What AI should and should not be doing in power sizing
The interesting use of AI in temporary power is not guessing the answer. It is removing the twenty minutes of questions before the answer and explaining the reasoning afterwards.
Every industry is currently being offered an AI version of itself, and construction is getting more than its share. Some of it is genuinely useful. Some of it is a language model asked to produce a kVA figure, which is a bad idea dressed up as innovation.
It is worth being precise about where the value is, because the distinction is not academic. If a number on a distribution schedule came out of a probabilistic text generator, nobody can defend it in a review, and nobody should try.
Where it helps
- /Intake: turning a description of a site into the specific inputs a calculation needs, and asking only for what is missing
- /Explanation: saying in plain English why diversity was applied here and not there
- /Consistency: applying the same assumptions on Friday afternoon as on Monday morning
- /Support: answering a question about regulations or kit without anyone having to find the right person on the phone
Let the model handle the conversation. Let the engineering handle the engineering.
Where it does not belong
Sizing itself should be deterministic. Given the same inputs, the output should be identical every time, traceable to a rule, and checkable by a competent person. That is not a limitation to be engineered away, it is the point. A calculation that cannot be reproduced cannot be reviewed, and a temporary power design that cannot be reviewed has no business being installed.
The same applies to pricing. Rates come from a catalogue, not from a model's impression of what a 150 kVA set costs per week. If the number on a quotation cannot be traced to a line in a price list, it is not a quotation, it is an estimate wearing a suit.
The test to apply
When anyone offers you AI in this space, ask a simple question: which parts of this are deterministic, and which parts are generated. A good answer is specific and slightly boring. A bad answer talks about the model.
Used that way, AI does not replace the engineer. It removes the admin that stops the engineer from doing the assessment properly on every job, rather than on the ones with enough time in the programme.
Try it on your next site
Sizr applies this thinking automatically on every job, and prints the assumptions alongside the answer.
Keep reading