AI, STRATEGY & THE 8% PROBLEM

I’ve sat in pitches where the deck was absolutely gorgeous. Clean type, charts you’d frame, the perfect amount of content on the slide, and timelines that were formatted so well, I couldn’t help but get jealous. Then the client asked one very simple question about the strategy, and no one could answer it. Not because the question was difficult. Because no one in the room had actually paid attention to what the deck was suggesting. I’ve also been in regroups after kickoffs and in the middle of projects where people can’t recall the simplest of details, and rather than being prepared or using the cliche, “I’ll circle back,” they tell someone higher up on the food chain that the “AI notes said,” to which my favorite response has always been, “Great, but I’m asking you.”

I saw new research from Mirel that pretty much quantifies everything we are thinking. 70% of agency marketers say they “trust” the output of AI, but only 8% say they trust its strategic recommendations. And 64% are using it for strategy support.

I like that 64%. I’m in the 64%. It’s also the figure that should nudge everyone.

Making things was never the hard part

Production is pretty checkable. You can take a headline, a resize, or a first-draft landing page, and, in four seconds, tell whether it works. If it doesn’t work, you fix it. You won’t lose an account over a clunky AI-written caption. Everyone has the same toolkit now. If any agency can produce 40 ad variations before lunch, speed isn’t a differentiator. It’s just the baseline.

One question everyone is going to start asking is, “What are you charging for if they can do it themselves?” If all you’re delivering is output, that’s a perfectly valid question. It’s true. And it’s a hard one to answer. What are you actually providing that they can’t do themselves?

The 92% aren’t wrong

Strategy isn’t about coming up with more ideas. It’s about killing the majority of them. It’s about working out what a brand won’t say and which audience you’re prepared to alienate.

AI only knows what you tell it. Unless you’re allowing it into meetings or feeding it transcripts, it doesn’t know the CMO made a joke about how the board can’t stand the campaign anymore. It doesn’t know that sales and marketing are two departments that refuse to communicate after Q2. And it doesn’t know how to repair a broken LinkedIn strategy that is betting on one thought leadership article for virality. Sure, these are slightly fictional scenarios. But you could swap in any of your own experiences where something wasn’t in a brief that completely changes the strategy, and AI still won’t have the sense that a vertical is getting tired of a trend before the internet has gotten around to bleeding it dry.

The greater risk is how good a bad strategy can look at the moment. AI doesn’t fumble. It’s beautifully formatted. Give it a fuzzy brief, and it will come back with something completely confident but completely indistinguishable from the output another agency across the country with the same brief could produce.

That’s how you end up with a 45-slide deck. Volume is being used as a proxy for rigor. Nobody asked whether slide 31 actually needed to be there, because it cost nothing to make it. And no one really cared to take the time to cut it. I recently sat in on a presentation where a series of slides laddered up to repeating the same thing 3 different ways.

Where I put the 64%

That presentation in particular wasn’t developed by someone who didn’t know what they were doing. The problem was that it turned out looking and feeling so good from the jump, they didn’t really feel the need to edit. I’m guilty of this, too. I’m not innocent. AI is a solid brainstorming partner. It helps me name things, find new angles, and it can create and review FDA documents to verify a claims matrix quicker than I ever could. It can spot holes in things before clients even get around to it. Most importantly, it can create so many ideas that I can immediately know which ones to toss out and which ones are worth fighting for.

That’s labor I’d gladly hand over.

But what I won’t ever hand over is my judgement. Working out what the actual business truly is is very rarely the one that turns up in an RFP. It comes from listening, from conversations, and from asking the questions that are sparked during human interaction. People love to tell you about their work, what their mission is, and every once in a while, you’ll find someone willing to admit that all the attempts they’ve made before have fallen short. You have to be able to be the one to stand up in front of a client when they push back, and if you’re strictly relying on AI to do everything, there’s no way you’ll ever be able to.

I ride or die by the ‘why.’ The ‘why’ is and always has been my North Star. Why this audience? Why this message? Why didn’t that approach work before? A tool can easily get me to that answer much quicker. It can analyze the data. It can point out something I may have missed. Hell, it can even help me explain a complicated concept without all of the jargon, so a 3rd grader could understand the impact of impressions and CPCs on a paid media campaign or how a Pixel stopped working so we momentarily lost the data. But there is one thing it can’t do: make me believe it outright. I would never just share an AI output and call it my work. Just the way I’d never say “Well, Claude said, so it is law.”

What Agencies Should Actually Be Selling

If you’re still pricing on hours spent doing the thing a client could do themselves, eventually they are going to do that calculation. And if you think they won’t, trust me. I’ve sat in on conversations where someone has specifically pointed out how much other people on the team LOVE using AI and “you can tell.”

What you need to add is the value. What they can’t get from a prompt is someone who is going to use their knowledge and experience to say, “Don’t do that,” and tell you precisely why. It also won’t take responsibility for it if it ends up being wrong. That’s the premium. So sure, go ahead and use the AI. Have it do the work you don’t want to waste extra time on. Have it develop the deck you’re going to use, but format it first. Build an outline. Or seriously, just edit the AI output at the end. Cut the deck down to 12 precise slides that don’t repeat themselves to death. Then spend the time you’ve saved making sure everyone who is presenting can defend every single slide - at a moment’s notice.

Here’s the test I’d give myself or anyone with a team: take the AI away tomorrow. Could you still explain why this was the right strategy? Could you do it without the deck?

If not, then the tool isn’t your issue.

 
Next
Next

OLD SOULS, NEW ROOMS