Ask a venue for their menu and you get a PDF. It looks organised. It is typeset, it is current, someone paid a designer for it. It feels like exactly the kind of thing a computer should be able to use. It is not.
To software, that file is a layout. It knows there is text at these coordinates in this font. It does not know that the line reading "Svíčková 320" is a dish with a price, that the small italic underneath is an allergen list, or that the dish relates to a recipe that relates to a supplier that relates to a prep task on the Tuesday list.
A human reads all of that instantly, because a human brings the context. The PDF never contained it.
What gets lost between the kitchen and the file
The menu is the visible tip. Underneath every line on it sits knowledge that never made it into any document:
- Why the dish is finished that way, and what happens if it is not
- Which substitution is acceptable when the supplier is short, and which one changes the dish
- What to tell a guest who asks whether it can be made without cream
- Who to call when the delivery is wrong at 07:00
- What the dish costs to make, and therefore whether the promotion is a good idea
None of that is in the PDF. All of it is in someone's head. And that is the actual asset: the menu is a marketing artefact, and the knowledge behind it is the operation.
You did not buy a menu from your designer. You bought a picture of the smallest part of what your kitchen knows.
Why scanning it does not fix it
The obvious response is to extract the text. Plenty of tools will do this, and the result is a wall of words with no relationships in it. You can now search for "svíčková" and find the line. You still cannot ask what to do when the cream is out, because that answer never existed in any file.
Text extraction solves the wrong half. The problem was never that the letters were locked up. It was that the meaning was never written down in the first place.
Take any dish on your menu. Ask what a new hire needs to know to serve it correctly on a busy Saturday. Then count how much of that answer exists in a file, and how much exists only in the person you just asked.
Structure is the thing you are missing
Structured knowledge means the dish knows what recipe it comes from, the recipe knows its allergens and its supplier, the supplier knows who to call, and the prep task knows which section owns it. Ask any question and the answer can be assembled, because the relationships exist.
This is not a technology problem. The technology to hold that structure has existed for decades. The problem is that somebody has to sit with the chef and get it out of him, in an order that does not waste his afternoon, and write it down in a form that holds those relationships.
That is the work. It is unfashionable, it is human, and it is the only part of the job that cannot be skipped.
The good news
It has to be done once. A PDF menu is redesigned every season and the knowledge underneath it barely changes: the svíčková is made the way it has been made for years, and the reason it is done that way is stable even when the price is not.
Capture that once and it outlives your menu, your designer, and eventually the chef who taught it. Which is rather the point.