The problem, as you feel it
The numbers exist. They are in your point-of-sale, your accounting software, your scheduling tool, your payroll, and a few spreadsheets. Looking at them together means exporting, pasting, fixing the dates, and building a chart, which is why it happens once a quarter, in a hurry, before a meeting.
So decisions get made from feel. Which location is actually the most profitable after labor. Which product line is quietly shrinking. Whether the new hire paid for herself. Whether Tuesdays are worth staying open. You have opinions on all of these. You do not have the numbers in front of you, and by the time you do, the quarter is over.
Hiring an analyst is a salary. A dashboard project is a six-month commitment to a tool you then have to maintain. What you actually want is simpler: the report you keep meaning to build, arriving on its own, and the ability to ask a question and get an answer without opening Excel.
What this looks like now
Three pieces, in order of how much they change your week:
- The report that builds itself. Every Monday at 6am, a page arrives with the dozen numbers you actually run the business on, pulled straight from the source systems, compared to last week, last month, and the same week last year, with the changes that matter called out in a sentence each. Not forty charts. Twelve numbers and the three things that moved.
- Questions in plain English. You type “which location had the highest labor as a percent of sales last month, and what was it the month before” and get the answer, the table behind it, and a chart, in a few seconds. Underneath, an AI is writing the query against your actual data and showing its work, so you can trust the number.
- Early warnings. Rules watching the data for the things you would want to know the day they happen: a location’s average ticket falling for two weeks in a row, a product’s margin dropping under a threshold, a customer who used to order weekly going quiet.
The data stays in your systems or in a private copy you own. Nothing is uploaded to a vendor’s dashboard. The AI reads, computes, and writes; it never changes a record.
Take a business like this one
The business in this section is a composite of the kind of company I talk to, not a named client. The numbers are the shape of the problem, not a case study.
Picture a restaurant group with four locations, one owner, one operations manager, and a bookkeeper who comes in Thursdays. Sales are in one point-of-sale system, labor in a scheduling app, purchasing in a mix of supplier portals and email, and the P&L in QuickBooks, usually three weeks behind.
The owner’s actual questions, when I ask, are: which location makes money after labor, whether the lunch service at the newest location is worth it, why food cost at one store is always two points higher, and whether the catering side is growing or just loud.
What I set up for a business shaped like this:
- A private data store that pulls nightly from the point-of-sale, the scheduling app, and QuickBooks, and cleans the three different ways they each name a location.
- The Monday page. Sales, covers, average ticket, labor percent, and food cost percent by location, with week-over-week and year-over-year, plus a short written summary of what changed. Delivered as a web page that reads well on a phone.
- A question box. The operations manager types questions in ordinary English. Behind it, an AI writes the queries and shows the table it used.
- Four alerts. Labor percent over target for a location for three days running, food cost up more than a point week over week, a lunch service under its break-even covers for two weeks, and any day where a location’s sales are more than a quarter below the same weekday last year.
- A monthly deeper cut on the questions the owner named, updated automatically, so the “is catering growing” conversation is a page, not a debate.
In a setup like this, the answers to the owner’s four questions arrive in the first two weeks, and then keep arriving. What changes is not the data; it is that decisions start being made on Monday morning from numbers that were true on Sunday night. In a business shaped like this it is normal to find one location whose lunch service has been losing money for a year, or a food-cost gap that turns out to be one supplier’s price creep, and either of those pays for the project many times over.
What it costs to run
The data store and the nightly pulls run on a small server for a few dollars a month. The AI answering questions costs cents per question. There is no per-seat dashboard license. The setup is the work: connecting your systems, reconciling how they each describe the same thing, and building the report around the questions you actually ask.
What I need from you
The five to twelve numbers you would want on one page every Monday. Read access to your systems, connected together on a screen share. An hour with whoever knows why the spreadsheets are the way they are. Then two weeks of looking at the report and telling me what is missing or wrong, because the first version is never quite right and the second one usually is.