For marketing teams, going from an idea to a full campaign can be time-consuming, difficult, and expensive. So Adobe Researchers collaborated with the Adobe Experience Manager (AEM) team to create Project Get Savvy. This experimental prototype lets marketers choose a team of AI agents who help them quickly move from ideation to a full marketing plan. The final result includes compelling content and visuals along with a strategy for using the best channels to reach target audiences. AI agents support marketers every step of the way, helping teams save time and costs as they build powerful campaigns.
In March, the Adobe Summit audience was treated to an early peek at Project Get Savvy during the Sneaks.
Building human-centered AI to help marketers design campaigns and get vital feedback
From the very beginning, the goal of Project Get Savvy was simple: to give marketers easy access to feedback so they can speed up the creative design process. After all, “It takes a lot of time and money to gather stakeholders and review a new marketing campaign,” explains Doga Dogan, Adobe Research Scientist and one of the main contributors to Project Get Savvy. So Dogan and his colleagues decided to use LLMs (large language models) to create AI agents that could represent the perspectives of target audience members and relevant experts. Then, marketers could easily put together a team of AI-powered experts for a real-time feedback session.
One of the big inspirations for the work was an Adobe Research intern project that was unfolding last summer in San Jose—right around the time Dogan began working with Adobe Research. Summer intern Donghoon Shin was developing AI agents to critique poster design with Research Scientist Gromit Chan. His idea was to allow users to upload a marketing brief, which the model would use to automatically generate AI-simulated customers who could share their thoughts on the poster’s design. The designer could ask questions, brainstorm, and get feedback from different viewpoints. During the session, the system would mark up the poster with suggested changes and the designer could click to automatically make the changes.
When Dogan learned about the project, he thought of his AEM product team colleagues who work just down the hall from him in Adobe’s Switzerland office. “Customers rely on AEM to empower their webpages. And they want to optimize those pages to make them more usable, faster, more interactive, and more engaging so they can attract more customers,” explains Dogan. “I thought that, since webpages are actually quite similar to posters in terms of conveying information and asking people to take action, I should talk with the AEM product team about how this Adobe Research project could help them.”
Dogan was also tracking an intern project in the Switzerland office. Intern Niccolò Toccane was working on data analysis to help users identify low-performing areas of their webpages—those spots where people weren’t clicking and engaging very often. Dogan realized that the two intern projects were pieces of the same puzzle, so he began working to connect them into an early-stage prototype.
The result was an early version of Project Get Savvy. It could automatically find the areas of a website that weren’t getting enough engagement and generate a set of relevant personas to give feedback on the issues. Then, a marketer could choose which personas they wanted and engage in a conversation about possible variations—from copy tweaks to layout changes—to help improve the page. From there, marketers could choose the suggestions they liked, add them to a queue of variations, and explore different iterations.
From Research to Summit Sneak
The team submitted their project as a possible Summit Sneak and when it was selected as a semi-finalist, they kept innovating and polishing the technology for a live, on-stage demo at Summit.
Project Get Savvy now takes a marketer through the entire design process, from ideation to a final marketing campaign or product, with a team of AI agents that represents potential audience members and subject matter experts.
“Marketers can choose audience simulation agents—for example, they can simulate the feedback of Baby Boomers versus Gen Z visitors. And they can choose specialists in areas such as social media, copywriting, SEO, and more, so they can get exactly what they need,” explains Dogan. From there, an AI moderator leads the conversation, so a user can understand what the AI agents are doing and participate in the whole process.
On the Summit stage, the team presented a demo to illustrate the rapid and collaborative creation process enabled by Project Get Savvy. In just minutes, a marketer was able to work with five different agents in real time to turn a rough campaign idea into approval-ready content—a process that would typically take weeks.
“This is only my sixth month at Adobe and I’m so grateful for this multicultural, multi-national company with all sorts of creative people and talented engineers and researchers who come together to connect the dots to make an impact,” says Dogan. “With Project Get Savvy, we’re addressing big research questions about how to make human-centered AI, which is why our agents explain their thinking and are transparent for users. It’s one way that we’re thinking about how to put marketers and creatives in the driver’s seat.”
Key contributors
Presenter: Ronald Oribio Ruiz
Adobe Research collaborators: Gromit Chan, Doga Dogan, Donghoon Shin (Adobe Research intern), and Niccolò Toccane (AEM intern)
Additional collaborators: Silvia Mulet Ferre, Audrey Kho, Chloe Liang, Dej Mejia, Tobias Reiss, and Guliz Sicotte
Wondering what else is happening inside Adobe Research? Check out our latest news here.
