Andrew Miller
- Episode Four, an ad agency in New York, created an AI tool called RYA to generate ideas for clients.
- The agency is leaning into the more nonsensical "hallucinations" to keep the ideas fresh.
- The tool has helped with ideas for clients in the financial services, auto, and travel sectors.
This as-told-to essay is based on a conversation with Leslie Walsh, head of product strategy and development at the New York-based advertising agency Episode Four. It has been edited for length and clarity.
The biggest thing about creativity is that ideas need to feel original and unique. I think everyone in the creative industry fears that if we're all using ChatGPT, we're going to have the same ideas — that we'll get into a sea of sameness. It's a race to the bottom.
A lot of people come at creative ideas in different ways. We like to say that we come at it from a place of, "Where do my customers actually want to engage?"
We needed something to base our ideas on, so that clients just didn't say, "Where is this coming from? Is this just something in your crazy brain? Is this something that's actually going to help drive my business?"
RYA is our creative AI tool. We position it as: radical ideas that are acceptable to your audience, because they're all grounded in data.
We put together a weekly survey that asks Americans: If you had extra time or money in your day, how would you spend it?
We serve up 180 genres and 20 different actions, things like "I want to go on a trip" or "I want to go out to eat." And then the genres add a little bit of nuance: "You said you like to go out to eat. What kind of food do you like? You'd like to go on a trip. What kind of traveler are you?" That's where we can really pinpoint passion points.
With that dataset, we then figured out how to train large language models to be creative and come up with ideas just as if our teams were coming up with ideas on their own — but at a rapid pace. We found that Anthropic's Claude is the best at generating creative ideas, but we use different ones for different things.
Typically, in an agency project process, you work with the clients, and then you work on the brief. You go back and forth, back and forth. You brief your teams, and then it takes a few weeks to craft those ideas, shape them, and get them to a good place. Then you present to the client, go back, revise, and come back again. That could take six to eight weeks, usually.
We have streamlined that process down to a couple of days. We like to say we can do it all in one day and, if our clients have time, actually do it in a couple of minutes. It just depends on how much you want to shape and mold the idea.
It's really helping us get to a good starting place, and it's solving for the bigger picture. It's less of that executional layer that I know a lot of creative tools are in the market right now.
RYA is not going to help you build a banner ad. RYA is not going to help you put out a social post. RYA will help you come up with a big idea that could then turn into a banner ad or social post.
Leaning into the 'weird' and 'nonsensical' AI outputs
We really like to lean into the weird combinations, the nonsensical places, the less obvious places that you might want to start your ideation from, because we believe that that leads to better work.
The way that you get AI to be a little bit more nonsensical is in your prompting strategy and the way you adjust the temperature settings. There are ways that you can give it guard rails to try to make sure that it's not going to hallucinate in a way that's going to get you in trouble, but hallucinate in a good way. You really need strong inputs. As we say, garbage in, garbage out.
For a financial services client, we turned financial education into a celebrity chef cooking series. And at a recent financial industry event, we turned product sales materials into custom pressed EDM vinyls.
For a travel and leisure client, we launched a cruise ship with a global digital treasure hunt that hid Google Maps clues across the web —transforming a product demo into a game.
For an auto lending company, we created a dating show that matched singles with both a car and a partner.
Everyone's coming up with different AI tools for different purposes, different verticals, and domain expertise. I'm not an engineer, but I do have a lot of domain expertise in terms of how to think about the best way to brief creatives, to work with clients, to crack a really tough brief, to conduct research, and get interesting insights.
Then, along with my creative partners, who have much better domain expertise in actually generating ideas, we put all of that input into a large language model so that it can replicate our thinking.
That is where we're headed. That is what's going to be exciting about this industry. There are opinions out there about whether AI can truly be creative. We believe it can.