AI estimating software: what it actually does and does not do

Every estimating product now has AI somewhere on the homepage. Some of it is real capability, some of it is a chatbot bolted to a spreadsheet. This article explains what the technology genuinely does inside a serious estimating tool, where the hard limits are, and how to tell the difference before you pay for anything.

The three jobs AI does inside an estimating tool

Under the hood, AI in estimating software does three distinct jobs. Knowing them separately makes the whole category easier to judge.

1. Reading. The AI takes unstructured input, a typed job description, a voice note, a PDF of architects' drawings, and turns it into structured facts: rooms, dimensions, wall build-ups, openings, specifications. This is the genuinely new capability. Five years ago software could not do this at a useful standard; now it mostly can, with important caveats covered below.

2. Interpreting. From those facts, the software works out what work is implied. A 4m by 5m single storey rear extension implies foundations, drainage alterations, cavity walls, a roof, insulation, first and second fix, decoration and more, even though nobody typed any of that. This is where a good tool encodes real quantity surveying knowledge rather than guesswork.

3. Pricing. Quantities get multiplied by rates to build a bill of quantities. Here is the part most people get wrong about the category: in a well built tool, this stage should not be AI at all. Pricing should be deterministic arithmetic, the same input always producing the same total. AI belongs at the reading and interpreting stages, not inside the multiplication.

What that means in practice

Take a concrete example of the whole pipeline working. You type: single storey rear extension, 4m deep by 5m wide, flat roof with two rooflights, bifold doors, knock through to existing kitchen, relocate one radiator. In under a minute a good tool returns a structured estimate: excavation and foundations measured from the footprint, cavity walling with your named blockwork rate, a steel and padstones for the knock-through, the roof build-up, first and second fix electrics and plumbing, plastering, decoration, and preliminaries. Each line carries a quantity, a rate and a subtotal you can interrogate.

In QS Quoter specifically, that output arrives as two documents: a clean client copy, and a private tradesman copy showing your cost build-up and margin. The rates are your own day rates and material assumptions, sitting on a rate book calibrated for 2026 UK pricing, so the tool starts sensible and gets more like you the more you adjust it.

What AI estimating software cannot do

Now the other half of the ledger, which matters more:

• It cannot see site conditions. No software knows the drains run the wrong way, the neighbour is difficult, or the ground is made up of rubble from a demolished outbuilding. Anything invisible on paper is still your risk to price. That is what site surveys are for.
• It cannot guarantee it read the drawing right. Drawing extraction is good but not perfect. Scale ambiguity, cluttered plans and hand annotations cause misreads. This is precisely why serious tools show you every extracted number for confirmation instead of hiding the workings.
• It cannot know your appetite. Whether you want this job, how busy you are, what this client is worth long term: these change the right price and no algorithm holds them.
• It cannot negotiate, walk the site, or stand behind the number. The signature on the quote is still yours.

Does and does not, side by side

How to judge a tool before you pay

Five questions separate substance from sticker:

1. Transparency. Does it show what it extracted from your input, before pricing? If you cannot see the workings you cannot check them.
2. Editability. Can you change any line, rate or quantity? A quote you cannot edit is a quote you cannot own.
3. Determinism. Run the same job twice. If the totals differ, the pricing itself is a language model guessing, and that is not an estimating engine.
4. Whose rates. Your day rates and your material costs, or an opaque national average? Averages lose money in both directions.
5. Cost of trying. A confident vendor lets you test on a real job before committing. QS Quoter's first quote is free for exactly this reason.

Run the test that settles it: take a job you have already priced by hand and won, feed it in, and compare line by line. Where the tool differs from you, one of you is wrong, and finding out which is instructive in both directions. For what accuracy you should realistically expect from that test, read how accurate is AI quoting, honestly.