The Creative Engine

Tom Ollerton, Founder of Automated Creative, on why best practice is just copying, how brand and performance are one team not two, why you only get data on what you test, and what eight years of building an AI creative platform has taught him.

Listen to the episode
Season 5, Episode 91

"AI will come down a pipe like electricity."

Eighteen years from Gumtree intern to founder of one of the most distinctive ad-tech platforms in the industry. Tom Ollerton on why best practice is just copying, how he and Alex Hobhouse founded Automated Creative seven years before AI was good enough to deliver on the thesis, what 3000 ads a week looks like in production, and his counter-pattern case study where the best ad targeted at mothers was a dad.

Tom Ollerton was a failed musician until 28, then a kitchen chef. A Gumtree intern listing at a mobile agency five minutes from his house became Head of Sales two weeks later, then roles at VI and Skive, then six years at We Are Social through Marketing Director and Innovation Director. In January 2018 he co-founded Automated Creative with Alex Hobhouse on a Kit Nicholas keynote line that stuck: AI will come down a pipe like electricity.

In this conversation with host John Horsley, Tom argues that the creative-versus-performance split is a managerial construct and a customer reality of one. He explains why his platform tags every ad with strategic markers so the CMO can ask the dashboard like an LLM why ads worked, walks through the baby-formula and Jack-Daniels case studies that demonstrate the counter-pattern playbook, and lays out a sober economic argument on the generative-AI cost curve and the question of where the money comes from when the price runs to the roof. The hopeful note: a creative industry where there is no physical limit on what an idea can be.

Tom's career arc: failed musician, kitchen chef, Gumtree intern at a mobile agency, two weeks later Head of Sales, then VI, Skive, six years at We Are Social through Marketing Director and Innovation Director, and Automated Creative co-founded in January 2018 with Alex Hobhouse and Rhoda Sell as co-CEOs.
Best practice is copying. Everyone copies. Therefore everyone's ads look the same and don't cut through. AC's job is to test the variants no one would have green-lit. The baby-formula case: best practice would put mum and baby in the ad. AC's data showed the best-performing ad to mothers was a dad with the baby.
Brand and performance are one team. The funnel is a model and like all models it is flawed. Brand thinking belongs in performance creative. Performance thinking belongs in brand creative. AC's technology delivers both at scale and tags every ad with strategic markers so the CMO can query the dashboard like an LLM.
Hungry algorithms. Meta, Google, retail media, digital OOH, email; every platform needs fresh inventory constantly, across every market and language. One person at one of AC's entertainment clients runs 3000 brand-new ads a week across 50+ markets, five languages, five platforms.
A sober economic note: ChatGPT-class AI businesses are spending five dollars to make one. Most of the money funding the build is coming from Gulf sovereign wealth funds. When the bill comes due, the price of AI-generated creative will rise. Brands relying on cheap generative output should ask themselves what happens then.
01Failed musician to kitchen chef to Gumtree-found intern to ad-tech founder in eighteen years
02Best practice is copying: the baby-formula counter-pattern and why everyone’s ads look the same
03Brand and performance as one team: bringing performance thinking to brand and brand thinking to performance
04Strategic tagging at scale: making the CMO dashboard answerable like an LLM
05The generative-AI economics question: who pays when ChatGPT spends $5 to make $1?
Key Exchanges 05
01 From kitchen to founder.

"I was a failed musician until I was 28. Then I got a job as a chef. Then a Gumtree advert."

Tom’s entry point to advertising was a girlfriend explaining her job designing a user experience for a Java app for Guinness. A few weeks later he answered a Gumtree intern listing at a mobile agency five minutes from his house, walked in and promised he would make himself indispensable. Two weeks later he persuaded the founder, Damian Kimmelman, to let him be Head of Sales.

02 Best practice is just copying.

"The best thing you could put in an ad targeted at a mom was a dad."

A baby-formula brand tested every conceivable variation through AC. Best practice would have produced a mother with a baby. The winning ad to mothers was a dad with the baby. The Jack Daniels example was the same shape: the messaging that drove sales for whisky as a gift was apology framing. Sorry. Make it up to them with whisky. Counter-pattern beats pattern-following because pattern-following is the safest way to underperform.

