A restaurant operator describes spending 8–10 hours per month manually exporting and comparing POS/PMIX, sales, labor and distributor purchase reports in Excel because POS systems provide limited dynamic reporting. They experimented with automating these analyses by creating AI prompts to process uploaded CSV data and iterate until the prompts produced the desired year-over-year, day-over-day, and weather-aware comparison reports.
Questions this video answers
8 to 10 hours a month.
Chapters
Anybody who owns a restaurant that's listening to this will absolutely understand what I'm about to say. I probably spend at least 8 to 10 hours a month pulling individual isoteric non dynamic reports off of my pmix, off of my sales, off of my labor matrix, off of my purchase spend from my distributor. And I have to go through so just like a sales report in general, POS companies suck at creating good dynamic reporting because no operator likes to look at stuff the same way. I guess. So it's all very generic but they also don't have. There's a very common piece of technology called bi. It's a business instrument and all it is is just a repository for all your data and every data is comma delimited and it's, it's got a commerce separated value CSV and so it's like sales day of the week, best check average total number of items of each cell. Well, our reporting doesn't get that. I can run a PMIX and I can run it over a period of time, but one of the things that my POS system won't do for me is it won't run a year over year, day over day with a trend line and a percentage up, down and it doesn't factor in weather. My POS system does run record historic weather, but it doesn't take that and give me a layover report like this. Same Monday last year you did $3,500 in sales and it didn't rain. So this Monday it's supposed to rain. So we look at the historic last time it rained on a Monday in this month or whatever. I'm pulling PMIX reports. Yeah, yeah. And then I'm going through exporting them to Excel, I'm resorting them and putting other comparative data up against them. And I spend all this time doing this. And so I decided, you know what, I'm just going to make this, I'm going to make this, this prompt code of a number of different codes that just measures what I'm doing and then I'm just going to upload the data and see if those prompts work to give me the same outcome. And they didn't and had to work on it a few times.
