In April the founder and the former CEO of Jungle Creations — Jamie Bolding and Nat Poulter — took their idea for an AI social agency to market.
Restless promises to ‘help brands grow by automating audience intelligence to drive more personalised content strategies.’
At the heart of the agency sits Antenna, a proprietary AI software that combines brand and public data to create synthetic audience personas, to help shape strategies and messages.
MediaCat spoke to Poulter, along with Restless’ managing director, Jamie Vaughan, about the shifting dynamics of social media and ad agencies.
What’s different about social media in an AI world and why does it require a new kind of agency?
Nat Poulter: I ran a social media company for six years starting a decade ago, and the barriers to entry for content creation were much much higher. We had to raise $4 million of investment to set up a studio. Now, we live in a world where everyone has access to tools on their iPhones and cheap SaaS products. You don’t need a massive Adobe Premiere suite to edit content. And there’s a whole host of AI tools to help think through concepts and scripts and hooks and things.
And with generative AI, my overall feeling is that there is just a world of noise now. Our position is that we’re an AI-native social agency very much focused on AI early in the process. So yes, we’ll definitely look at generative AI, but we’re thinking more about it earlier on in that workflow — thinking about AI as a creative intelligence tool.
How do you see people responding to AI-generated ad content? Do you think we’ll get to the point where they can’t tell the difference?
Jamie Vaughan: I think we’re already in a place where people would struggle to tell whether some things are AI-generated or not. That moment has arrived. We just delivered some AI assets for a luxury brand client of ours and they wouldn’t have known they were AI-generated but for us labelling them.
That’s maybe the tension that we as a business exist in more than most. There’s clearly a massive opportunity with AI-generated content in terms of cost, scale and potentially quality for many brands, and then at the same time, there’s the rise of creator-generated stuff, whether that be user or influencer-generated, that feeds the authenticity that people still crave.
Do you think social platforms will start to penalise content that is AI-generated?
Poulter: A lot of this content is driven through algorithmic discovery, right? The KPIs that underpin that are engagement based metrics and so if the content, be it AI or human-generated, is delivering more engagement, then that is going to be promoted by the platform.
Are you concerned at all about creating content for platforms that don’t appear to want agencies involved in the marketing process at all?
Poulter: I would push back and say I don’t necessarily agree that is the case. A lot of these platforms are adopting a workflow to automate the process of buying — and rightly so because AI’s really good at pattern recognition. We should be coming up with AI buying platforms that take all of the historic data and then make decisions going forward. That is the direction of travel and it’s ultimately the reason that we have launched this agency, because our view is that the value is moving higher up the chain. Media activation is almost the last point in the chain. The things that happen before that — creative ideas, creative content and strategy — that’s really what we’re focusing on.
What kind of clients will use an agency like yours?
Vaughan: We’ve definitely come to market with an ideal customer in mind, which we have described as mid-market consumer — consumer brands that have found product-market fit and are scaling or those that have been in the market for a long time and that want to grow.
I’ve been in agencies dealing with that customer continuously through the last 10 years or so, and the thing that came up in the research we were doing when developing Antenna was that brands have got incredibly poor discipline in understanding who their existing customers are. And they’ve got even worse discipline around understanding who their future customers are. And then the problem that they’re all describing is how they effectively prospect new audiences to turn them into customers that come back over and over again. So it still feels to us very much like they need an effective creative performance social agency, or whatever you want to classify it as.
In your press release about the new agency you described AI as a ‘force for good’. What makes you so confident about that?
Vaughan: I think we were probably speaking a bit more specifically around it being a massive opportunity for the sorts of brands that we’re going to be working with. There’s energy, unemployment, and all sorts of other macro-things that could come up. But for all brands there’s an enormous amount of opportunity that AI brings. What we do within Antenna would normally require a team of creatives, data analysts and developers, and very few brands want to invest in all of that, even in a low-committal way. So there’s clearly a massive upgrade to the opportunity that brands have. And consumers then benefit from having much more personalised, optimised stuff.
Why is it better for clients to use your Antenna tool rather than Gemini or Claude or one of the other main models?
