Editor’s Note: This article is part of Food On Demand’s Summer Series, where the editorial team explores a different topic each month. August’s focus is Tech Lessons, for which one article each week features tech-focused lessons from restaurant industry professionals.
While Science On Call bills itself as an AI-powered support platform for restaurant operators, one of co-founder and CEO Andy Freivogel’s biggest takeaways from years in restaurant tech is surprisingly low-tech: Start with the basics, then troubleshoot from there.
That lesson, Freivogel said during an August 19 interview with Food On Demand, remains as true today for AI integration as it did when he worked with restaurants and retail operators to design Wi-Fi infrastructure two decades ago.
“AI has limitations, just like wireless networking did 20 years ago,” Freivogel said. “In order for Wi-Fi to be successful, you actually have to have a good wired network first. I would say in order for AI to be successful in this space, you need to have that foundation of skilled human beings with domain expertise who understand what output is supposed to look like, and are able to do things like quality assurance on the output of their AI solutions.”
Freivogel said the best implementations of AI services within restaurant tech stacks are those with human domain expertise layers attached to them, and he predicted that would be the case for at least the next three years.
“Twenty years ago, when wireless internet was starting to become a thing, I would talk to people about how to design their infrastructure and how to build out the right physical plant to support their office, restaurant or retail shop,” Freivogel said. “There were people that would just kind of wave their hands at you and say, ‘Oh, isn’t this all going to become wireless anyhow?’ They didn’t realize that Wi-Fi network connectivity was not mature enough for everything to be wireless at the time—and I would actually maintain even today, it’s not quite there.”
Freivogel said he has noticed that same faulty approach more recently from operators looking to AI for solutions.
“People are starting to say things like, ‘Well, can’t we just do this with AI,’ or ‘Can’t we just wait until there’s some kind of AI-powered service to do this?’” Freivogel said. “But, there are principles that really are essential to success there, and one of them is good data. If you have a mixed bag of data drawn from multiple systems without everybody speaking the same language, that’s really confusing for some of the AI-powered solutions that are out there.”
Ultimately, Freivogel said AI-powered solutions require good data from the start and good data hygiene. Similarly, he said robots benefit from human managers, as automation requires quality assurance by people.
“You have to have that foundation in place to really leverage all the benefits of all this incredible software that we have available to us in the restaurant space,” Freivogel said.
Regarding the ideal approach to tech stacks, Freivogel suggested that operators approach the topic less like a list with a bunch of boxes to check or a multiple-choice test and more like a series of essay questions.
“The reason that they want to provide a long-form answer on that test is that it allows for the eccentricities of being a human,” Freivogel said. “It allows it to create a broader margin for error.”
