How Yummly is monetizing AI-powered personalization
Brian Witlin, CEO of Yummly, calls the service the ultimate decision engine for food by using personalization, powered by AI, machine learning, and a lot of data science, to answer the question “what’s for dinner?” before it’s even asked.
“That’s part of the magic of personalization,” Witlin says. “Answering people’s questions before they need to ask them.”
One of the things that Yummly tries to convey to users right out the gate is that the more you share with the site, the better your experience is, since that information is leveraged specifically to help users, and is not collected with any ulterior agenda. And users are sophisticated enough today to fully understand that value proposition and be happy to share.
“People are willing to fill out a fairly lengthy onboarding process because they know that at the end of it, they’ll get personalized results,” he explains.
The company has spent quite a lot of time designing their onboarding process to eliminate that taking-a-survey feeling, he says, focusing on user testing and iteration to ensure the experience is fun, playful, and engaging. In fact, it’s now used as a reference both at Google and Apple as an example of a best-in-show onboarding experience, which helps create a personal connection with their users.
“Getting these inputs, conveying to people that they’re going to get a great experience, the better the information they share, and giving them instant gratification at the end of that, creates this virtuous cycle,” Witlin says. “Now we can ask other types of questions because we’ve already built that level of trust with our users, and they see the benefit in the usage of the product. The first time you use Yummly, it feels personalized to you, and it gets better and better as you use it.”
That’s because AI is front and center in the product, he explains.
As soon as a new user goes through onboarding, they’ll land on a page of recipe suggestions that are personalized to them. If they’re a vegetarian and they don’t eat peanuts and are allergic to milk, they’ll be presented only with things that meet those preferences. When they search, they’ll get subsequent optional questions to help them narrow their search, and those questions are constantly changing, being tuned by an AI system to essentially ask the best questions for any query.
It’s training itself and getting better and better with every interaction with people, Witlin explains, because it can see if someone skips a question or if they go back, which might indicate that a question may not be as useful as another question, so a new one will get swapped in, and the system gets smarter.
That’s also true for services like Pandora, Netflix, and Spotify, where the more you use them, the better they get — it’s a symbiotic relationship that is fairly clear to users. But some companies don’t necessarily have that benefit right out of the gate — for instance, a retail clothing site.
“You just have to go the extra mile,” he says. “I think it could start with just surveying existing users and seeing if there are patterns in things they like.”
First, explicit input: What are the types of products you like? What are you interested in? What’s your style? But then you may find that there are purchase patterns in your typical shoppers. And you can segment these into different archetypes, which might lead to questions that may not seem directly relevant, but end up predicting what people may want to see, depending on the season, depending on how trendy or classic they are, or how formal versus informal.
The end goal might be to present to customers a curated set of products that you think they’ll like in a feed, or delivered by email, in order to help customers find things as quickly as possible, and potentially even find pieces that go really well together, which can be based on what other people have purchased collectively.
“By conveying that in some elegant and simple way to the end user, someone might say, ‘I’m willing to tell a store about my style,’” Witlin explains. “I’ll tell them whether I’m big into camping, or if I have to dress up for work, what my job is. Because that’s all going to inform this greater good of helping me, the customer, get to what I want. Even sometimes helping me be inspired without even needing to search for that particular item.”
At Yummly, the goal is to go beyond what most recipe sites offer, which is generally a search field where a user can type “chicken,” because they need a recipe, and they get back a huge list, but still don’t know what they want to make. In Yummly, once you type in chicken, it might say, do you want sweet, salty, or spicy chicken? How much time do you have?
“We ask questions to narrow down,” he explains. “That’s something that can also be leveraged by pretty much any service interacting with customers to continuously improve their experience.”
This kind of carefully constructed personalization has also boosted their monetization efforts, Witlin says. The focus has gone from advertising to developing a monetization model that aligns with the user experience, working with brands it in a way that’s purposeful and useful to the end consumer.
That includes marketing emails in which Yummly recommends kitchen gadgets and tools, which, he notes, seems very disparate from finding a recipe. Initially, when they tested it, it was unclear whether they’d actually be looked at as experts in this arena, or if their recommendations would be trusted. Spoiler alert: they are.
“In fact, because we’ve built that rapport with people, and we’re showing them something in a simple, curated way, people will use our emails and purchase a lot of the items that we showcase,” Witlin says. “We’ve seen our influence has massive impact for brands who are selling on Amazon. They have a huge spike in traffic. Sometimes we’ve seen things completely sell out due to what we reference in our emails.”
Though Yummly doesn’t work with the freemium model, Witlin says it does work well for personalized services, particularly when you’re asking your users for consistent feedback to continue to increase the accuracy of their results.
“They’ve already invested their time, and if they see it working, then they’ll be able to see the unique value that your company brings to the table,” he says. “That’s a really great time to say, hey, by the way, if you pay this monthly subscription, you’ll get these added benefits, or the elimination of advertising, or you’ll have the ability to do this extra level of personalization, or you can add other people to the service. You can find ways to upsell them to other services.”