane is in the air — and we're expected to land on time and on budget. There's no playbook yet that's been pressure-tested across enough industries and team sizes to trust completely. That's not a reason to stop. It is a reason to be honest about it.

2) Teams operating in silos, in disconnected tools

Every CMO I know has the same vision: a truly integrated marketing motion, personalized across the full journey, optimized over time. And every CMO I know has the same reality: content in one tool, demand gen in another, analytics somewhere else, and no single view of what's actually working. AI layered on top of that fragmentation doesn't fix it. It accelerates it. I joined Optimizely knowing this was something we needed to fix internally, not just for our customers. Broken foundations are broken foundations, regardless of what product you sell.

3) Measurement that actually holds up

Our North Star at Optimizely is return on marketing investment — are we moving more pipeline per dollar over time? But we're still building the measurement framework alongside everything else. And I think any CMO being honest will say the same. We're all trying to justify every dollar while the rulebook for what those dollars should do is still being written.

Actually, I'm gonna go ahead and add a fourth one — and it's the one that gets talked about least in boardrooms: keeping your team intact. Not just headcount. I mean keeping the people who are genuinely good at this work from quietly burning out while you're scaling the tech around them. The expectation that AI = infinite capacity is doing real damage. Meet your team where they are, not where the org chart says they should be.

What we got wrong: the pendulum swing

Before I get to where we're going, I want to share the biggest internal lesson from the last year, because I think it'll resonate.

We call it the pendulum swing.

When we first leaned into AI adoption, we did what a lot of teams do: ran hackathons, encouraged everyone to experiment, told people to go build. Great for energy and curiosity. Not great for governance, consistency, or actually figuring out what worked. The output was exciting, but the signal-to-noise ratio was not.

So we swung the other way. Centralized ownership. Standardization. Tighter control. That killed the momentum.

We're now landing somewhere in the middle — teams are still building, but inside a more structured, orchestrated system. That balance is genuinely hard to get right, and I don't think we're done figuring it out.

This is similar to the overcorrection after personal burnout: you pull back too hard, over-control, and lose the energy that made things work in the first place. Getting the balance right isn't a straight line.

The governance piece matters more than most people admit. AI without governance isn't innovation; really, it's just chaos with a better interface. Brand consistency, compliance, quality control: these aren't optional, and they don't work when you bolt them on at the end of a project. They have to be baked in from the start. That's the only version that sticks — and it's the version that earns back the trust of the teams who've already been burned by tools that promised efficiency and delivered more work.

It's the same challenge we see with our customers, which is exactly why we built Opal U | AI Marketing University. Today, Opal U customers have built over 800 agents, saved over 2,000 hours a week, and we've grown a community of more than 2,100 marketing AI practitioners working through exactly this together. And this is growing on a weekly basis (!!) — not because the technology is magic, but because structure, community, and shared learning make the difference between a pilot and a practice.

 

 

Why our big bets are growth plays, not efficiency plays

We've set ourselves a North Star: to 2x our pipeline impact within two years. And I'm letting you in on our plan on how to get there, and spoiler: no cost-cutting plays are involved. 

1) AI in the SDR motion

We're reimagining how our SDR team operates with AI embedded from the ground up. More intelligence, more leverage, better conversations with more of the right people. The key word is embedded. We're bringing AI to where the work is already happening, inside the tools they use every day — not asking them to switch to something new on top of an already-full plate. If it creates friction, it won't get used. Full stop.

2) Personalization at scale — but like, actual personalization

"One-to-one personalization at scale" has been on marketing slides for fifteen years. For fifteen years, we couldn't deliver it because everything was disconnected. What's different now is that we can do this inside one platform, where marketers are actually doing the work. Our team has built a segment builder that analyzes behavioral data and surfaces audience suggestions the team wouldn't have thought to look for on their own. That's data working as hard as the people.

3) Micro-targeting by vertical

Getting surgical about which verticals we go after, how we speak to them, and how we deploy resources to go deeper in the segments that matter most. This is a growth driver — it's about expanding reach, not just maintaining it.

4) And running through all three: real-time analytics at marketers' fingertips

Not a report that arrives three days later. Not a dashboard only two people know how to read. Insights surfaced in the same place where the work is happening, when the decision actually needs to be made.

"Data-driven marketing" and "personalization at scale" have been aspirational for a decade. What's different now is that we're building toward them in one governed, connected system — where marketers are doing the work, not handing off to other teams and waiting.

 

The agents my team have already shipped

I think it matters that we're not just talking about what's possible, so let me talk you through what we've already done. 

Our team built a GA4 reporting agent that takes what used to be half a day of work and turns it into three minutes. Raw data to executive summary, channel breakdown, and ranked action plan. Anyone can now self-serve. The analyst who was the bottleneck gets their week back. That's not a small thing — that's time reinvested into the work that actually needs a human: the judgment calls, the creative decisions, the craft that no agent can replicate.

We built a content performance loop — an agent that reviews every piece of content at the 30-day mark, flags underperformers, and surfaces them for a human decision. No more set it and forget it. The human is still in the loop. The agent just makes sure the human doesn't have to remember to look.

We built a GEO/AEO recommendations report that used to take 40+ hours manually across our full site. Now it runs in minutes, with plain-language fixes anyone can act on — no developer required.

None of these are moonshots. They're the result of a disciplined process: find the right problem, solve that specific problem, measure the before and after. Small changes, outsized impact.

 

The honest version of the story

I started this piece by mentioning burnout. I want to come back to it — because I don't think we can have an honest conversation about AI in marketing without it.

When I wrote about my own experience — the ER visit, the moment my son told me I'd been in the room but I wasn't there — the response I got wasn't what I expected. It wasn't sympathy. It was recognition. Hundreds of messages from marketers saying: me too. I thought it was just me.

It wasn't just them. It's structural. We've been asked to do more with less for years. AI arrived and instead of relieving that pressure, it often added a new layer: learn this, adopt this, prove this works, do it fast, do it publicly, do it right. That's a lot. It's okay to say it's a lot.

But here's what I genuinely believe: AI, done right, gives us the first real opportunity in a long time to get back to the work that actually matters. Not the reporting. Not the form-filling. Not the manual QA at ten o'clock on a Thursday night. The actual work: the judgment, the positioning, the message that cuts through because a human decided it should.

That's what I'm most optimistic about. Not the automation. The reclamation.

The teams I see making real progress aren't the ones with the flashiest AI strategy decks. They're the ones asking the harder questions: What do we actually need? What can we stop doing? What does this process look like when we start from scratch?

I got to test that question live recently, hosting the keynote at Agents in the Mix, Optimizely's AI marketing event. I asked the audience the same thing I ask every marketing leader I talk to — no scale of one to ten, just: where are you right now? The answer that came back, over and over, in the chat window in real time, was the same one I hear everywhere else. Messy middle. That was the conversation we were trying to have. I think it landed. Because naming where you actually are is the first step to getting somewhere better.

Start small. One agent. One workflow. One process you stop doing because a machine can do it better. And use the time you get back for the thing that only you can do.

The messy middle is where most of us are right now. Watch the Agents in the Mix sessions that show what the other side looks like.