Don’t try this at home

I would trust an AI to take my child to school. I would trust an AI to fly me across the Atlantic and back. I'd trust an AI to help me wire an electric plug in a foreign land or help me find the best place to eat on a wet Wednesday night in rural France.  

But I would never, repeat never, trust an AI with my marketing strategy, brand planning or GTM.

I say that for two reasons, one theoretical and the other practical. 

Firstly, unlike cars, planes, plugs or maps, marketing AI is trained on incomplete data sets. Marketing is partial, it's siloed, the results are proprietary from one business to another. The only people who have a decent view of the whole show are Google and Meta, and even their understanding is partial and very last mile. 

Marketing AI was trained on self-serving HBR reviews, shonky awards submissions, agency blogs and other marketing-of-marketing content. There are no peer reviewed submissions to the Lancet that helped healthcare professionals with evidence-based decisions, no millions of miles driven in test conditions for driverless cars, no accumulated body of thousands of restaurant reviews that will save my Wednesday night dinner out. 

A friend of mine who is a very senior BA pilot once told me that every single take-off and landing a pilot makes is tracked and logged. If you have a couple of bouncy ones or misjudge your approach, someone will put their hand on your shoulder (I seem to remember it was the union in this instance) and have a quiet word that perhaps you should spend a week off frontline duties, and head into the flight simulator to sharpen things up. That's what I call a proper data source. That's why I will fly AI to New York and back. 

Where's the marketing equivalent? We just don’t have it. 

So much for the theoretical, now the actual. 

LLM-averaged marketing is shit. Forgive the profanity but there really is no better word for it. If you want to drive your business (and your reputation) off a cliff by following the uninformed entreaties of a poorly trained but highly intelligent AI, be my guest. But don’t say I didn’t warn you. 

Even if you aim off with all the habitual health warnings about disabling the sycophancy setting and never taking the first answer as anything more than an opportunity to think again harder, off-the-shelf AI simply doesn’t have enough real-world data or complete human experience to draw upon. If you’re working with AI in marketing and you haven't worked across dozens of sectors and seen the pitfalls and perils of dead-ends and blind alleys, it won't. 

If that wasn’t bad enough, then here is the worst thing of all. Creating clever prompts, agents and bots that automate a series of complex, intricate tasks, before producing what looks like a perfectly serviceable marketing won't save you if the source code is effluent. In the hands of the naïve or paying clients they can look beguiling. You read it here first but beware the rise of the prompt-jockeys. They are after your marketing money, and they will set back the cause of truly inspired AI-powered marketing five years or more, and blight careers. 

So why, you may reasonably ask, are you guys any different? I'll tell you. 

First, our business is built on the source code of 40 PE marketing turnarounds that we ran in the eight years prior to launch. It's also buttressed by thirty-five years of hard yards on brands like O2, Peroni, Battersea Power Station, Nokia, and more, across every sector imaginable. Based on that experience, our model is built around a unique framework for understanding, evaluating, and optimising a company’s whole marketing system. 

Secondly, our model sits alongside the LLMs, with a continuous feedback loop, informing and learning all the time? We are not educating the LLMs. We are building our own reasoning engine. 

Thirdly, our thinking and our products are human things, created using our real-world experience, and we use LLMs as an exoskeleton, rather than outsourced thinking machines. The LLMs work for us, not the other way round. 

And finally, we are experienced, old and wise enough to test the hell out of everything - and check whether it works or not - before we inflict it on a client. That is one of the last great privileges of age. 

AI and marketing are natural bedfellows, but they need (if you excuse the mixed metaphors) a midwife who knows which end is which. As Penn & Teller used to say, don’t try this at home.