SecretSauce on AI Marketing Companion with Sandy Carter

Every marketer now has access to AI. Every brand can generate content at scale. So why does so much of it feel exactly the same? Or worse, feel like it belongs to someone else's brand entirely?

In a recent episode of The AI Marketing Companion, host Sandy Carter sat down with Simon Davis, Co-Founder and CEO of wearemighty.ai, to unpack one of the biggest challenges facing marketing teams right now: how to stay original and on brand in a world flooded with AI-generated content.

Before founding wearemighty, Simon helped create more than 45 games played by millions of people through partnerships with Apple, Disney, and Razer. Then came a brief that changed everything: 25 million branded assets. The challenge exposed a problem most organizations are only beginning to recognize. AI can create content endlessly, but maintaining a brand's identity at scale is a fundamentally different problem.

The brief that broke everything

Back in 2022, Simon's game studio was trying to generate unique hero characters across millions of possible combinations. They turned to the earliest generative AI tools - Stable Diffusion, Midjourney - because there was simply no other way to hit that scale.

What they found was that nothing held together.

"You might get one asset you're happy with and then when you regenerate, something just breaks. The clothing is all of a sudden wrong, or the shoulder would be off. And even if you regenerate the same asset, the faces start to drift and get artifacts."

The team spent three years building internal tooling to solve the problem. When they demoed it to a friend at one of the world's largest advertising agencies, the response was immediate: "Wow, this is amazing. Can I have access?"

"I was like, 'Uh, no. This is an internal tool. We have no way of being able to support you and run this.'" But the interest kept coming. "We realized actually that this was a much bigger opportunity. This is a tool that could help not just our business but literally any business out there that has any sort of branded assets, social media assets, any kind of content needs."

That internal tool became SecretSauce.

The triple bind every marketing team is facing

Simon broke down the pressure marketers are under right now into what he calls "the triple bind."

1. The first layer is increased expectations. "In the era of AI, every business leader is under pressure to generate more content than ever and to cover more surfaces than ever." He described a meeting that same day with someone in London asking for help creating content for Chinese social networks - because in the era of AI, leadership expects you to be everywhere.
2. The second layer is that there are only two ways to meet that demand, and both are bad. "You can scale with AI, but off-brand AI will create stuff that doesn't look like your brand and diminishes your brand value. Or you can throw money at the problem and have hundreds of humans just churning out as many assets as you can." Either way, "your profitability goes down because your input cost is going up or you're devaluing your brand."
3. The third layer is competitive. Simon called it "the brand equivalent of the Manhattan Project.” Every brand owner racing to be the first to accurately deploy AI at scale within their vertical. "The one that does that is poised to validate hundreds of thousands of different variants and then find the creative that really works for them and capture the market."

The ROI trap

AI made content generation fast and cheap. But Simon argues most teams aren't measuring the real cost.

"A business that spends loads of money on creators, but only two videos out of 10 are actually usable. If you're spending $100 per video, then actually your effective cost per video is $500 plus the review time. But people are not really thinking of AI in these terms. They're just thinking of the input cost, but the real human cost and business cost is much much higher."

Sandy reinforced the point with examples from the wild. Nike recently caught heat for an ad that leaned too hard on AI-style language. Another company ran a trust-focused campaign using AI-generated actors - one of whom had six fingers. The comments were brutal: how can I trust a company that can't even use a real human?

"It doesn't really matter how big you are," Sandy said. "Whether you're Nike or a smaller company, there is a lot of risk to doing this."

Brand Drift is the silent killer

Sandy brought up a viral example: someone took a photo of Dwayne "The Rock" Johnson and asked AI to simply redraw the same picture, over and over. By the end, "he basically turns into an egg,” and one of the final versions had him wearing a blue Marvel costume that had nothing to do with the original image.

This is brand drift in action. And it's what happens when AI tools reference their own output instead of a fixed source of truth.

"Generic big models, they are fine for creating one-off work, but they will not look like your brand. They will not retain what your brand's memory is," Simon said. "I wouldn't put my personal brand or our company brand in the hands of a tool that's built for everyone. I would use a solution that's either been built with my business in mind or built with our specific use case in mind."

SecretSauce's answer is what Simon calls the Brand Brain - a system of record that holds the logo, product subsections, tone of voice, audience, values, layout rules, dos and don'ts. "When it keeps regenerating content for you, it's not using the previous photo and generating a new asset based off of that. It's always referring to the source of truth."

