How Can AI Help Quantity Surveyors? BoQ Checks, Takeoffs and Cost Reports

AI helps a quantity surveyor most with checking and drafting: it can flag odd BoQ descriptions and arithmetic slips, propose a first takeoff for you to verify, and turn your figures into a cost report draft. It does not take professional responsibility. Scale, measurement rules, rates and final totals stay with you.

The useful question is not "what can it do?" but "what happens when it is wrong?" A wrong adjective in a report draft costs you a minute. A wrong wall area carried into a tender costs real money. So sort the work by consequence first, try AI where a miss is cheap and visible, and keep a short list of checks manual however good the tool looks in a demo.

Follow me on Instagram@sagnikteaches

Sort the job by what a mistake costs

Vendors sell the speed. Sorting by consequence is what keeps your name safe. Here is the split I would use before touching any tool.

Connect on LinkedInSagnik Bhattacharya

Notice that the AI column is never "decides". That is the line to hold throughout.

Subscribe on YouTube@sagnikteaches

A main example: checking a 40-line BoQ extract

Say a small contractor is pricing a single-storey rear extension and has received a bill of quantities from a designer with 40 lines in the groundworks, masonry and roofing sections. You want a second pair of eyes before pricing. The first job is a wording and consistency check, run on a copy with names removed.

You are checking a bill of quantities extract for clarity, not pricing it.
For each line, tell me if the description is vague, duplicates another line,
contradicts another line, or is missing a unit or specification.
Do not invent quantities or rates. Return a table with columns:
line number, issue type, what is unclear, a question I should ask the designer.
If a line looks fine, leave it out.

[paste lines here]

An illustrative result on a 40-line extract might be five flagged lines:

• Line 6, "Excavate for foundations": no depth and no disposal instruction. Question: is spoil removed from site or reused?
• Line 11, "Concrete to foundations": no strength class, and no note on whether it is poured or pumped.
• Line 19, "Blockwork to cavity": the unit is m but the item reads as a wall area.
• Line 27 and line 33, both "Insulation to walls": possibly the same item entered twice.
• Line 38, "Allow for flashings": no length and no material.

That output is not an answer, it is a list of questions. Each one still needs you to open the drawing and decide. But it takes a minute instead of an hour to get a prompt list, and it catches the duplicate pair that a tired reader skims past. Two of the five might be false alarms, which is acceptable; a check that finds three real queries in a minute earns its place.

Add the arithmetic check, but do not trust the chatbot to multiply

Language models are unreliable at long runs of multiplication and addition. They are good at reading a table and telling you which row looks wrong, and they are good at writing the spreadsheet formula that checks the rows. Ask for the formula, then let the spreadsheet do the sums.

I have a BoQ in a spreadsheet: column B is quantity, column C is unit rate,
column D is the stated amount. Write a formula for column E that returns
the difference between B*C and D rounded to 2 decimal places, and a
conditional formatting rule that highlights any row where the absolute
difference is more than 1. Explain each step in one line.

Say it flags rows 14 and 31. In row 14 the quantity is 85 m2, the rate is $42.00 and the stated amount is $3,750. The calculation gives $3,570, so the amount is out by $180: a transposition. In row 31 the amount has been typed over as a lump sum with no rate. Both are things a formula finds in seconds and a human eye misses at the end of a long day. If you are newer to this, cleaning messy data covers getting a pasted table into a sensible shape first.

Then check the section totals and the grand total with a plain SUM on a separate cell. Never accept a total from a chatbot's text reply. The bill of quantities template tutorial walks a whole sample through the same checks, line by line.

Takeoffs: what the software proposes and what you must confirm

Takeoff tools now read plan sets and propose measures: areas for floors and roofs, wall lengths, counts of doors and fittings. Vendors such as Togal, Bluebeam, Kreo and iBeam sell versions of this, with different levels of automation and a range of prices (Togal Growth is listed at $299 a user a month, billed yearly; Kreo Pro at $175 a user a month, billed yearly, with AI takeoff and symbol counting that its cheaper Essential plan lacks; and Bluebeam's Max tier, the only one with its AI drawing checks, at an introductory $590 a user a year, all as of November 2026). The accuracy figures on their pages are the vendors' own claims on their own tests. Treat them as marketing until a tool has done a project of yours and you have checked it.

