AI Will Not Save a Bad Strategy. Here Is Where It Actually Helps.

The question nobody is answering honestly

Every agency is telling their clients they use AI. Every pitch deck has a slide about it. Every newsletter has a section on how the team has "integrated AI into the workflow."

Nobody is explaining what that means. And almost nobody is asking the harder question: has the thinking gotten better or worse since the tools arrived?

There are plenty of situations where AI has made the thinking worse, not better. Not because AI is bad. Because AI makes whatever quality of thinking already exists move faster. Strong thinking plus AI is genuinely powerful. Weak thinking plus AI creates more output, more activity and more apparent progress, without necessarily producing better decisions. 

This is the conversation the industry is not having. So here it is.

What actually happens when agencies "use AI"

A brand we know went through an agency transition last year. The new agency came in with a polished onboarding process, a technology stack slide, and a clear emphasis on how they used AI to accelerate creative production and reporting.

Three months in, the client had more deliverables than they had ever received. Weekly reports with charts and commentary. 200 new static ad images a week. Detailed strategy documents. 

Nothing was working better. In fact, performance had declined slightly. When the client pushed for a strategic conversation, the agency produced another document.

The problem was not the AI. The problem was that nobody was thinking. The briefs going into the AI were weak because nobody had done the hard work of understanding the customer, the offer, and the actual constraint on growth. The AI produced output that looked competent and was strategically empty. The volume of that output made it harder to see the emptiness, not easier.

That is the trap. And it is everywhere right now.

Where AI is genuinely useful

Before the hard truth, the fair picture. AI is a real productivity multiplier in a specific set of tasks.

Production speed on well-defined work.

If the task is clearly scoped, the output format is known, and a human can quality-check the result quickly, AI is excellent. Writing twenty hook variations from a strong brief. Summarising a month of performance data. Drafting the first version of an email sequence. Turning a call transcript into structured notes.

These are real gains. A task that took two hours takes twenty minutes. The key word is well-defined. The brief has to be good. The judgment about what good looks like has to be human.

At Flat Circle, this shows up in creative production. When we are building out a testing pipeline, we need volume quickly. AI accelerates the early ideation stage meaningfully. Not because the output is the final product. Because reacting to twenty rough starting points is faster than generating them from scratch. The human layer decides which directions are worth pursuing, rewrites the ones that are close, and kills the ones that are not.

Without that human layer, you get twenty ads that are grammatically correct and strategically identical because they all came from the same underlying brief that nobody interrogated properly.

Synthesising large amounts of information.

Reading 400 customer reviews to extract the recurring objections, the specific language patterns, the purchase motivations that nobody in the internal team would have guessed. AI does this quickly and accurately. This is one of the most underused applications in performance marketing.

A client came to us convinced their customers bought their product for one reason. We ran the reviews through an AI synthesis process and found that the second most common motivation, which appeared in about a third of reviews, was something the internal team had never once mentioned in a brief. That insight changed three months of creative direction.

The AI found the pattern. A strategist decided what to do with it. Both parts were necessary.

First drafts, structuring, formatting.

Reports, briefs, documentation, proposals. AI removes the blank page problem and handles the mechanical work of assembly. The thinking behind the structure is still human. The time saved is real.

Where the value disappears

Strategy.

AI cannot tell you what your brand should stand for, which customer segment is worth going after, or whether your offer is actually compelling to a cold audience. It can produce a strategy-shaped document. That document will be coherent, well-formatted, and built from the average of what is generically true across all brands rather than what is specifically true about yours.

We have seen these documents. They read well. They contain recommendations like "focus on emotional storytelling" and "test price-anchoring in the creative" and "consider retargeting users who visited the product page." These are not wrong. They are also not useful because they are not connected to the actual situation. There is no judgment in them. There is no prioritisation based on real context. There is no point of view on what the specific business needs to do next.

Strategy requires understanding the margins, the unit economics, the product reality, the competitive landscape, and the customer who actually buys rather than the customer the brand imagines it is selling to. You cannot prompt your way to that. You have to build it through real engagement with the business.

Creative judgment.