The platform also has a Brand Guard, computer vision checks on both the input and the output. "If the output is wrong, without involving you, the agent will go back and redo their work and keep redoing it until it passes the brand guard check."

"It's a machine. It's automation. There are still times where it comes out and you're like, 'No, I don't like this.' And you'll send it back again." But the guardrails dramatically reduce waste and drift.

The 80/20 rule of AI content

One of the most practical takeaways from the conversation was Simon's framing of where time should actually go.

"I always tell people you want to be spending 80 to 90% of your time feeding it information and only 10 to 20% of your time actually generating content. Because the more time and effort you put into feeding it, the faster and the more accurate the content will be and the less money you'll waste on failed generations."

He walked through a real example: building a sample video for a potential client earlier that day. "The amount of time I spent actually generating the video was probably 20% of the time. The other 80% was me saying, 'Here's a picture of this place. These are the type of people that go to it.' Basically just giving it as much context and information as possible and then saying, 'Don't generate video. Storyboard this for me first.'"

The parallel to how most people use any AI tool was sharp. "AI is the world's most powerful intern, but at the end of the day, it's still an intern. The more information and context you feed it and you really make the instructions super clear, you get amazing results. But if it's like, 'Hey, give me a 30-second video for this theme park' and it doesn't know anything else, you're probably not going to be happy."

He called the alternative "the prompt lottery." "Maybe one in 100 times you get something great first time, but I don't think luck is really a great business strategy when you're trying to allocate your time."

Why humans still matter

Sandy cited BCG and McKinsey research suggesting that only 20–30% of a company's AI success comes from the technology itself. The rest is people and process. Simon agreed and pointed to a clear pattern in who gets the best results from SecretSauce.

"Anyone that has a strategy background or is a strategic thinker absolutely crushes it on there."

He also made a distinction between the work AI should handle and the work humans should protect. "Something like over 85% of agency budgets or production budgets are spent on stuff like resizing, doing localization. I call that design administration. We really want to automate as much as possible so that people can spend 80% of their time thinking of an amazing campaign or new strategies for reaching their audience, and not doing the boring stuff."

As for what AI still can't do well? Copywriting. "I'm very particular about language. I'm not 100% satisfied with our copywriting yet within the bot. I think that's still hard to get right." He acknowledged the obvious AI tells that plague everyone's feeds: the m-dashes, the formulaic openers. "Great copywriters are worth their weight in gold, and actually, like the strategist, they tend to have very good results because they can communicate very clearly and communicate their artistic vision."

Should you let agents publish on their own?

Sandy raised the question a lot of marketing teams are asking right now: what happens when AI agents don't just generate content but approve and publish it autonomously?

Simon didn't hesitate. "I wouldn't trust an agent to publish today. Maybe we'll get there, but I actually don't think that will happen anytime soon."

His reasoning went beyond quality control. "Agents are not going to be aware of the zeitgeist or things that are happening around them. It could be that something that's fine to publish one day, because of an event in the news, becomes a very bad idea. An agent's not going to catch that."

He shared a story he said he'd never told publicly before. "I had an agent running on a machine doing some analysis on social media posts. I was like, 'Just look at this, give me feedback on what's working.' And without me approving it, it just went off and posted something that was completely false."

"I think that just shows you how primitive these agents are and how bad they are at sticking to guardrails. I would not under any circumstances recommend anyone trust an agent to approve and publish anything."

SecretSauce has what Simon calls "relentless agents" that work around the clock: scraping news, generating campaign ideas, creating assets. But the human is always the final checkpoint. "The idea is that you receive that, you vet it, you give it feedback, and then you decide if you want to publish."

"Damaging your brand can be a multi-million dollar outcome. I wouldn't trust them with just publishing stuff."

Just start

When Sandy asked what one move marketers should make this quarter, Simon kept it simple.

"Just start. AI seems very scary, but just start with one thing and start with something very small. These baby steps start to compound and before you know it, you're doing all kinds of things that you thought were unbelievable, that felt like magic."

He did add one caveat: the productivity trap is real. "I've never worked this many hours. I'm probably five to 10 times more productive than I was a few years back in terms of output. But I'm also working longer hours because it can be quite addictive."