There is a reason the claims need checking on each job, not once. A takeoff is only as good as the inputs:

• Scale and revision. A drawing issued at A1 and printed at A3 looks fine until the scale bar is wrong. A superseded revision can look identical.
• Measurement rules. Your contract or standard method of measurement decides what is included: openings over a certain size deducted, skirtings measured as a run, and so on. A tool measures geometry. It does not know which rule applies unless you tell it and then check.
• Hidden work. Foundations below ground, services in ducts and anything shown only in a specification do not come off a floor plan.
• Drawing quality. Scans, overlapping layers and hand-drawn changes are where automated recognition drifts most.

A spot-check routine for any AI takeoff

Do not check everything. Check what matters. This takes twenty minutes on a typical house-sized job and tells you whether to trust the rest.

1. Rank the lines by value. Sort the takeoff output by extended cost, not by quantity. A small count of expensive items beats a big count of cheap ones.
2. Re-measure the top five by hand. Use a scale rule or the measuring tool in your PDF viewer on the original drawing.
3. Record the gap as a percentage. Say the tool gave 148 m2 of roof area and you get 152. That is a 2.6 percent gap, and you also know the direction.
4. Apply your tolerance. A sensible tolerance depends on the trade; you decide it before you look. If any top item breaks it, recheck the whole trade.
5. Check the count items with a different method. For doors, sockets or windows, count them in the schedule as well as on the plan. Disagreement points to a missed room or a duplicated symbol.
6. Write down what you checked. One line in the job file: items checked, gap found, date, initials. That is your audit trail if the price is questioned.

For the full picture of tool types and where each fits, see five AI tool types compared by job. Tool names and prices are covered more fully in AI tools for quantity surveying; the focus here is the checking around them.

A mistake that looks like a success

Here is an illustrative failure that is easy to miss: a tool measures the external walls of a two-storey house as one long perimeter and applies the height of the ground floor to both storeys, because the first-floor plan sits on a separate sheet that was not selected. The wall area comes out about 40 percent low, and the output looks clean and confident. A spot check on the biggest line catches it in two minutes: you measure one elevation by hand and the number is nowhere near. Without the check, the tender goes out too low. The lesson is that the error is silent. The software does not flag what it was never given.

Cost reports: let AI write the commentary, not the conclusions

A monthly cost report has two parts: figures and commentary. The figures come from your cost plan, the contractor's valuation and your own forecasts. The commentary explains movement: why a package is over, what changed, what is at risk. Drafting commentary is the safest high-value use of a chatbot, as long as you give it the numbers and forbid it from adding any.

Draft the commentary section of a monthly cost report for a client who is not
a cost professional. Use only the figures below. Do not add numbers, causes
or forecasts that I have not given you. If something is unexplained, write
"cause to be confirmed" rather than guessing. Keep each package to two
sentences, plain English.

Package, budget, forecast, change since last report, reason I know:
Groundworks, 48,000, 52,500, +4,500, rock found at 1.2 m depth
Superstructure, 120,000, 120,000, 0, none
Roofing, 36,000, 34,800, -1,200, supplier discount agreed
Preliminaries, 22,000, 24,200, +2,200, two extra weeks on programme (cause to be confirmed)

An illustrative draft back might read: "Groundworks are forecast at $52,500 against a budget of $48,000, an increase of $4,500 following the discovery of rock at 1.2 m depth. Superstructure remains on budget. Roofing is forecast $1,200 below budget after an agreed supplier discount. Preliminaries are forecast up by $2,200 because of a two-week extension to the programme; the cause is to be confirmed."

