Using AI, and using it well

Everyone in marketing has the same tools open in a browser tab, which means the tools have stopped being the advantage. Graeme Stiles, CEO and Chief AI Officer at Algorithm Agency, says the gap that matters now opens up somewhere else entirely.

The misconception that costs the most

Graeme Stiles founded Algorithm Agency and is also its Chief AI Officer, a title we had not met in this market before. The agency is building a second arm in AI transformation and capability training, so his week goes into working out what is genuinely worth selling other businesses about AI, and what is noise.

We assumed a Chief AI Officer would be all-in on the tools. The first thing he did was name the belief he spends most of his time correcting. "The biggest misconception is probably the one around it's just going to replace people or replace your agency or replace any service provider." His correction carries a condition most of the hype leaves out: "It can be a significant productivity enhancer, can give you essentially superpowers in your space, but you still have to be the expert. You still have to know what you're doing."

The businesses quietly rolling their spend back, in his reading, are the ones that treated it as "a plug and play technology". Here it is, go use it. What comes back is a strategy that arrives as fifty pages out of a chatbot and stops holding up the moment anyone digs into it.

Two divides, and only one of them matters

"There's the massive divide between people who are using it and people who aren't using it. And then there's an even bigger divide between people who are using it and people who are using it well."

Most of the industry's attention sits on the first divide, which is the one that closes on its own. Everyone gets a licence eventually. The second one is the career-shaped gap, and nothing about buying a seat moves you across it.

Our own analogy for why marketers keep falling into it is uncomfortable but fair. Everyone thinks they understand marketing, because everyone sees ads, so everyone assumes a chatbot plus an opinion is enough. Nobody tries that argument with the finance team. You would not fire the CFO because you now have a ChatGPT licence.

Map it out before you hand it over

His fix is a prompt habit, and it is deliberately boring. Before you ask the tool to do a job, work out how you would do that job yourself: the steps you would take, the tools you would reach for, what a good answer even looks like.

"Give it that context, give it that download of your brain. And that's where you start getting the value. Don't give it permission to think for you. Don't be lazy is kind of a good way to say it."

The division of labour that follows is clean. You keep the strategy and the process. The grunt work goes to the model. Skip the mapping step and you have quietly accepted whatever the model decided the job was, which is a decision you never made and cannot defend.

The same test applies before you build anything. Asked how to know whether a chat agent or a dashboard is worth constructing, he goes back to the same place: how would you do it as a user, before you get your AI to do it or get your team to do it?

Where the value actually comes from

None of this is about the tool being better. It is about what you bring to it. "Plug in your brain, your systems, your workflows into AI, you'll get a better result versus just kind of using it like Google."

Which is why he is unconvinced by the licence-and-hope approach to rollout. It "can't just be, we've got a ChatGPT licence and we use it like Google". The work is understanding how the thing plugs into the business ecosystem, and that is an operating question rather than a software one.

His advice on data follows the same shape: start with what you already have, which for almost everyone is Google Analytics and Search Console, then layer a CRM on when it arrives. Waiting for a perfect data warehouse before you begin is just a longer way of not beginning.

Try this week: take the next job you were going to hand to a model and write the three-line version of how you would do it yourself first, then paste that in ahead of the request. Compare the output to what you would have got without it. And find one place in your workflow where the tool is quietly producing average work, and decide whether it belongs there at all.

Sourced from recorded, on-the-record interviews on the South African Digital Marketing Podcast. Every practitioner quoted above is named in the text.

First published on the SA Marketers' Circle.