When the spreadsheet stopped working

The first system I built for my cafe was a Google Sheet with more than ten tabs. Receipts. Cost and profit for every recipe. A marketing todo list. Inventory. A supplier list. Bookkeeping.

By the second month, one tab was still alive. Bookkeeping, and only at the end of the month.

The other nine did not die of laziness. They died because nobody could reach them. I wanted the staff filling the sheet, and the staff did not have a computer. The ones who could get to it did not understand what I wanted in each column, which was my fault rather than theirs.

So the cafe routed around the sheet, the way teams always do. Everyone recorded what happened in our LINE group, and at the end of each month I sat down and copied it out of the chat into the spreadsheet by hand.

That is the part worth sitting with. The sheet was still the system of record. It just had a person wedged in front of it doing the job of the interface, and the person was me.

The workbook is quietly doing two jobs

A spreadsheet that lasts is almost always doing two jobs at once. It is the database: the place the owner trusts when the month closes. And it is the user interface: the thing staff are supposed to open on a phone to log what just happened.

Those two jobs pull in opposite directions.

A good database wants narrow, boring, consistent rows. One fact per row, one meaning per column. A good interface wants the opposite: big obvious actions, fewer fields, forgiveness when someone types “2 boxes” instead of 2.

Our tabs were the compromises. Recipe costs. A marketing list. Inventory. Suppliers. None of those were mistakes. Each was the smallest fix at the time. Staff still could not use the sheet.

What actually broke

That hand-copy held for two or three months. Then it was too slow. I started logging only the price, so I could still get a total. I lost the category. I could not see what was costing too much.

The total is the painkiller. The category is a vitamin. A profitable month is fine. When revenue drops and you have to dig, you cannot.

Bookkeeping at Zendai is 600+ items a month, 300+ receipt images, and inventory in and out every day. That is what I was scrolling.

One month, even the price-only close took three to four hours. That was enough. I started writing the chatbot.

The usual fix does not fit

The standard advice is to buy software. Get a real POS, a real inventory app, a real booking tool. For some businesses that is right.

Local shops are broad and shallow. A cafe, a clinic, a studio, and a retailer all need a system. They almost never need the same one. Off-the-shelf tools are either too rigid or too general. The owner ends up typing into three apps that do not know about each other. That is worse than the sheet.

The other advice is to hire someone to build it. That works. It costs more than the problem.

Separate the two jobs instead

Spreadsheets are not the villain. One artefact should not be both the database and the interface.

Keep the database. Replace the interface.

The database stays a structured sheet in your own Drive, with rows that mean one thing. The interface becomes the group chat the team is already in. Someone sends a receipt photo, a stock-in, a question. The agent records it, keeps the structure clean, and answers in plain language.

Nobody learns a new app. The person who knows the business does not have to become a systems designer after close.

That chatbot became ZendaiBills: bookkeeping only, inside the LINE group. Shops asked for more than books. That is why we built Nossy.

The hour that used to take a week

This year the Revenue Department asked for documents because I applied for a tax refund. I spent one hour exporting the summary from ZendaiBills and sending it. That was the whole job.

One day the barista was off. A manager covered the bar. The barista still had to take calls: less sweet, where is this item, how this regular likes it. The knowledge was in one head. We now teach Nossy once, so the floor shares a brain. That is how we use it at the cafe, not a feature list.

A Tuesday in the group chat is what the day looks like once the interface lives in LINE.