Vaughan: There’s a functionality answer to that question, which is a bit dull, but I think the key point of difference is to get the same level of context that we’re able to give Antenna would take someone an incredibly long amount of time to set up. And it’s not to say that we don’t use Claude and Gemini as well for different tasks. We’ve not invented anything in terms of generative LLMs. We’re leveraging that technology by giving it enormous amounts of structure and enormous amounts of context, so that it yields far superior results.
What’s the user-experience of Antenna like? Is it conversational?
Poulter: There are different elements to it. The first part of it is immersion, where we upload the context. That could be anything from brand guidelines through to connecting to an API with Meta — so we can pull in performance data — and scraping comments, and click-stream data, so we can look at how people are navigating the content. Context is king in this new world and we’ve basically built out a framework within that immersion tab so that we can aggregate as much information [as possible] at the click of a button.
Then it goes into micro-personas, which is where we take all of that context. We understand audience clusters based on the different signals within all of those different packs of data and build out enriched micro-personas. That will give us an inferred understanding of what their demographic breakdown is, what their pain points are, what their voice is. And then it goes into the chatbot functionality where you can then start to have a real-time conversation with one of these synthetic personas, to test a concept or hook that might work on social.
All of this is to try and build hypotheses, right? A lot of this is rooted in client data, some of this is based on public data. No one has a crystal ball but what it allows us to do is generate hypotheses, to say we think this concept with this hook with this storyboard with this type of creator works, and then build a really detailed brief which then either goes to a creator, an in-house social producer or indeed our AI studio.
We start running paid spend against these assets as quickly as possible, and that data, through the Meta API, comes back into Antenna to build that feedback loop. Then we can validate whether these personas are working or not.
How accurate or effective are the personas that Antenna produces? Do you have any figures?
Poulter: The way we see it is we are validated that these personas are working and they are correct if the paid performance scales. And what we’ve already proven with a host of our clients is that we’ve been able to essentially double paid spend and maintain the cost of acquisition. So for all intents and purposes we have doubled efficiency.
Are you committed to the agency model, or are you looking at licensing Antenna?
Vaughan: It’s a question that’s very front of mind. We’ve had conversations where it could become a SaaS business. But for the first six months we’ve made a pact to keep Antenna as a managed software. There are a few elements of it that will become accessible by our clients, though.
Three out of four people that we spoke to don’t want to just take on more software costs because they are not short of that. I don’t think there’s many people in the world that are thinking, ‘We’ve got really good consolidated software bills at the moment’.
Do you need juniors in the agency, given that Antenna does a lot of the grunt work? If not, how do you develop talent?
Poulter: Different agencies take different forms. I was in one before where we were 150 people and the average age was 23, and it was hectic. We’re a team of experienced experts, and we’re trying to make sure that a lot of the heavy lifting can be done by Antenna, so then we’re bringing in heavy-hitting creative strategists and other senior people that ordinarily wouldn’t be sitting on client accounts of these sizes, so that we can deliver significantly more value.
But, yeah. There’s a creative talent pipeline that needs to be managed and so we are actually trying to work out what is our core organic ramp.
How do you manage the cost of using AI?
Poulter: Smartly… I’ll give you an example. We build out these micro-personas and we have an API into Meta’s creative marketplace, so we can beautifully match a persona with a potential creator for a perfect brand partnership. You can pull those creators and then you can start to enrich those profiles using AI. It can get very expensive to a point where it’s costing you $10 per potential creator, and there are thousands of creators. So a lot of the skill is the ability to use the right models at the right time with the right prompts to make sure that you are being as efficient with your tokens as possible. The use of tokens is going to be a skill in its own right as time goes by because we know that these LLMs chat bots are being subsidised by serious amounts of capital just to drive growth. So there’s definitely a level of nervousness.
We binned off OpenClaw (an AI assistant that can create videos from text prompts) within about two weeks because it ended up costing us a few hundred quid per day per user. Things can spiral out of all control.
Your press release didn’t mention creators. Was that an oversight?
Vaughan: I think it was implicit.
