Joe & The Juice wants one thing out of its Facebook campaigns: the lowest cost per install possible.
The brand is still adjusting its Facebook strategy in the wake of Apple’s iOS 14.5 release, when the ability to track cost per action with any level of confidence “went out the window,” said Miguel Martin, global head of digital marketing at the coffee and juice brand. Since then, it’s become open to anything that can improve its new core metric, cost per install.
Martin allocates 90% of the digital marketing budget toward driving mobile app installs, dedicating most of that customer acquisition spend to Facebook and Instagram, because when new users join the app, they visit stores 2.4 times more often. The company also runs ads on Google and Reddit.
The idea of AI appealed to Joe & the Juice, so it ran an A/B test with Pixis, which promises to use AI to optimize campaigns for efficiency better than Meta can. The test showed that, in comparison to Meta, the cost per install fell 12% with Pixis, while conversion rates jumped 14%. For Joe & The Juice campaigns in Europe, the conversion rate ranged between 40% and 44%, compared to 55% in the US.
Under the hood
To maximize Joe & The Juice’s budget, Pixis’ system moves bids and budgets on Facebook and Google up and down.
“Let’s say the systems know that, on Wednesday at 2 p.m., if we scaled the budget for one of the assets on Facebook, the purchase volume can be higher at the same cost. It will go ahead and [increase the Facebook ad budget],” said Pixis Co-Founder and Chief Business Officer Hari Valiyath.
Martin noticed that even on weekends and holidays, the tool automatically shifted dollars from one channel to another to optimize spend. “You’re away, but the AI is working for you. It’s nice to see,” he said.
Performance AI (PAI) pulls from historical campaign spending data to come up with recommendations for efficient spend across ad networks, including Meta, Google, TikTok, Bing, LinkedIn, DV360 and Google Display Network (with plans in the works to go live on The Trade Desk and Amazon DSP). Other signals the tool relies upon include seasonality-based trends, attribution, analytics and live performance data. It also measures how it’s allocating its marketing dollars against AppsFlyer’s attribution tool.
PAI uses natural language processing to generate a number of relevant clusters of users based on their behavioral patterns, preferences and engagement style. Then, with the help of these clustering models, a targeting engine called Targeting AI finds new, more profitable audience cohorts on Facebook and Google. A brand like Joe & The Juice can feed its ideal customer profile into the targeting tool so the tool can look for “premium,” more high-intent audiences, Valiyath said.
In addition to the performance and targeting assists, Pixis’ deep learning and AI optimization tools surfaced useful insights for Joe & The Juice.
For instance, Facebook ads for the brand’s app used to offer a free sandwich to new customers after their first app purchase by default. But thanks to Pixis, they learned what motivated different regions’ consumers. Free sandwiches incentivized US users to install the app. In Europe, free juice drove installs. And in the Nordics, free shakes carried the day.
B2C ecommerce businesses like Joe & The Juice can see incremental lift within seven weeks of deployment when using Pixis, according to Valiyath.
Pixis streamlined tedious work and improved KPIs at a time when Martin had to decide between growing his small team or finding a tech solution to automate some of the manual work. Future tests with the Pixis tech are on hold until the team has the bandwidth to try them, but it’s a matter of when, not if.
“AI has arrived to stay,” Martin said. “It’s going to change a lot – the industry and our jobs. I’m intrigued to see where we’re going to be in two years.”