Three questions in one: the AI question(s) every agency boss can't dodge

By Sarah Keith, Managing Director, Involved Media

The trouble is that this question is really three very different questions wearing a giant trench coat.

I get asked some version of "so, what are you doing about AI?" most weeks. Whether it's clients, team members, pitch doctors, or other agency rivals over coffee or a meal, it's the lead-in to something that inevitably turns into a comparative therapy session.

And every time, I notice the question is doing a lot of work it shouldn't be allowed to do, because it's actually three separate, quite different problems bundled into one giant one.

Understanding the three is important: there's AI in our own businesses, whether it's the tools, the workflow, or the agents quietly simplifying some of the more monotonous or repetitive tasks in the office that we all used to do.

 There's AI as a discipline we need to understand well enough to advise clients properly eg. marketing to the machines, not just with them. And then there's AI as a media channel in its own right, the thing everyone's suddenly whispering about when Reddit and Quora come up in the same sentence as media plans.

For many agency leaders, it is difficult, if not impossible, to do all three well at once. So which one earns the time, the budget and the headspace first? I don't think there's a universally correct order, but I do think there's a wrong way to sequence it.

First: fix the plumbing, but don't automate a mess

The obvious starting point is turning AI inward - emails, media confirmations, brief writing, brief response, campaign optimisation. The commercial logic here is unarguable. Margins in this industry have been under pressure for years, and if AI can help turn around a media confirmation or a brief response in minutes instead of hours, that's real money and real time saved.

But here's the trap that awaits us: we can't vibe-code our way into good process. Handing a messy, inconsistent, poorly documented workflow to an AI tool doesn't fix it; it just automates the mess and bakes in bad habits at a speed and scale a human never could.

Before AI delivers value, someone still needs to sit down and actually design the process properly: eg. what does a good brief response look like, what does a correctly checked media confirmation actually require?

That's the uncomfortable bit. AI implementation in-house isn't really a technology project first. It's a process and expertise project that technology then accelerates. Skip the process design, and you haven't automated efficiency; you've automated your worst habits and given them a much bigger platform.

Second: know enough to be dangerous, on our clients' behalf

The second front is trickier to sequence because it's less visible, less measurable and constantly evolving.

Marketing to machines is a genuinely different discipline from marketing to people, and it's evolving rapidly. What actually moves the needle when large language models (LLMs) are increasingly the layer between a brand and a consumer's decision?

At the recent IAB AI and Search Summit, we were reminded:

SEO = is getting your shop onto the main street.

GEO = is getting your shop recommended by a trusted local guide.

We need to ensure LLMs are the trusted local guide, helping the right consumers find our clients' brands and representing them the right way.

There is no single source of truth; AI stitches one together, and we have to remember that there are fewer human metrics than machine metrics.

Kantar's Karin Du Chenne explained that agencies and marketers should keep the acronym ACCE front of mind. 

ACCE stands for: Authoritative, Concise (no corporate speak), Conversational and Evidence-based.

I think this framework works because it highlights that we need to make our clients' websites answer-ready for machines, which means they need the right hooks and links.

It lets us ask key questions like: Is it structured data? Are there earned mentions in the right third-party sources? Is it simply being the kind of brand that gets talked about clearly and consistently online?

Many clients don't know, and I suspect many agencies are still working it out too. The risk of under-investing here isn't abstract; it's the same risk print faced when digital arrived, and some publishers assumed their audience and their authority would simply carry over. It hasn't, at least not for everyone.

This differs from our internal efficiency project because it demands genuine expertise, not just tool adoption.

Third: understand the channel before you buy it

Which brings me to the bit generating the most noise in the industry but probably with the least clarity: AI as a media channel.

Why is everyone suddenly talking about Reddit and Quora in the same breath as media planning? Because those platforms, rightly or wrongly, have become disproportionately influential sources that LLMs draw on when forming answers. 

This means a mention, a thread, a well-ranked answer on those platforms can carry weight in a way a traditional display ad buy never will and probably a comment on a well-trafficked site could never achieve.

Whether that's a genuinely durable media channel or more a moment of platform-related opportunism is an open question. I'd be wary of anyone claiming total certainty either way.

This is the question I'd argue matters least right now for an independent agency, not because it's unimportant, but because it's the most immature and the most likely to evolve shape entirely. Understanding it is essential. Building a media product around it today, before the ecosystem settles, is a bet, not a long-term viable media strategy.

So which question do you tackle first?

If I'm honest, I'm prioritising the second question: understanding AI deeply enough to advise clients well is the question many of my peers and I have rightly focused on, even before we've perfected our own internal tooling. Clients are asking us the AI question right now, this week, regardless of whether or not our own house is fully in order.

An agency that can't speak with authority on marketing to machines risks becoming irrelevant to the conversations that matter most, regardless of how efficient our internal brief-writing process becomes.

That doesn't mean ignoring the internal efficiency work; it means running it in parallel, properly, with the process discipline it deserves, rather than chasing the media-channel story that generates the most headlines but the least certainty.

Turning AI into an advantage for a small agency isn't a technology question. It's a sequencing question, and getting the order wrong is the easiest way to spend a lot of money proving very little.