Ask a deputy head what the timetable costs and you will get an answer measured in July: two weeks, sometimes three, usually including a chunk of what was supposed to be annual leave. That number is real, but it is the smallest of the costs and the only one anyone counts.
The larger costs are the ones that do not appear on anyone’s calendar.
The first is the change cost. A manually built timetable is a finished artefact — a grid that encodes hundreds of decisions, most of which exist only in the head of the person who made them. When a teacher resigns in November, that person has to reconstruct the reasoning before they can safely move anything. Which lessons were placed where because of a room constraint? Which were a favour to a part-time colleague? Nothing in the grid records why. So every mid-year change is a partial rebuild, and the changes never stop coming.
The second is the single-point-of-failure cost. In most schools one person can build the timetable. If they are ill in July, or leave in June, the institution has a genuine problem, and the handover is close to impossible because the knowledge is not written down anywhere. Schools manage this risk by hoping.
The third is the quality cost, and it is the least visible. A manual timetable stops when it becomes valid — when the clashes are gone and the grid is full. It almost never continues to "and now let us make it good." So nobody checks whether one teacher ended up with five consecutive periods while a colleague got a comfortable spread, or whether Year 9 has Maths last thing on Friday every week, or whether the same member of staff has every early start. Those decisions were not made. They are just what fell out of the order things happened to be placed in, and they are then locked in for a year.
The fourth is the divergence cost. The master grid gets copied into a staff handout, a parent letter, a room booking sheet, and the cover system. Then one lesson moves, and four of those five artefacts are quietly wrong. Anyone who has watched a class arrive at a room that another class is already in has seen this cost being paid.
Automation changes these costs unevenly, and it is worth being precise about which.
It largely removes the change cost, because the timetable stops being an artefact and becomes a result. The constraints are the thing you maintain; the grid is regenerated from them. A teacher leaving means updating one constraint and regenerating, with the option to lock everything that is working and let the software rearrange only what must move.
It removes the single-point-of-failure cost, because the constraints are written down. Anyone competent can read them, change them, and regenerate. The institutional knowledge lives in the system rather than in one person’s memory.
It substantially improves the quality cost, because a computer will happily evaluate ten thousand valid arrangements and pick a balanced one, which no human has the patience to do. This is where the gap is widest: not in whether a valid timetable can be found, but in how good the one you settle on is.
It removes the divergence cost only if the timetable is connected to everything downstream. A generator that produces a better grid you still export into four places has fixed one problem and left another. This is the thing to check.
What automation does not remove is judgement. Software cannot tell you that a particular teacher is going through a hard year and should not have the difficult Friday afternoon group, or that two members of staff should not be timetabled to share a room. Those decisions stay human. The right division of labour is that the software handles the combinatorics — the part that is mechanical and that humans do badly — and the humans handle the context, which software cannot see. A product that tries to automate the judgement is overreaching; a product that makes you do the combinatorics by hand is underdelivering.
Skoolia’s AI scheduling is built for that division: constraints in, conflict-free and workload-balanced timetable out, locked lessons respected, and the result flowing straight into attendance, cover and the parent portal so there is nothing to re-export and nothing to diverge. The two weeks in July is the cost everyone sees. It is worth asking what the other four are costing.