The AI-First Marketing Operation

Transcript

[SLIDE 1 — Title]

Right. Let me start with a confession that should worry anyone who sells AI for a living.

Most of what gets called an “AI-first marketing team” in 2026 is a team that bought ChatGPT licences and a few clever tools, and changed almost nothing about how the work gets done. The logos changed. The org chart didn’t. The way a campaign moves from idea to live to measured is exactly what it was three years ago, only now there’s a chatbot drafting the first version of the email nobody should have sent.

So this talk is about the difference between those two things. Where AI genuinely replaces process in a marketing team. Where it absolutely does not. And why the teams getting real leverage are building systems, not buying tools.

I run a consultancy called GAMEPLAN. We’re AI-first, which is an overused phrase, so let me define it by what we actually do rather than what we say. We ship custom websites in forty-eight hours. We run paid media across Google, Meta and TikTok with a measurement layer most agencies don’t have. And we do it with a headcount you could fit in a taxi. None of that is because we found a magic tool. It’s because we rebuilt the process.

[SLIDE 2 — The gap]

Here’s the number that frames everything.

This year, BCG surveyed three hundred chief marketing officers globally. Ninety-six per cent of them said AI is driving end-to-end transformation of their function. Ninety-six. That’s not a trend, that’s a standing ovation.

Then they asked what people had actually built. Forty-two per cent are using generative AI only to help a human with discrete tasks. Write this email. Summarise this brief. Just under a third have moved to anything you could call an agent-led workflow. And eight per cent, eight, run campaigns where multiple agents operate with any real autonomy.

So the honest picture is this. Ninety-six per cent claim the transformation. Eight per cent have done it. The gap between those two numbers is the most interesting thing in marketing right now, and it’s the entire subject of this talk.

The good news, if you’re in the ninety-two per cent, is that the bar is on the floor. The teams that have actually rebuilt are not winning because they have smarter people or bigger budgets. They’re winning because they did the unglamorous work everyone else skipped.

[SLIDE 3 — Tools assist, agents execute]

Let me draw the line that the whole industry blurs on purpose.

Tools assist. Agents execute. They are not the same thing, and most of the spend in marketing right now is on the first while the brochure promises the second.

A tool waits for you. You prompt it, it answers, you do something with the answer. ChatGPT drafting copy is a tool. An image generator is a tool. A subject-line tester is a tool. Useful, all of them. But the human is still pushing every button and carrying every handoff between them.

An agent does the thing. It monitors, it decides within limits you set, it acts, and it does the next step without you in the loop for each one. The work moves while you sleep.

The reason this matters is that almost nobody bought what they think they bought. Gartner reckons that of the thousands of vendors waving the word “agentic” around, roughly a hundred and thirty are actually doing it. The rest is what they politely call agent washing. An old chatbot with a new sticker. So when a marketing leader tells me they’ve “gone agentic,” my first question is boring and deliberate: show me a workflow that completes without a human touching each step. Usually there isn’t one. That’s fine. But let’s not confuse a licence with a transformation.

[SLIDE 4 — Where AI replaces process]

So where does AI genuinely replace process? Not where you’d hope, if you sell content generation.

The real leverage is in operations. The plumbing. The checks, triggers, monitoring and handoffs that make a marketing function actually run. This is the stuff nobody puts on a conference slide because it isn’t sexy, and it’s exactly where the returns are.

Five that work, today, in real teams.

Competitive intelligence. Instead of asking a junior to “keep an eye on competitors,” which means they look once a month in a panic, you run an agent that watches rival pricing pages, job listings and launches every day and drops a summary in Slack each morning. Constant intelligence, zero human effort.

Briefing. An agent that builds a structured content or campaign brief, with the competitive context, the keyword data and the internal linking already in it, before a human writes a word. That alone removes the weeks teams waste aligning on scope.

Quality enforcement. Automated scans across your site for broken links, missing metadata, the tell-tale patterns of careless AI writing. The boring audit that never gets done because no human enjoys it.

Signal routing. Watching intent and CRM activity continuously and surfacing the warm accounts automatically, so leads don’t go cold while a human gets round to the spreadsheet.