What you fix: check every figure against your sheet (they should match, but the model might round or reorder), check the sign of every movement, and read the phrase "following the discovery of" as a statement of fact you are putting your name to. Add the total by hand. Here the forecast is $231,500 against a budget of $226,000, which is $5,500 over, a figure the AI was not asked to calculate and should not be asked to. If you want the draft to carry a total, put the total in the input.

The paragraph you write yourself

Leave the risk and recommendation paragraph blank in the prompt and write it yourself. This is where the client pays for your judgement: which risks are likely, what contingency remains, what decisions are needed this month. A model with no sight of the site visits, the conversations or the contractor's history can only produce plausible filler, and a reader may take that filler as your assessment.

Rewriting a weak BoQ description: before and after

When you write or amend a bill yourself, vague descriptions are where disputes start. A chatbot is a decent editor here because it has no attachment to the draft. Give it the line, the specification extract and the unit, and ask for a tighter description that adds nothing you have not supplied.

Illustrative before: "Brickwork to external walls, as drawings." After the pass and your edit: "Facing brickwork in half-brick thickness to external walls of rear extension, laid in stretcher bond with 10 mm weather-struck joints, measured net of openings over 0.5 m2, m2." Three changes came from the specification, one from your measurement rule, and the unit is now stated. Check the new wording against the specification yourself: if the model has made the bond or the joint finish sound more precise than the documents are, you have introduced a contradiction instead of removing one.

A quick sum shows why this matters. A wall area of 62 m2 at $95 a square metre is $5,890. If a vague line lets one party read it as including the damp-proof course and the other does not, a $600 argument starts before any bricks arrive. Ten minutes of tightening is cheap by comparison.

Five more places AI saves time in a surveyor's week

Comparing two revisions of a specification

Paste the old and new clauses of a spec section and ask for a list of changes only. Then check the list against a document compare in your PDF or Word tool, because a model can skip a line. The value is in the plain-English summary of what changed and which items might affect price.

Turning site-visit notes into a measured summary

Dictate or type rough notes ("east gable brickwork about 60 percent, scaffold lift three struck, skip full") and ask for a tidy progress note with no added claims. You then compare the stated percentages with your valuation.

Building a query list from a tender return

When a subcontractor's return arrives with qualifications and exclusions, ask the model to list every qualification, exclusion and assumption as a table with a column for "affects scope or price". Levelling the returns afterwards is still your job, and the choice of winner is yours.

Checking a contractor's valuation application

This is the one where the stakes rise: AI can find arithmetic mismatches, items claimed above 100 percent and omitted retentions, but the assessment is yours. A fuller test sits in Can AI Check an Interim Valuation?

Drafting standard letters and notices

Payment notices and similar letters have deadlines that are set by contract. AI can tidy wording from your template, but it cannot know your contract dates. See drafting notices and checking deadlines and always run dates against the contract yourself.

The checks that stay manual, listed

If you keep one list by the desk, make it this one.

• Drawing revision and scale on every sheet used for measure.
• Rule of measurement applied for each item, including what is deducted and what is included.
• Totals and subtotals calculated in a spreadsheet, not read from a chat reply.
• Rates and allowances, including any adjustments for the site, access and phasing.
• Provisional sums and risk allowances, which are judgements about uncertainty, not outputs.
• Anything contractual: notice dates, entitlement, extension of time, certification.
• The sign-off: the report carries your name, so you read every line.

If a vendor says its tool removes the need for any of these, ask for a reference from a user in your trade and test it on a job you have already priced by hand.

Running a one-job trial to see if a tool earns its fee

Pick a finished job where you have a hand takeoff and a final account. Run the AI takeoff on the same drawings, with no peeking at your numbers. Compare the top ten lines by value, then the total by trade. Log time taken, including the time spent correcting. Say the tool takes 35 minutes plus 25 minutes of checking, against four hours by hand: you save about three hours on that job. If the checking takes longer than the measuring would have, the tool is not helping on your type of drawing. Repeat on a second, different job before you commit to a yearly plan, since one good result on a tidy set tells you little.

