Coffee robots: stability matters more than smarts

I spent two days ordering from that SIASUN coffee robot in the campus lobby, start to finish. Over two days I bought seven cups myself — two in the morning, three in the afternoon, and the remaining two were me asking the person behind me to order other categories. Not an institutional test method, just squatting next to it watching.

Conclusion first. This thing is worth using, but only worth buying for specific people.

The interface is a vertical screen, categories on the left — americano, latte, cappuccino — and on the right you can adjust sweetness and ice level. First cup took me about twenty seconds because ice level is buried in a second-level menu, you have to tap into "Customize" to see it. There's an A4 sheet taped to the side of the machine with handwritten operating steps. Clearly I'm not the only one who got stuck.

Order, scan, pay, the arm starts moving. Grab cup, drop ice, grind beans, extract — not fast. I timed it: from payment to cup out, between a minute thirty and two minutes. When there's no queue that speed is fine. At the 9:30am rush, four people ahead of me, the guy behind me took two looks and left.

What surprised me is consistency. Three lattes in a row, the foam thickness, temperature, sweetness barely differed. At another place with an arm I'd been to before, the first cup and second cup tasted like two different baristas made them. This one, no.

What really caught me off guard was ops. That afternoon the guy came to swap the milk, and I chatted with him a bit.

He said what he fears most isn't the machine making a mistake, it's nobody knowing how to handle a stuck cup. One stuck cup and everyone in line scatters.

Milk swap plus cleaning, ten-plus minutes total, machine down. So it's not "unmanned," it's "fewer-manned." That one point determines its cost structure, and where it can open.

The downsides, clearly. The milk line needs daily cleaning, the grounds bin needs emptying — neither can be skipped. The ordering flow isn't friendly for first-timers, I didn't see voice guidance on screen, not sure if this version just didn't enable it. Cup-out speed means it can't handle instant peaks, only suits locations with smooth foot traffic.

From a business angle, per-cup pricing is about the same as a street chain, material cost is easy to calculate, what's hard is equipment depreciation, space, ops, and waste. Amortize those and if you don't hit a certain cup count per day you're losing money. Domestic parts have pushed whole-machine cost down these past two years, but that's hardware — it doesn't push down people.

So what it sells isn't coffee, it's turning a stall that needs people into one that doesn't need many. The ceiling is in location density, not in the robot. Anyone can source an arm; the recipe and the ops network are the moat. I've looked at quite a few in this space — the hardware gap isn't as big as the marketing claims.

Who it suits: campuses, office buildings, hospitals — locations with steady foot traffic where you can arrange one person to check on it daily. Not suited for small teams wanting to build brand premium off one machine, nor for fully unattended scenarios.

Actionable advice. If you're going to deploy, negotiate monthly rental or per-cup revenue share first, don't buy outright — run a month and look at real cup volume and failure count before deciding. Those two data points are more solid than any spec sheet.