Repurposing. One strong long-form piece, turned into the blog post, the LinkedIn version, the email section, the carousel. The format conversion is automated. And note the line I’ll keep coming back to: the conversion is automated, the strategy and the voice stay human.

Notice what these have in common. None of them is “have the machine think for you.” All of them are “have the machine run the system so the humans can think.”

[SLIDE 5 — Where it doesn’t]

Now the other side of the line, which is the part the vendors are quiet about.

There is work AI does not replace, and pretending otherwise is how you lose a client or a quarter.

It doesn’t do judgment. The decision about what this brand is for, which market to enter, which campaign to kill, that is yours. An agent can lay out the options beautifully. It cannot own the consequence.

It doesn’t do taste. AI produces the competent average of everything it has seen, which is precisely the thing a strong brand is trying not to be. Distinctiveness is a human act. The machine pulls you toward the mean.

It doesn’t do the money conversation. When your finance director asks why you want another hundred thousand in spend, “the model suggested it” is not an answer that survives the room. That’s a human standing behind a number.

And it doesn’t do the relationship. The trust, the read on a client’s actual mood versus their stated brief, the moment you say “I wouldn’t do that” to someone paying you. That is the job. It always was.

So the operating principle is this. Delegate the process. Keep the judgment. The teams that get this wrong don’t fail because the AI is bad. They fail because they handed it the one thing it was never going to be good at.

[SLIDE 6 — Build, don’t buy]

Here’s the part that genuinely changed in the last eighteen months, and it’s the reason small teams are about to eat some large ones.

The cost of building your own system fell by an order of magnitude. A year ago, connecting an agent to your CMS or your ad platform meant custom engineering. Today there’s a standard connector for it. Building a multi-step workflow meant a dev team. Today the orchestration comes out of the box. The protocols standardised. The plumbing got cheap.

What that means in practice is that “should we buy the enterprise platform or build our own” stopped being a question with an obvious answer. For a lot of teams, building is now faster, cheaper and a far better fit than buying a bloated suite you’ll use a tenth of.

And here’s the counterintuitive bit. The companies best placed to do this are not the Fortune 500s with the biggest martech stacks. It’s the small ones. Because they have no legacy. No eighteen-month implementation. No committee. McKinsey’s own data has most organisations still stuck in the experimental phase, and only around a quarter actually scaling. The blocker is never the technology. It’s the org wrapped around it.

We run GAMEPLAN. this way deliberately. We don’t sell you a tool. We find where the process is broken, and we build the system that fixes it. That’s the whole offer.

[SLIDE 7 — Garbage in, faster]

One principle, and if you take nothing else, take this.

Agents amplify whatever you feed them. Good process or bad.

If your content is mediocre and you bolt an agent onto it, you now produce mediocre content faster and at greater volume. You haven’t solved the problem. You’ve scaled it. You’ve automated the part you should have fixed.

This is the most common way AI projects quietly fail. Someone automates a broken workflow, the broken output multiplies, and six months later the conclusion is “AI didn’t work for us.” AI worked perfectly. It did exactly what you asked. The process was the problem.

So the order of operations is not negotiable. Fix the process first. Then automate it. Never the other way round. The sequence is the strategy.

[SLIDE 8 — Anatomy of one system]

Let me make this concrete with a single system, because abstraction is where these talks go to die.

Take content operations. The old way is a person prompts a tool, gets a draft, and a different person does something else with a different tool. Disconnected steps, no shared context, a human carrying every handoff.

The system version is a loop. Research the topic and find the genuine gap. Build the structured brief. Draft against it. Run it through SEO and answer-engine optimisation. Publish to the CMS. Track how it performs. Feed that back into what to make next. One connected loop, where the machine runs the chain and the human sets the strategy and approves at the points that matter.

The difference isn’t that the system writes better. Sometimes it writes worse. The difference is that the work moves continuously instead of stalling at every desk, and the human attention goes to the two or three decisions that actually need a brain, not the twenty handoffs that don’t.

Build one of these. Prove it. That’s worth more than a year of licences.

[SLIDE 9 — How this fails]

I want to be honest about the failure modes, because the hype merchants won’t be.

