What is the true state of AI adoption in business today? In this unscripted keynote from the Mediaspace.global Leadership Club & Mixer, James Welch (Growth Lead at WPP Choreograph Tech & AI Consulting) breaks down the realities of artificial intelligence using a classic cinematic lens: Sergio Leone’s The Good, the Bad, and the Ugly.

From the dangers of AI hallucinations and the impending "slopocalypse," to the financial traps of legacy "celery tech," James shares candid, behind-the-scenes insights on how companies are actually using machine learning. Discover why true AI success requires more than just buying off-the-shelf tools—it demands prior preparation, solid guardrails, and a privacy-first approach to data.

Recorded live in London at an exclusive roundtable of top marketers, media lawyers, and business innovators.

🎧 Listen to the full conversation on the "After Dinner Chats" podcast wherever you get your pods!

📌 Chapters / Key Moments:

• 0:00 - Introduction & The Sergio Leone AI Analogy
• 03:45 - The Good: Prior Preparation & Privacy-First Workflows
• 06:45 - The Bad: Bad Actors, Hallucinations & The AI "Slopocalypse"
• 13:05 - The Ugly: Legacy Systems & The Danger of "Celery Tech"
• 15:15 - Conclusion: Chasing the AI Gold

🔗 Links & Mentions:

• Mediaspace Global: https://mediaspace.global/
• Video Production: Baked Bean Media
• Special thanks to our hosts & sponsors: Hamish Sandison, Kinga Incze, and Cornelia Reitinger (SAS).

#AIAdoption #ArtificialIntelligence #JamesWelch #WPP #TechConsulting #GenerativeAI #MachineLearning #BusinessStrategy #MediaspaceGlobal #AfterDinnerChats

James Welch growth lead at WPP tech and AI consulting. Hamish thank you. And thank you Kinga for the invitation. And lovely to meet everybody here today at the Media Space Leadership Club. My background so I've lived Sydney, Singapore, Dubai and then back in London here where I grew up over the last 20 years and seen various things in different places. So while I was in Singapore I ran a machine learning media buying business across the Asia Pacific for WPP back in those days. And then moved to Dubai to help people get ripped off less by Facebook and Google. I've whilst being on my travels I set up a dinner party event just so that I could meet local people who are running different exciting things in different parts of the globe. And then on the back of that 12 months ago I started doing a podcast with people that I'd had at the dinner party and that was called after dinner chats. And the reason for doing that was I meet some really fascinating people at dinner and it'd be great to share those people with my network on LinkedIn and elsewhere. And it's with those 27 conversations over the last 12 months that I draw some of the insights and some of the inspiration for today's talk which I'll talk to you about.

So the theme the good the bad and the ugly. I don't know if anybody here has seen the 1966 film, Sergio Leone's film. Has anybody here seen that? Well if you have seen it you'll remember that haunting music by Ennio Morricone that score as they have that standoff. So I was thinking about what am I going to say here I just had that over and over in my head. The quest of that film is to find the gold. That is what we're using AI for today.

So you'll remember that there was Blondie that was Clint Eastwood, Angel Eyes and Tuco for the good the bad and the ugly. But the good wasn't really that good. He was just very very well prepared. Angel Eyes was ruthless and took no hostages and was very happy to cull whatever needed to be culled to get the information that he needed to find out where that gold was. The happy go lucky Tuco was the ugly guy who was really fun but hapless because he wasn't well prepared. And in today's world that's my thesis is that we need to be really well prepared. Because otherwise we're using AI to grow something that is not well formed in the first place. And as they say if you don't know where you're going all roads look just as good.

So going to the good. He wasn't that good as I was saying. And it's not that glamorous to talk about preparation. And what were the seven Ps? Prior preparation prevents poor performance I think they said in military circles. And that's what we need to work out how we get the workflows understood. If you're trying to automate something and you don't know exactly what you're automating you can make it go faster but you're not getting to the right space in the right place and the right time.

If you're looking at different data sets and how you can then start to mix those data sets and you're trying to pull them in from different places. You have privacy issues. So if you prepare yourself very carefully and look at where those data sets are and then work out how you can set up the tools in the right way and you can then have a federated learning from where you can then see what is in the different places. We can start to have a privacy first solution within the automated arena. And I use the word automated as much as I do artificial intelligence because it's the first step. Because to Philippa's point we talk about agentic AI flippantly nowadays and the large language models that we're all experimenting with day in day out. But back to the point that I was saying before I was using what we now call artificial intelligence when it was just the machine learning piece of things to look at the algorithms to see what works so we could optimise the budgets for media buying when you're spending a huge amount of money how do you optimise that in one way or the other.

So the way to be good is to be very very prepared. Now how Blondie in Sergio Leone's film was very very prepared right before the final scene he stole the bullets out of his competitor's gun. And if you're going to go to a gunfight and you know your competitor doesn't have any bullets in their gun that's pretty good preparation. It's slightly underhand does that make him good? Well you can decide on that.

