Before Your Company Invests in AI, Ask Whether There’s a Real Customer Need

Editor’s note: This piece is part of our Columnist Network series, which explores the tactical thoughts and actions from Adweek’s community of high-level experts. Today, Shutterstock CEO Stan Pavlovsky shares the strategy behind the company's most recent tech initiative. Below, in his own words, he cautions against investing in AI just because you can. The challenge Ad creative is the single most important attribute in driving performance, which means there’s enormous pressure on digital marketers to select the right imagery for campaigns. At Shutterstock, we realized that our customers were getting overwhelmed by choice when perusing our library of more than 500 million images. They didn’t know where to begin. They would often refer back to previous campaigns for inspiration, but selecting content for campaigns seemed to be driven by gut instinct or guesswork. Internally, we identified this challenge as the “cold start” problem and recognized that AI could help. By analyzing hundreds of millions of images, AI could help creators marry their creative intuition with tangible, data-driven insights into how creative assets are likely to perform for specific objectives like brand awareness or with specific audiences in different markets. At a higher level, our leadership and technology teams were already thinking about how to integrate AI into our tech stack to improve the user experience and make our platform feel like an innate part of the creative process. The process We knew that improving the creative selection process would be a long game, and we started by hiring an experienced team of data scientists and AI experts and establishing a new business unit, Shutterstock.AI, to tackle the challenge. Next, we got out of their way, giving them time to build and train models with the ability to parse creative assets into hundreds of thousands of data signals based on parameters like layout, copy, color, style and sentiment. The goal was to join this asset-level data with a brand’s campaign data or Shutterstock’s anonymized data constellation (from across more than 7,000 brand partners) to determine which creative choices correlate with high click-through rates and other KPIs. Based on this work, we’ve already launched audience resonance scoring to help our customers discover and select the content that is relevant and resonates with their target audiences. And early next year, we will offer a set of AI-powered tools. Our vision is for users to enter their campaign objectives, channels, location, KPIs, goals and audience information to generate impactful content recommendations. Alternatively, users will be able to upload existing creative regardless of source into our catalog tool, and our AI will analyze and rank based on expected performance against chosen goals. This will give them a valuable heads up before using underperforming assets in the real world and wasting ad dollars. The takeaway We strongly believe that creativity starts and ends with humans, and we designed these tools to augment creative intuition—not replace it. Shutterstock creators are still in full control of the look and feel of campaigns, and they are now able to make creative decisions with confidence as data validates their choices, reducing decision fatigue and helping them create more engaging content. AI is most effective when it solves a specific customer problem, so don’t build and train models just because you can to address an edge case. This should go without saying, but there are too many examples of companies investing in technology for the sake of impressing investors or shareholders, even if it only helps a small subset of their users. Start with building something your customers will love, and then remember to be patient. It’s natural to want to see ROI quickly after making a sizable investment in engineering and data science talent, but testing and training models is an ongoing process. It takes time and effort, but when done well, it can not only transform a business but also change an industry.