03 Brand and performance are the same team.

"Bring performance thinking to brand and brand thinking to performance. They are one team. Not two."

The funnel is a model. Like all models it is flawed but useful. People become aware of brands through performance ads. People buy products from awareness ads. Tom’s argument is that you need a platform, data and a strategy that lets both motions run through the same creative pipeline. The first second of a YouTube ad has to be optimised or no message lands and no memory structure builds. Performance discipline applied to brand work.

04 Hungry algorithms.

"One entertainment client. 3000 brand-new ads a week. 50+ markets, five languages, five platforms. One person."

Every platform now has its own creative format and they all need fresh inventory constantly. Meta, Google, retail media, digital OOH, email. AC's thesis is that no manual or partly-manual process can keep up with the volume at compliance, which is why MARS Petcare runs nine brands across five regions through the platform. Designers get into the industry to design. AC takes the variance and scale work off their plates and lets them do the beautiful work.

05 The generative-AI economics question.

"What happens when the price goes through the roof?"

Tom’s sober question on generative AI: ChatGPT-class businesses are spending five dollars to make one. Most of the money funding the build is coming from Gulf sovereign wealth funds. At some point Microsoft, Google and Meta will want their money back. The price of AI-generated creative will rise. Tom’s prompt to anyone using vibe-coded or vibe-generated marketing: if you paid full price and waited weeks for the output, would you still be happy with it? Most people say no.

32 Minutes
S5 E91 Season & Episode
8 yrs Running Automated Creative Since January 2018
3000 Ads a Week Run Through the Platform by One Person at an Entertainment Client

"The best ad targeted at a mum was a dad."

Hear Tom on
The Business of Marketing
Season 5 Episode 91
More Episodes
Full Transcript SEO & AI indexed
Season 5 E89  ·  Tom Ollerton, Founder, Automated Creative
Lightly edited for readability.

Host What first sparked your interest in advertising?

Ollerton I was a failed musician until 28. Then I got a job as a chef. I ended up dating a girl who worked at a mobile agency. I asked her what she did and she said she had designed a user experience for a Java app for Guinness. I realised you could work in a creative industry and earn decent money. I applied to agencies and kept failing. Then a Gumtree advert came up for an intern at an agency five minutes from my house. I walked in and said give me the job, I will make myself indispensable. Two weeks later I persuaded Damian Kimmelman to let me be Head of Sales.

Host The moment you decided to start your own business.

Ollerton South by Southwest, years before. A man called Kit Nicholas said AI would come down a pipe like electricity. That stuck. Years later Alex Hobhouse and I were running an event series called I'll Be Back, on the intersection of creativity, ads and AI. We’d have a brand, an agency, an academic and a supplier on stage. It became obvious AI would eventually support human creativity. So we founded Automated Creative to fix the wrongs we’d seen. We didn’t realise it would take eight years for the technology to catch up.

Host What you wanted to fix.

Ollerton Removing the guesswork. The hunch. Alex was frustrated that strategy ideas would go out based on what felt right. We always wanted to combine the hard ones and zeros of data with the soft fluffy nature of creativity, so every decision we make for our brands comes back with concrete confidence. Bose, MARS, Formula One, Jack Daniels, Specsavers, Tony's, to name a few.

Host The co-founder triangle.

Ollerton Rhoda Sell is the third corner. Co-CEOs at AC. Alex gravitated to product. I gravitated to marketing and sales. Rhoda gravitated to finance and operations. We overlap in the middle on direction and strategy.

Host Is there a creative vacuum in performance marketing?

Ollerton No. Performance has a bad name with traditional creatives, but both disciplines require creativity. The funnel is a model, like all models flawed. You can become aware of a brand through a performance ad. You can buy something off an awareness ad. Bring performance thinking to brand and brand thinking to performance. They are one team. Not two.

Host When ads do not work, is it the creative?

Ollerton Meta yo-yos between thirty and seventy per cent. We are not an agency. We sit between creative and media as an ad-tech platform. If both are not singing together, nothing works.

Host The typical impact of being in the middle.

Ollerton Hungry algorithms. Every platform has its own format. New format on Google, then Meta has to have its own. The explosion of retail media. Digital out-of-home, email. Massive content for one platform, one market, one audience, one funnel stage. MARS Petcare runs nine brands across five regions through us; I do not think anyone makes more cat and dog ads than us. Designers get into this industry to design. We do the variance and the speed.

