How Stop & Shop is using AI, not cookies, to target customers

Could AI be the key to replacing cookies?

For over two decades, brands like grocery retailer Stop & Shop have depended on third-party “cookies” – the small text files saved through a web browser as the user browses a website – to discern customer behavior and power their programmatic, or software-driven, digital advertising.

But those days are nearly over – thanks to Google’s plans to phase out support for third-party cookies on Chrome in 2023, Apple’s limits on the use of its mobile device ID in iOS apps, as well as the need for GDPR compliance. Soon, the $152 billion U.S. digital advertising industry will no longer have access to most third-party cookie data.

Retailers and brands are not waiting around for the ax to fall, however. Instead, many are already testing a variety of possible ways to target the right customers once access to third-party cookies ends. These include AI-powered solutions that learn from first and third-party data to spot patterns that form customer profiles or segments.

Stop & Shop, for example, a Massachusetts-based supermarket chain (and subsidiary of holding company Ahold) with 415 stores in the Northeastern U.S., has spent the past year and a half testing new approaches with digital marketing firm AMP Agency.

“We wanted to figure out what would change in a post third-party cookie world,” said Samantha Weiss, VP, data strategy and programmatic at AMP Agency. “How can we adapt benchmarks and grow and enhance?”

Stop & Shop needed a new way to target customers

According to Meghan Galligan, digital marketing director of Stop & Shop, the company knew it needed to find other ways to target customers with relevant content in a privacy-friendly way, as the deprecation of cookies approaches.

“We wanted to be early hand-raisers and test as much as we could this year while we still had the safety of cookies to fall back on,” she said.

Most importantly, Stop & Shop has made sure its first-party data strategy is sound, she explained. Now, they are also benchmarking new solutions against cookie-based segments, as well as against each other.

Her team has also been preparing Stop & Shop’s leadership for the post-cookie landscape. “For the last two years, we’ve told them that everything that they’re used to seeing from us is about to shift,” she said. “We’re letting them know that when 2023 comes we will have a solid plan that we feel good about.”

One approach AMP Agency has taken on behalf of Stop & Shop is increasing their use of contextual targeting, or the practice of displaying ads based on a website’s content. Another tactic is partnering with companies offering new customer identifiers that use first- and third-party data to replace cookies. For example, AMP Agency is testing LiveRamp’s ID, which uses machine learning algorithms to perform identity resolution by combining a company’s data with its own massive database of over 250 million U.S. consumers.

Finally, the agency had already worked with Dstillery, a customer audiences solution, on AI-driven custom targeting. This year, they tested Dstillery’s recently-launched programmatic advertising solution, ID-free Custom AI, which uses machine learning and predictive analysis to help companies reach their best audiences, but does not track users at all.

An AI solution with no user tracking

Melinda Han Williams, chief data scientist at Dstillery, told VentureBeat that its approach is predictive behavioral targeting, but without tracking the users that are going to be targeted.

Instead of trying to use AI to understand the user, she explained, the solution uses AI to find the best impressions that are most likely to lead to conversion.

Dstillery trains the AI model on the client’s first-party data, and then looks for “privacy-safe signals” in order to make programmatic advertising bids. That may include looking at the DMA (designated market area), the URL and time of day in order to serve the ad to the best customer.

The approach, Williams claims, picks up on purchase intent behaviors such as research and planning, as well as lifestyle behaviors.

“I think that’s already pretty different from most of the approaches out there,” she said, adding that most other post-cookie solutions are developing new identity-based solutions. But what about all the web traffic that has no ID?

“We already know that some people don’t want to be tracked on the internet. And all of those new identifiers require some amount of explicit opt-in,” she said. “So there’s going to be a lot of people out there that are unreachable with those solutions.”

Dstillery’s ID-free solution “has some flavors of what you expect from those ID-based approaches, because we’re using a lot of information to make predictions about where’s the best place to show the ad,” she said. “But I think we’ve been trained to think the goal is to reach users and how much information do I have about the user and how privacy-friendly is the information about the user – but ID-free Custom AI is not trying to reach users at all.”

AI neural network focuses on behavioral trends

Dstillery, she says, goes in a different direction. “We’re really taking at face value that this user doesn’t want us to know anything about them, so we’re not going to try and figure out something about the user,” she explained. “The only thing we know about the user is what they’re doing at this very moment, because that’s the inventory that we have the option to show an ad at.”

Instead, Dstillery tries to figure out how much it can infer just based on that one piece of information – what is the user doing right now? “Based on that, what’s their likelihood of being a good customer for this brand in the future or converting for this brand in the future,” she said.

Then, the solution looks at the digital journeys of those who have fully opted in for research purposes, where they allow companies to see every single site they’re going to. “We’re looking at millions of these journeys, and this is where the AI comes in,” she said. “We run it through a neural network – we came up with this big model that we call the ‘map of the internet’ that tells information about every single site and how it’s related to every other site.”

Once this foundational model was built, Dstillery could understand that when they see a user at one website, they can understand a great deal about a customer without identifying them in any way.

“We can build a whole custom model specific to an advertiser – for every single website, we have a score that places a value in this moment in whatever broad-based geography that the person is located in, to determine their likelihood for converting for that brand,” she said. “It’s a behavioral prediction based on inventory, on a specific impression, rather than based on a profile of the person.”

Testing AI solutions for a post-cookie world

Williams points out that even Google’s own efforts to provide a cookie replacement, as part of its Privacy Sandbox initiative, has focused on customer identity.

“Its first approach was FLoC, which groups users together based on their behaviors, but that wasn’t privacy-sensitive enough,” she explained. “Now they are doing Topics, which provides more broad-bucket information, but it’s not as useful and it’s based on the same framing a lot of the industry has – that to be effective the advertiser needs to know more about the user, but the user wants to share less information. These two ideas are inherently competing against each other.”

But Dstillery claims its results are on par with cookie-based approaches. “It’s impressive that even without all of that information, you can use AI to do pretty much the same level of prediction, just from the impression,” she said. “You have to shift your mindset.”

And, she adds, the Chrome team has encouraged this type of development from other companies. “This is the type of internet they would like to be creating with the Privacy Sandbox,” she said.

Right now, people are looking to be educated, so Williams says Dstillery is doing a lot of outreach. She also recommends that people test now while cookies are still around to compare measurements. “The most common way we have people testing is head-to-head against other approaches, and then measuring it against their favorite cookie-based KPIs.”

For Stop & Shop, this testing of AI-powered solutions has been essential. “It’s given us the ability to build learning agendas and prepare ourselves for next year,” said Galligan. “It’s something that’s going to provide us a lot of strength as we go into this new post-cookie future.”