The Starbucks App, Sysco and AI

Probably not what you're thinking off the bat

First, a bit of history is required here. I worked for CAKE (restaurant point of sale and software) and was focused on contemplating a consumer app for dish discovery in addition to the software platform / POS.

Maybe more on that later, as I still think that’s an unsolved problem, but irrelevant for the moment.

Sysco acquired CAKE, and after discussions with the CEO, we pivoted my team to working on internal tech for Sysco to improve the customer experience - order tracking, online ordering, inventory management etc.

Another aside - it’s fascinating to see Sysco being in the Reddit discourse due to their Michelin Guide affiliation as well as their acquisition of Jetro. There are so many levels to fundamentally misunderstanding the realities of how food gets from a farm/manufacturer to a plate, but another post for another day.

What I’m focused on today is something that I couldn’t help but create a correlation to in my brain from my experience while working at Sysco and what’s happening now with Starbucks rolling back their AI powered inventory app because it basically sucked at its core premise.

This isn’t a shot at NomadGo, the company upon whose tech this was built per se. It is a commentary on truly understanding your customer and building the right product, which is where the tie in to Sysco is.

“According to Fast Company, the technical breakdowns stemmed from both software limitations and outdated infrastructure. While NomadGo’s computer vision achieved 99% accuracy in controlled tests, CEO David Greschler noted that computer vision inherently struggles when inventory changes — requiring up to six weeks of retraining for seasonal holiday cups or limited-time packaging that NomadGo developers sometimes only learned about once items hit store shelves. ”¹

While at Sysco, we had a lot of real breakthroughs on the technical side of things that moved the company forward. Moving onboarding online to bring time to onboard a customer down from around two months to about 48 hours, real time tracking of trucks and inventory on trucks (both harder problem than they sound) and modernizing online ordering for restaurants.

The last one is the focus here. In brief, as the tech started to work and that turned into real money for the business, everyone wanted to be touching what we were doing. Fine, but the customer has to be at the center of everything, which got lost along the way somewhere.

A few things are critical for restaurant tech of this sort. First, has to be dead simple. I shadowed many a restaurant manager or chef who rarely used a computer or ordered things online (yes really), and on the other end of the spectrum even the most tech savvy operators just need the software to work fast every time they use it.

This is where things got interesting. When companies build tech for their customers to solve real problems, they will win. Duh. However, when companies get large enough they often start to impose their own will onto customers, sometimes for logical reasons, often due to misaligned incentives.

Sysco decided that now that this whole technology thing was working so well that it was time to really bring in some big brains around it to “take it to the next level” etc. Enter BCG and McKinsey along the way. I’m not here to poo poo big consulting groups en masse, but I will say that the syndrome of trusting external groups to know your customer and produce solutions for those customers can be a slippery slope.

I started sitting in meetings being shown the Starbucks app and being told “just build this but for restaurants!” A fascinating thought, given that quickly placing your pickup order for your venti pumpkin spice latte with eight shots of espresso, seven pumps of pumpkin sauce, and one pump of maple pecan sauce so you can earn points in your app is fundamentally different than creating lists of goods, optimizing for ordering hundreds of line items of product at precision accuracy and doing all this in a few minutes.

Oh, and also ensuring the tech hands off between web and mobile and works offline in your walk-in freezer. Knowing your customer.

But here we were being lectured about how we could sell more house brand product (a fine internal goal) if we just surfaced ads for it in the ordering process, how building a one click “order my favorite” style app would solve so many problems.

After some amount of time experiencing this, it became clear that the momentum we had built solving real customer problems and understanding the user was being less valued than a cool presentation on a topic that served as a dopamine hit for someone. I’m not a luddite. I’m not sitting here saying that there’s nothing to learn from great products nor that Starbucks didn’t create a multi-billion dollar bank effectively with their loyalty program.

What I am saying is that if you do not deeply understand your customer, their business and their needs you end up with a cool toy that scans some shelves and counts things but doesn’t actually work for the users who it was supposedly built to help.

If you’re in a position of solving any type of problem, think about this. Understand why AI use cases would or would not help your user. These are amazingly powerful tools that we now have at our disposal and as the cost of building theoretically comes down, the need for good product judgment goes up just as quickly, if not more so.

1. https://apple.news/AZyen4Zf0QpecmlOjX0cnhA