Payment UpdateWeek of Oct 12
What AI is actually doing behind the counter of a Brooklyn bodega
TL;DR
amNY visited Brooklyn bodegas running an AI price tool. It suggests a price, shows the nearby average and the range down to a ZIP code, and never changes a price by itself. Full disclosure, the distributor amNY quoted is Xtraspots. Here is what the tool does, what the counter still decides, and how the independent restaurant data says to adopt AI.
What happened
On October 7, amNY published a visit to two Brooklyn bodegas, in Sunset Park and Greenpoint, running a point-of-sale system with an AI pricing feature. For a product on the shelf, the screen shows a suggested price, the average price at nearby stores, and the range, and the comparison can be narrowed down to a ZIP code. In one store a Coke Zero was entered at $2.49; the tool suggested $4.95 and showed a neighborhood range of $2.48 to $7.43.
Nothing changes on its own. Charles Sepulveda of Xtraspots, the distributor quoted in the piece, told amNY: "It'll give you recommendations, but it'll never change itself." Full disclosure: that is us, and we have placed about 50 of these systems.
The workers were not uncritical. The Sunset Park worker said his store starts from the delivery receipt and its own markup, then compares that to the benchmark, and that some averages are off. He also said: "I could see people taking advantage of it."
For restaurants, the James Beard Foundation's 2026 report on more than 380 independent operators found the top wanted uses of AI were inventory and supply chain, staffing and scheduling, and reservations. Over 80% plan to increase AI spending next year, and 38% said AI tools currently feel irrelevant to their business. Restaurants with moderate, intentional adoption reported stronger performance than both the low-tech and high-tech extremes.
What it means for you
The useful part is not the robot. It is seeing what the block charges before you print the tag, and doing it in seconds instead of walking three stores. The worker's order of operations is the right one: your invoice, your markup, then the benchmark as a check.
The same lesson shows up in the restaurant data. The operators who did best gave AI a few specific jobs and watched the results. The ones who bought everything, and the ones who bought nothing, did worse.
What to do
Give AI one job and check its work for a month. Pick the task that costs you the most time: reordering, the weekly schedule, or pricing against the block. Ask your POS what it can do there without a data scientist. Keep your own rules, compare the suggestion, and keep the final say. If a suggestion looks wrong, it probably is.
Questions owners ask
Does the tool raise my prices automatically?
No. It recommends. You change the tag.
Where does the benchmark come from?
From other users of the system and public sources, which is why some averages look off. Treat it as a reference, not a rule.
Is this a sales pitch?
Partly, which is why we are saying so. The advice stands with any system: one job, one month, your call.
Sources
