Importing a schedule from a photo — what works and what does not

The single biggest reason people abandon a scheduling system is data entry. A month of shifts is twelve or thirteen events, each with a date, a start and an end, copied from a grid designed for a noticeboard. People do it once, sometimes twice, and then stop.

Photo import removes that step: you photograph the schedule, and the shifts are read off the picture.

How it works

The picture is processed to find the structure — rows, columns, dates, times, names — and turned into events. In Freeish that happens on the server rather than on your phone, using a vision model, and the image is discarded rather than stored.

What you get back is a proposal, not a saved schedule. You see what it found, correct anything wrong, and then save. That distinction matters, and it is worth insisting on in any tool that does this: silently saving a guess is much worse than showing you one.

What reads well

• Printed rota grids — the standard noticeboard month or fortnight. The best case by a distance.
• Screenshots of scheduling apps. Clean, high-contrast, consistently laid out.
• Spreadsheet grids, screenshotted. Structurally the same thing.
• Whiteboards, if written reasonably clearly and photographed straight on.

What reads badly

• Photographs at an angle. Perspective distortion breaks the row-column structure that the whole thing depends on. Stand square to it.
• Bad lighting. Glare across a laminated rota, or a photo taken in a dim corridor.
• Handwriting, particularly cursive or crossings-out.
• Rotas with no year on them, which is most of them — a grid saying “Mon 3” does not say which March.
• Ambiguous times. “7-7” could be twelve hours or a typo. It has to guess.
• Multiple people’s schedules in one grid, where yours is one row among twenty. It may take the wrong row.

Practical technique

Square on, whole grid in frame. Include the headers — the dates and the column labels are what make it readable.

Two photos beat one bad one. A fortnight each, straight on, reads better than a whole month at an angle.

Take it in decent light. Under the corridor light rather than in a dark break room.

Check the year and the edges. The first and last rows are where errors cluster, because they are the ones the crop cuts.

Always check before saving

Worth saying plainly, because it is the step people skip once they trust it: look at what it read.

The errors are rarely dramatic. They are a 7 read as a 1, a row offset by one, a shift assigned to the wrong day at a month boundary. Individually small, and a wrong date is exactly the kind of error that causes somebody to turn up at the wrong time.

Ten seconds of checking, against a minute of typing that you did not have to do, is a good trade.

What it does not do

It does not connect to your employer’s system, and in Freeish that is deliberate. There is no integration with any scheduling platform, which is also why nothing you do is visible to your employer. The photo is the interface precisely because it requires no relationship with them.

It also does not know what your shifts mean. If your rota uses codes — E, L, N, LD — you may need to tell it once what those are, or correct the first import and let subsequent ones follow the pattern.

When typing is still better

If you work a regular rotation — 4-on-4-off, Pitman, a fixed pattern — do not photograph anything. Set the rotation once and let it generate forward. That is less work than importing a picture every month, and it stays right without you doing anything.

Photo import is for schedules that are assigned rather than patterned: the self-scheduled ward rota, the retail week posted on Thursday, the fortnight on the staffroom wall. Those cannot be generated, which is exactly why they are the ones that get retyped.