One AI. Five Reasons to Use It.
Over the very hot summer, I spoke with over a dozen agency leaders and experts about AI, and what leads the conversation about how to use it. Not about models, features or whether agencies should use it. That ship has sailed. I was interested in something closer to their daily reality: what is becoming central in the conversations they have with clients about using AI? One answer came back remarkably often, and practically always first in line: efficiency.
It makes sense, in a way. Agencies are finding ways to research faster, develop first drafts faster, produce variations faster and automate work that previously consumed expensive human hours. Clients know this too. So a perfectly reasonable commercial question follows: if AI makes the work more efficient, who benefits from that efficiency?
But something interesting happened when I kept asking what efficiency in this context actually meant. The answers became less precise. Not because these people had not thought seriously about AI. They had. It was because a seemingly concrete word turned out to hide quite a few different questions.
Efficient at what?
Suppose AI saves three hours. Sounds good. But three hours of what? Research, exploring ideas, developing an argument, writing a first draft, editing, translation, producing fifteen variations or moving copy through approval? Those are not always interchangeable hours.
Saving three hours of repetitive adaptation may free a senior strategist to spend more time understanding the client’s problem. Saving three hours by shortening that understanding process might achieve exactly the opposite. And then there is another, million-dollar-or-more question: efficient for whom? The agency may save time while the client gets the same output for the same price. The client may expect the saving to be reflected in the fee. A team may use the saving to produce more. Or everyone may simply fill the liberated hours with more meetings.
“Saving three hours isn’t value yet. What happens to those three hours is.”
Efficiency, in other words, is not the end of a conversation with one homogeneous meaning. It should be the beginning of a value conversation.
Every gain can hide a loss
That became clearer in those agency conversations when we moved beyond the first answer. We tend to name what we want to gain without naming what we might lose in getting it. Faster research sounds attractive, but not if it produces shallower understanding. Faster writing sounds attractive, but not if the efficiency comes from removing the thinking that made the piece worth reading. Cheaper production sounds attractive, but not if the client was actually paying for expertise, judgement and a point of view rather than the physical production of 1,200 words.
“We want AI to increase X without sacrificing Y.”
That sentence makes “we want to use AI for efficiency” suddenly insufficient. We might want AI to reduce production time without reducing the strategic thinking the client is paying for. Or make research faster without making our understanding shallower. Or reduce routine work so people can spend more time on work where human judgement matters. Now we are having a useful conversation.
There is a useful precedent here. Digital tools have never created only one kind of value. Some made work possible or better in ways that changed the craft itself. Think Photoshop or Procreate. Others made an existing level of output easier, faster or accessible to many more people. Think Canva, or more recently tools such as Jasper. The boundaries are not clean, and the products keep evolving, but the distinction matters: technology can improve the work, improve the economics of the work, or do both.
Once I started pulling efficiency apart this way, another question began bothering me: why are we assuming efficiency is what we want from AI in the first place?
Five peas, different destinations
There are already excellent inventories of what marketers can do with AI. We do not need another list of 50, 100 or 230 use cases. The question I find more useful comes one step earlier: what are we actually trying to improve by bringing AI into this work? I currently see five closely related questions. Five peas in the same AI pod, perhaps, but each can lead the work in a different direction.
| The ask | The advantage | Potential hidden trade-off | |
|---|---|---|---|
| Efficiency | What should become easier? | Less time, cost, effort or friction. | Depth, judgement or valuable human time gets cut with the waste. |
| Quality | What exactly should become better? | Clearer, sharper, more accurate, useful or consistent work. | Polish disguises weak thinking or erases a recognisable voice. |
| Exploration | What do we want to see that we might otherwise miss? | More alternatives, perspectives, questions and counterarguments. | More possibilities become noise, convergence or outsourced judgement. |
| Scale | What becomes worth doing because AI makes more possible? | More markets, formats, audiences or versions become practical. | Errors, sameness and mediocrity scale with the output. |
| Learning | Who should become better afterwards? | Stronger human understanding, skill and judgement. | A better immediate output leaves the human less capable. |
The categories are deliberately simple. The same task can serve several of them. Research, for example, can become faster, better, more exploratory and more educational at the same time. The point is not to force every use case into one box. It is to make the intended gain explicit enough that we can also see what deserves protection.
Before you invite AI in
This does not need another elaborate framework. Before using AI for a meaningful piece of work, ask three questions: what are we trying to achieve? Which gain matters most here: efficiency, quality, exploration, scale or learning? And what are we unwilling to sacrifice to get it?
A CEO perspective might be designed primarily for exploration, using AI to challenge assumptions and expose alternative interpretations while explicitly protecting human ownership of the eventual argument and voice. Once that perspective is finished, using AI to adapt it for several markets may be primarily about scale, while protecting factual accuracy and the meaning of the original.
Same person, same AI, even the same original piece of content, but a very different reason for inviting AI in and therefore quite possibly a very different way of using it.
The other half of the brief
The conversation about AI has understandably spent a lot of time on capability: can it do this yet? Increasingly, the answer is yes. The more useful question now may be what we want that capability to improve. Efficiency is a perfectly good answer. So are quality, exploration, scale and learning. But naming the gain is only half the brief.
The other half is naming what must survive the improvement. Faster without shallower. Better without blander. More without multiplying mediocrity. Broader exploration without outsourcing judgement. More help without less learning.
“The workable idea: every AI ambition needs two answers. What do we want to gain, and what are we unwilling to lose to get it?”
That may be a more useful starting point than asking how much AI we should use. Because the value of AI is not in how much human work it can remove. It is in what becomes better when we choose, deliberately, what should change and what should remain human.