Reviewing an invoice or a quote with AI
A refurbishment quote, the energy bill that’s gone up with no explanation, the one from the garage with six line items you don’t understand. They all have something in common: they’re written by whoever is charging, not by whoever is paying.
There needn’t be any bad faith. It’s enough that they’re written from inside the trade for you to be unable to tell the necessary from the optional.
What an AI does well here
Translating the line items. “Builder’s work in connection”, “standing charge”, “diagnostic labour”. Knowing what each line is comes to half the job, and it’s what nobody explains to you.
Spotting what’s duplicated. Items appearing twice under different names, labour billed inside one item and again separately, materials counted twice. It’s a common error and not always deliberate.
Comparing two quotes that look nothing alike. The most useful case of all: two firms quote for the same work with different structures and there’s no way to compare them by eye. Ask for them in a single table, line by line, flagging what appears in only one.
Finding what’s missing. “What does a quote of this type usually include that isn’t here?” This question is the one that avoids “that wasn’t included” three weeks later, which is where nearly all overruns come from.
The questions that surface what’s padding
- “Explain each line as if I knew nothing about the trade.”
- “Which items are essential and which are optional or could be scaled back?”
- “Is anything billed twice under different names?”
- “What is NOT included here that usually ends up charged separately?”
- “What should I ask this supplier before signing?”
- “Is VAT included, and is the right rate applied to each item?”
The fourth is the golden one. A cheap quote missing three items isn’t cheap: it’s incomplete.
Where it stops
It doesn’t know prices in your area. It can tell you whether a quote’s structure is odd, but not whether £40 per square metre is expensive in your city this year. If it gives you a price range, treat it as a vague reference, not a valuation — that’s exactly the kind of figure these systems invent with poise.
It hasn’t seen the job. Two different quotes can both be right because one includes work the other hasn’t spotted is needed.
It doesn’t replace a third quote. The most reliable way to know whether something is expensive is still to ask for another.
Why more than one opinion helps
If you give the same quote to several AIs, something revealing happens: they agree on the odd line items and disagree on whether the price is high. That’s telling you something precise — what they detected by reasoning (the structure) is reliable; what’s market opinion (the amount) isn’t.
That’s what The Judge does with any document: several AIs read it separately and a fixed judge tells you where they agree, where they don’t, and with how much confidence. On a quote, that distinction between “this is wrong” and “this looks expensive to me” is exactly what you need to decide whether to push back or get another quote.
Before uploading anything
A quote carries the firm’s name, sometimes yours and your address. If they aren’t needed for the question, strip them out: it reads the same. We go into it in documents you shouldn’t upload to an AI.
What you take away
An AI isn’t going to tell you whether you’re being overcharged. It’s going to tell you exactly what you’re paying for, and that’s what lets you ask.
With the answers in hand, the conversation with whoever signs it changes tone.