So the good is about the privacy first making sure that the practice that you've got is evolving. And I nicked this line from one of my colleagues who heads up commerce globally and I love this line. Practice evolves and promise holds. So practice of what you're doing you need to change and iterate within the changing environment that we're in. But the promise and the premise that we will do the right thing with your budgets, right thing with your data that must hold. So we need to keep our ethics, our morals, and our promise to our partners up high whilst evolving within the world that we're in so that we can then have a look at what is good and right for now.

Meanwhile the bad is all around us. Angel Eyes, Lee Van Cleef was the torturer who had no qualms taking hostages and torturing his way to find out where that gold was buried. We are also surrounded by some bad actors. It's those bad actors and I think those bad actors have made us better because we've learned to work around those bad actors to make sure that we put up the guardrails to make sure that those things don't happen. And we're putting up those guardrails everybody around this table whether we realise it or not are putting up guardrails now to make sure that we work within the constraints.

So when we were talking Nick you were talking about how to stop information leaking out or hallucinations. The hallucinations for example is something that one of the people that I interviewed on my podcast Kate Devlin who's a professor of AI at King's College London was talking about how AI just makes stuff up.

I've got a question for you here. Has anybody got an idea of what would happen if you asked your LLM you know whether it's ChatGPT or Gemini or one of the others Claude if you think of a number, random number. Give me a random number between one to ten. Any idea how that works? Indeed. Number seven is the right answer fixed model. Because it's not going to answer the question of what is the random number but what is the best random number that it could possibly give you. And the best random number that it could possibly give you for various reasons that we can all hypothesise would be the number seven.

And so if it's making stuff up and there are limitations to what those tools are doing and they are going to feed back the you know their biases already within the as you if you wanted to make sure that your ideas are put out onto all of the LLMs make sure that you get a lot of people posting stuff onto Reddit because Reddit is the most obvious place for LLMs to be getting their information from today. And then so you need a lot of human beings in a lot of different parts of the world posting the right stuff and that's a whole different type of public relations.

We putting up guardrails. The UK Sovereign Wealth Fund has put up a budget right now to fund ways for people to work out what are the right things creatively and from a business sense and then to promote those ideas across the board. We need to stop and I came across this lovely expression the slop I can't say it slopocalypse. So the apocalypse, the slopocalypse. There is a lot of slop being generated and I am guilty of this.

There's at conferences right now people are standing up on the stage to be very excited about the latest ad that they have created. And the ad that they've created purely with AI is beautiful but really really not memorable. And we're not there yet and it's still adding to the slopocalypse and I've I'm saying it more often out loud and I'm getting the hang of saying it now.

So I think there's a lot of things that could be made better and that we are making better. So the bad things that are out there we're working out how to make sure we work around them. Does anybody here have an Alexa at home? If you ask Alexa something that is out of its realms the same sort of thing that you might ask ChatGPT it'll say I'm sorry I can't help you with that. And it's really frustrating and my eight year old told me how he thinks that Alexa is a bit stupid. And you know Alexa Plus is coming soon well we're told I don't know when.

So there's a lot of bad out there and the hype cycle is Gartner's hype cycle if you remember that one where it starts with a trigger, a technology trigger and then it goes up to the peak of inflated expectations then you fall down into the trough of disillusionment you got the slope of enlightenment plateau the plateau of productivity. That was in 1999 that Gartner started plotting different technologies across the hype cycle. And of course the hype cycle is really dangerous because while you're hyping things people are starting to believe things. So people will go in and ask their LLM a question and believe the answer that they're given back. If they asked their consultants the same thing they would you know question that consultant over and over and again. So people are starting to believe what they're getting in a quick response rather than things that they're paying for elsewhere which is a pity but we're changing those things because we're putting the guardrails in place and we're all getting smarter with what we're doing.

So we will learn how to avoid the torturers and the bad actors that are out there.

The ugly. Tuco was played by Eli Wallach and he was entertaining but he doesn't double check before he goes into the fight. And that is one of the problems that we're living with at the moment is the legacy tech. I worked with a chap in Sydney who talked about this some 15 years ago he talked about celery tech and I love this expression. Celery is the piece of food that you can eat that gives you no sustenance. And it's great for dieting because you feel like you're eating something clearly I haven't had enough of it recently. Celery tech is the tech that you pay good money for and doesn't do anything for you but you're paying good money so you tell everybody how good it is.

There is legacy tech and celery tech that is out there and we need to be aware of what is there and work out how we use the tech that is there, the tech that we paid good money for and how it can integrate well with the tech that is coming round the corner. We're putting a lot of duct tape up at the moment as an industry. But there are better ways to do things and it's to work with some really bright people who know how the different technologies can work together and how they can integrate together. And I think that's an important part of working out what the workflows are, being prepared and making sure that everything that you're doing is clear and that you are double checking to make sure that there are bullets in your gun.

The other thing I wanted to just say about the ugly side of things, it's not AI that made business ugly. AI just puts a light on it. And so we need to work out how to make everything work better so that we can get there faster. Back to that line of if you don't know where you're going all roads look just as good.

And my concluding point is that everyone is chasing the same gold right now. The bad will torture the system, the ugly gets exposed and the good does the quiet work but the others have skipped. Thank you.