Keep the sums honest: include setup, learning time and the fee. A seat at $299 a month (about $3,600 a year, committed up front) that saves three hours a week pays back quickly for a surveyor billing by the hour; the same seat used twice a month does not.

Data handling and professional responsibility

Tender documents often carry confidentiality terms. Before you paste a BoQ into a chatbot, read the terms of your plan, switch off training on your data if the plan allows, remove client names and anything restricted, and tell your client if your appointment terms require it. If you are unsure, ask your professional indemnity insurer or professional body how it views use of AI tools; this is an area where guidance is still moving, and I would not guess. For a broader look at which jobs deserve a tool at all, the AI hub collects the related tutorials.

Professional duty does not change because a tool produced the first draft. If a measure is wrong, the person who issued the number answers for it. That is a good reason to treat AI output as a junior's work: useful, fast, in need of review.

Questions quantity surveyors ask before trusting AI with measures

Can AI replace a quantity surveyor's takeoff?

Not safely. Takeoff software can propose areas, lengths and counts from a drawing, and you can speed up a first pass, but a person must confirm scale, drawing revision, inclusions and measurement rules. The vendor accuracy figures you see are claims on selected drawings, not guarantees on your set.

Is it safe to paste a BoQ into a general chatbot?

Only after you check the confidentiality terms of the plan you use and remove client names, rates you consider commercially sensitive and anything your contract restricts. A business or team plan with data controls is the usual minimum. If you cannot say where the text goes, do not paste it.

What should I never let AI do on a cost report?

Never let it decide a forecast final cost, a risk allowance, a contingency figure or a recommendation to pay or reject a claim. It can draft wording around numbers you have checked. The professional judgement, and the signature, stay with you.

How do I spot check an AI takeoff quickly?

Pick three to five items by value, not by ease. Re-measure each by hand against the drawing, compare the totals, and note the percentage gap. If any high-value item is out by more than your tolerance, recheck the whole trade rather than fixing one line.

Further reads

• Use of AI in Quantity Surveying: Comparing Subcontractor Quotes — Levelling subcontractor quotes with AI, line by line.
• Construction Final Account: Variations, Deductions and Sign-Off With AI — Variations, deductions and sign-off on a final account.
• AI Tools for Construction Estimating: Bid Checks After the Takeoff — The bid checks that come after a takeoff.
• AI Software for Construction Drawings: Catch Changes Between Revisions — Catching changes between drawing revisions before you measure.
• ChatGPT for Construction Estimating: Check a Quote, Do Not Write It — Using ChatGPT to check a quote without letting it write one.
• Construction Cash Flow Forecast: Build One in a Spreadsheet With AI — A project cash flow forecast built in a spreadsheet.
• Dilapidations Schedule: Draft Items From Lease Clauses With AI — A worked example of mapping lease clauses to dilapidations schedule items with AI, with the RICS format explained and cost and legal checks left to people.
• Can AI Do a Construction Takeoff From Drawings? A Small Firm's Test — A repeatable test for AI takeoffs on one set of plans you already know: what it reads well, where counts and measures slip, and when it is safe to price from.
• HubSpot AI for Heavy Civil Contractors: Does a CRM Fit Bid Work? — HubSpot can track bids and owner follow-up for a heavy civil firm, but its AI helps less than the pipeline. Where it fits and where it stops.
• Masonry Takeoff With AI: Counting Bricks and Blocks From an Elevation — A worked brick and block count from an elevation, with openings, wastage and a quick hand check on the AI's number.
• AI Property Development Feasibility: A First-Pass Appraisal — A worked first-pass appraisal for a small developer: inputs, GDV, costs, sensitivity and planning risks, with AI as challenger and a spreadsheet as calculator.
• RICS AI Standard for Small Firms: Register, Assessment, Client Notice — Turn the effective RICS standard into a written pre-use assessment, two practical registers and client wording that can be checked.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: vendor product and pricing pages for Togal, Bluebeam, Kreo and iBeam (accuracy figures are vendor claims), checked November 2026.