Host Mission control.

Ollerton Amy Wright, our global head of strategy, has the analogy: it is not enough to launch the rocket, you need mission control. We saved one client two million dollars on a single brand in a single market just through tagging and tracking.

Host Understanding why an ad worked.

Ollerton Nine out of ten brands do not know why their ads worked last year. With our strategic tags they can. For alcohol you tag against serve, flavour, ingredients, heritage. The CMO can query the dashboard like an LLM: what drove conversions in France on Pinterest? Flavour cues. What drove add-to-basket for Gen Z in North America? Provenance. Concrete causality instead of chin-scratching about what to do next month.

Host Real-time optimisation.

Ollerton People confuse automation with optimisation. Optimisation is the optimal thing you could do. Automated is the easiest. Traditional DCO just churns images and text without thinking. For highly compliant brands the value we bring is the strategic oversight. Marketing is one human speaking to another human. We sometimes forget that.

Host Originality vs homogenisation.

Ollerton ChatGPT is trained on words. It is a predictive text guess machine. It does not understand anything. It only knows that a glass smashes when it falls because the words say so. New AIs trained on video will have a different understanding of the world. Marketers have ingested a lifetime of multisensory data that ChatGPT has not. That is the moat for now.

Host Memorable case studies.

Ollerton Reckitt formula. Best practice for a formula ad targeted at mothers is a mum with a baby. We generated every variation: grandparents, other kids, no kids, just text. The winning ad to mothers was a dad with the baby. Counter-pattern beats pattern-following. Jack Daniels: the messaging theme that drove whisky-as-a-gift sales was apology. Sorry. Make it up to them with whisky. You only get data on the things you test.

Host Reeducating brands to zag when others zig.

Ollerton We start with speed and scale, the low-hanging fruit. Once the brand sees what we have done across all the variants, all the markets, all the platforms, we have got the data to talk about flavour, ingredients, provenance, whatever. The conversation moves from playbook to evidence.

Host The typical client journey.

Ollerton Two patterns. One entertainment client has to produce 3000 brand-new ads every week in 52 markets, five languages, five platforms. One person runs it. Another client with a heritage portfolio needs hand-holding through a process where everything is brand-safe and hitting the funnel cleanly. Two totally different journeys, one platform underneath.

Host Skills marketers need today.

Ollerton When we came up you followed a pattern. Get into a game, junior creative, two years there, two years there. Now a junior person can vibe an idea on their phone on the bus and have a functioning version by the time they get home. So you can circumnavigate the entire industry with an idea. The skill is to come up with something no one else has thought of, and then ship it tomorrow.

Host Where is the industry on the maturity curve.

Ollerton All the AI tools come down a pipe. ChatGPT, Anthropic, the rest. We are all plugged into the pipe. Every update sends everyone back to zero. Remember when Snapchat launched and you could be the agency Snapchat expert by spending half a Saturday on it? It is like that with every model update. So we are constantly relearning. As a business we think experiences not platforms now. The end of a campaign report could be a deck. Or you could vibe-code an 8-bit platform game featuring the client's product. We have the tools.

Host The economics of AI.

Ollerton Anthropic made nine billion in revenue spending fifty to build it. ChatGPT, similar shape. The data-centre cost is enormous. One estimate: ChatGPT needs subscriptions from seventy per cent of the world to break even. Most of the money funding the build is coming from the Gulf sovereign wealth funds. The Microsofts, Googles and Metas will want their money back eventually. Then the price of AI-generated content will rise. My test for anyone using generative output: if you paid full price and waited weeks for it, would you still be happy? Mostly: no.

Host The hopeful note.

Ollerton My daughter asked me to make her a platform game featuring Teen Titans. I voiced it to Claude. It built one in seconds. She said it was rubbish. I said tell it what you want fixed. She prompted the next version. My daughter is already in a world where her imagination can be realised. It is the first time in this industry we have been able to say that with credibility. There is no physical limitation on what an idea can be. Anyone who tells you they know what happens next is blind. We are all making it up. But we are in a position to be part of the story.