An Adobe executive on privacy, getting customers to use new tools, AI-generated video, and more
LAS VEGAS — Adobe $ADBE revealed a suite of new AI tools at its Las Vegas conference last week. The new products’ generative AI abilities focused on personalization, such as in Custom Models, which lets companies train Adobe’s AI image generator on their brand’s imagery so the outputs stay on-brand. Other new tools can create personalized content at scale and bespoke ad campaigns for smaller audiences than ever, even down to individuals.
There are plenty of use-cases for generative AI, but marketing could be one that benefits early, said Alexandru Costin, Adobe’s vice president of generative AI.
“The reality in content production and marketing today is that ROI improvements are bottlenecked by the amount of content that can be created to do personalized marketing,” he said.
Marketing is in flux, both because of generative AI and the phasing out of internet cookies, which companies have long relied on to track consumers as they move around the web. These changes mean brands wanting to reach specific types of consumers need to find new ways to do so — but they also have some new tools to try.
Adobe is hoping to cash in on the moment and meet both evolving needs. Costin talked to Quartz about how these changes could affect privacy, the struggles to get customers to use new AI tools, how Adobe thinks about AI-generated video, and more.
This interview has been condensed and lightly edited for clarity.
Alexandru Costin: “The advent of these cookie changes in browsers means we need to actually make personalization saner and give the user more control in how his or her data is being used. But it also gives enterprises the opportunity to engage directly with their customers if they have a direct connection.
For example, in Experience Cloud we have the Adobe Experience Platform, where many large Fortune 5000 companies are keeping their own customer interaction data. Every enterprise has its own dataset and then we can connect that to lead data. We can connect that data lake with their own content generation mechanisms so they can generate personalized content for their users.
We can do that in a way that privacy is respected because we’re not following you across the internet. I can recommend and generate content just for you because I have some demographic information about who you are.”
“We’ve engaged with many customers across media production, and they all have this content supply chain or content velocity problem. They all want to create on-brand content featuring their products or their style at scale. And there is a spectrum of usability of those outputs right now.
You can use them for ideation, which is a very low bar that can be achieved very fast. You can use them for pre-production, meaning you almost get what you want. You need to take it into Photoshop and change it a little bit to get the final result. Or you can use it for production and that’s the highest bar to achieve.
That’s also the one that we think will require some mentality changes. I don’t think that today any marketer will be comfortable letting a model generate content and put it out on the internet without a manual review process. We think humans should be the ones at every level approving what gets published because of the risk to revenue and the brand’s reputation. “
“There’s still a lot of education that needs to happen for enterprises to understand how we’re different from other models out there. I’ve been in so many meetings, explaining to customers that our models are actually safe to use commercially because there’s a grass-roots adoption happening for some of these other models.
The legal departments in large companies find out that these models are used and they’re worried about bringing in all this risk. The legal department intervenes and says it is blocking generative AI adoption, and they blanket approach where they say ‘Let’s not use any generative AI.’
We feel our models are very different from the other models trained on internet data because we offer that guarantee not only through our AI transparency, but also through something called indemnification, [in] which we give them the offer to pay their legal fees if they get sued for us generating copyright material.”
“The main generative AI demos you see today generate raw content that needs to be edited in our tools. But we’re working on our own models, too. We’re working on both text-to-vector video generation and video editing in our tools.
But we think that when other models launch and customers want to use them, we’ll welcome those videos generated into our tools because they still need editing. We have an open, welcoming attitude towards other models generating content that needs our tools.
The expectations from our professional customers are very high. So we haven’t launched models yet because we don’t feel we have achieved the level of quality that’s needed, but we’re actively working on those.”
Join 500,000+ readers who start their day with Quartz.
By subscribing, you agree to our Terms of Service and Privacy Policy.