AI can help identify patterns and generate hypotheses. But deciding which signal matters, how much confidence to place in it and what risk to take next is still an operating decision. AI can generate creative. It cannot tell you which creative will move a cold audience on Meta in Q4 against increased competition and category fatigue. That judgment comes from pattern recognition built through running real campaigns, losing real budget on things that did not work, and developing an intuition for what the data is telling you underneath the headline metrics.

The challenge is not that AI can never access the context. Increasingly, it can. The challenge is that most organisations have not captured, structured or connected the context required for reliable decisions.

A model is only as context-rich as the operating system around it.

Better models will improve the quality of the analysis. They will not remove the need for someone to own the decision, understand the trade-offs and be accountable for the outcome.

Anything where context is the whole point.

Should we scale this campaign this week? Is this a creative problem or a structural problem? Is this flat performance a signal we should act on or variance we should wait out?

These questions require judgment grounded in specific context. AI applied to them without that context can produce a confident answer from incomplete information. The risk isn't just a bad answer. It's the illusion that the decision has been properly reasoned through.

Where AI actively makes things worse

It is being used to avoid thinking.

The most common misuse of AI in marketing is using it to skip the work the job actually requires. Brief the AI, take the output, ship it. No hypothesis. No rationale. No human judgment about whether this is actually the right direction.

This produces fast, mediocre work at scale. And fast, mediocre work at scale is exactly what you are competing against in every ad auction. The brands generating AI creative without a human layer of strategic judgment are contributing to the noise. They are not cutting through it. They are making the noise louder.

It dilutes the specific into the generic.

AI output without strong proprietary context tends toward the average. The average hook. The average offer framing. The average recommendation. Average is invisible in a feed. Average does not convert cold audiences who have no prior relationship with the brand.

The specific insight, the customer language that sounds exactly like what your buyer thinks before they purchase, the angle that is slightly counterintuitive, the observation about your product that nobody else has made — these do not come from a model that has ingested everything. They come from people who have done the hard work of understanding one specific brand deeply.

A team we know ran an AI creative sprint. Two weeks, one brief, fifty ad variations. Not one outperformed the existing control. When they looked at the outputs, every single hook was a variation of the same underlying idea. The brief had a hypothesis in it but it was a weak one and nobody had challenged it before they started generating. The AI executed the brief faithfully. The brief was wrong. Fifty ads confirmed the wrong hypothesis efficiently.

It makes junior work look like senior work.

AI makes a weak brief look like a considered document. It makes a shallow strategy look like a thorough one. It makes an inexperienced team look more capable than they are at the deliverable level.

Clients see more output and mistake it for more thinking. The reports are longer. The creative batches are bigger. The documentation is more polished. But the quality of the judgment behind all of it has not improved and in many cases has gotten worse because the friction that forced slower, more deliberate thinking has been removed.

You are often paying for the appearance of a senior operation. The AI is providing the appearance. The seniority is not there. This is why output volume is becoming an increasingly poor proxy for agency seniority. AI can dramatically improve the appearance of the operation. It cannot tell a client who is actually making the decisions behind it.

How to actually use it well

The teams getting real value from AI are using it as a multiplier for human judgment, not a substitute for it.

They define the task precisely before briefing the model. Vague briefs produce vague outputs. The work of writing a specific, well-reasoned brief is most of the value. The AI execution is fast. The thinking that makes the execution useful is still slow and human.

They treat every output as a first draft. The value is reducing the time from zero to something reviewable. Whether that something is actually good is a human judgment.

They keep clear human ownership of the strategic layer: what to test and why, what the result means and what the business should do next. AI can contribute to all three. It should not remove accountability for any of them.

And they are clear about where AI is creating leverage versus where it is simply creating more output.

The question worth asking your agency

Not: do you use AI?

That question is meaningless now. Everyone uses it for something.

The question is: has the quality of your strategic thinking improved since you started using it, or has it just gotten faster? 

If the answer is that things are moving faster and the output looks more polished, that is not the same thing as the work getting better. It might mean the opposite.

The brands that will win with AI are the ones where the tools are accelerating strong thinking, not replacing it. Those brands exist. They are a minority.

The rest are moving faster. They do not always know where they are going.

The advantage won't come from who uses the most AI. It will come from who builds the fastest loop between insight, judgment, execution and learning. AI can accelerate that loop. It cannot replace it.