Gartner’s projection is that more than forty per cent of agentic AI projects will be cancelled by the end of 2027. Forty per cent. So the base rate of failure here is high, and it’s worth knowing why, so you’re not in it.

Three reasons it falls over, and they’re always the same three.

Data. Fragmented data is the number one killer. If your customer information lives across disconnected spreadsheets and three CRMs nobody trusts, your agent has nothing solid to stand on. Fix the data foundation or don’t start.

No human in the loop. Two-thirds of companies report real quality and safety problems with agents. The teams that survive build guardrails: spending limits, approval thresholds, the ability to roll back when an automated change makes things worse. You start agents on low-risk actions, pausing a bad ad, not high-stakes ones, launching a campaign.

And agent washing, again. Buying the sticker instead of the system. If you bought a tool and expected an agent, the disappointment was priced in from day one.

None of these is an AI problem. All of them are an operations problem. Which is the theme of this entire talk.

[SLIDE 10 — The new operating model]

So what does the team actually look like on the other side of this?

The marketer stops being the person who does the tasks and becomes the person who designs the system and conducts it. You set the strategy. You set the guardrails. You decide where a human has to sign off. And then you let the agents run the parts that don’t need you.

The most valuable skill on a marketing team is changing under our feet. It used to be proficiency in a specific tool. Being the person who really knew the ads platform. That skill is depreciating fast, because the tool now half-runs itself. The skill that’s appreciating is the ability to look at how work flows through your team, see where it stalls, and design the system that fixes it. Systems thinking, not button-pushing.

That’s a genuinely good deal for the people who embrace it, by the way. The machine takes the repetitive, the data-heavy, the things humans do badly and resent. You get back the strategy, the creative judgment, the relationships. The job gets more human, not less. But only if you actively claim that reclaimed time, which, as it happens, is a whole other talk.

[SLIDE 11 — Start Monday]

So what do you do on Monday morning. Not in some eighteen-month transformation programme. Monday.

One bottleneck. Find the single workflow that hurts most right now. Can’t keep up with competitors? Start there. Errors slipping into published work? Start there. Leads going cold? Start there. You’re not boiling the ocean. You’re picking one thing.

One agent, one workflow. Build the smallest system that fixes that one thing. Resist the urge to do all five at once. You will fail at all five.

Prove it. Run it for a few weeks. Measure the time it gives back, the errors it catches, the revenue it routes. Get the evidence.

Then add the next one. And only then.

The teams winning in 2026 are not the ones with the most AI. They’re the ones with the most operational discipline. AI doesn’t replace the thinking. It executes the system. So build the system first.

[SLIDE 12 — Close]

Let me leave you with the whole talk in one line.

Stop buying tools. Start building systems.

Almost everyone in your market is going to spend the next year buying licences and calling it transformation. A small number are going to do the boring, structural work of rebuilding how the work gets done. In two years those two groups will not be competing in the same weight class.

The technology is no longer the hard part. It’s cheap, it’s standardised, it’s sitting right there. The hard part is having the discipline to fix your process before you automate it, and the judgment to keep the decisions that were always yours.

If that’s the work you want to do, that’s the work I do. Thank you.

---

Tom Goodwin is the founder of GAMEPLAN., an AI-first performance, media and technology consultancy in London. He is not the author of Digital Darwinism; that is a different Tom Goodwin. Book him to speak at tomgoodwin.london/speaking.

Sources grounding the talk

• BCG, Making the Agentic Marketing Transformation a Reality, 2026 survey of 300 global CMOs (96% claim transformation; 42% task-assist only; ~⅓ agent-led; 8% multi-agent autonomous).
• Gartner: >40% of agentic AI projects cancelled by end of 2027; ~130 of thousands of vendors genuinely agentic (“agent washing”).
• McKinsey: agents projected >60% of future AI value in marketing; ~62% of organisations experimental, ~23% scaling.
• Reported pattern across 2026 practitioner sources: leverage sits in operational execution (monitoring, briefing, QA, routing, repurposing), not content generation; “agents amplify whatever you feed them”; two-thirds of companies cite agent quality/safety issues.