At the bleeding edge of robotic restaurant operations, Blendid is rethinking the smoothie bar. It’s out in the wild, ahead of the tests and prototypes across the robotic restaurant space, and there are some interesting learnings from the real world.

Blendid is essentially a robot arm surrounded by a bunch of blenders and chutes of fruit, veggies and other mix-ins. The arm grabs a cup, collects the goodies and blends it all up, delivering pre-made or custom smoothies. As with many robotic-arm concepts, there’s a hearty dose of novelty. Co-founder and CEO Vipin Jain said that does wear off and then it’s all about the product.

“The novelty is going to wear off. After the second or third try, you just want your food,” said Jain. “This is something we have been very strongly opinionated about from the get-go. We should not build a product for novelty. Robotics is a means to an end, not the core business. It helps people come in, but the product has to resonate from both the expectation and the cost for them to reengage. We see strong repurchase behaviors at all our locations.”

He said roughly 66 percent of orders at the University of San Francisco come from highly engaged customers who have worked a Blendid smoothie into their day-to-day walk to class or work. In a mall location, there’s more of the novelty, but mall staff has really gravitated toward the machines at all hours of the day.

“That was one so overwhelming for me. After we launched and this janitorial staff, a person that works most of the nights. He came and said thank you for being here, and he was actually the first customer before we had even launched. We left the machine on overnight and he just came and ordered and got his drink. So that was very satisfying to see that happen,” said Jain. “Serving the mall in the wee hours, we didn’t imagine that as the use case.”

For anyone who has worked late nights—QSR hours and not much else—something like a healthy smoothie at work is a dream come true.

In both instances, as Jain said, it’s not the robot but the food. Blendid has a culinary team to develop smoothies, but also some help from AI. For every order flowing through the three existing locations, there is a ton of data. Today, is the most popular item and it was developed not by a chef but a computer. To the self-proclaimed “trekkie,” it’s akin to a Star Trek smoothie replicator, but you don’t even have to know what you want.

“It’s a recipe that a machine learning system came up with. It was not initiated by a chef, this was built on data. The system said we could create this concoction with this and some of that,” said Jain. “You can do that with AI, you can build it with millions and millions of variations and create a highly personalized product.”

As the company continues to expand—currently it’s working on the Atlanta market—Jain said wrangling all that data, developing new menu items and watching how customers utilize the machines is a chief goal. He envisions an experience based on each customer’s health and diet needs, paired with their tastes and individual health biome. That, he said, is where AI can truly shine to get customers precisely what their bodies need, without tasting like the inside of a lawnmower.

On the business end, he envisions partners or franchisees building entirely automated food courts. Blendid already has a partnership with Jamba, two of the three robots fall under the Jamba branding. Jain expects more partnerships like that in the future. The company is also integrated with third-party delivery, drivers just walk up grab the prepared smoothie and go, something he’s seen grow even from inside mall-based outlets.

The company just wrapped up its second crowdfunding round, raising $1.4 million on the investment platform Start Engine. It raised $1.9 million in a prior round. The funding will go toward more scale and investments in the team. So far, the company has raised just over $20 million.