Use a different tool type for each job: AI takeoff software (Kreo Pro at $175 per user a month) for quantities, spreadsheet AI such as Copilot in Excel for cost checks, and a business-plan chat assistant for reports. Contract-review tools help with payment checks. Test each on a finished project, and record your due diligence, as RICS's AI standard expects.
Takeoff accuracy is the figure to distrust. Vendors measure it on their own test plans, and a tool that is right on walls and floors can still be 10% out on skirting, so your own test on a finished job is the only number that counts. RICS's standard on responsible use of AI has applied since 9 March 2026 to members and regulated firms, and it expects due diligence on each tool, so the results of your tests belong in the file.
Scoring five tool types against six jobs
A quantity surveyor does not buy "AI". They buy help with specific jobs. The matrix scores each tool type from 0 to 3: 0 is no use, 1 helps with a part of the job, 2 does most of it but needs structured review, and 3 does it well with ordinary checks. The scores are my judgement from what the vendors' own pages say each type does and from the failure modes below. They are not test results, so re-score them on your own work.
| Job | AI takeoff | Spreadsheet AI | Chat assistant | Contract and payment review | Phone and site capture |
|---|---|---|---|---|---|
| Takeoff from drawings | 2 | 0 | 1 | 0 | 0 |
| Comparing tender returns | 0 | 3 | 2 | 0 | 0 |
| Rate and bill-of-quantities checks | 0 | 3 | 2 | 0 | 0 |
| Interim payment checks | 0 | 2 | 1 | 2 | 0 |
| Cost reports and client letters | 0 | 1 | 3 | 1 | 0 |
| Measuring an existing building | 0 | 0 | 0 | 0 | 2 |
No type scores 3 on takeoff, and that is deliberate: the worked test below shows why. Here is what each type costs, from the vendors' pages in late October 2026.
| Tool type | Examples and list prices (USD) | Billing and trial |
|---|---|---|
| AI takeoff | Kreo Pro $175 per user a month (Essential at $100 has no AI); Togal.AI Growth $299 per user a month; Bluebeam Revu $260 to $590 per user a year, with AI drawing review and comparison on Max | Kreo: annual billing only, 7-day trial. Togal.AI: billed yearly, demo first. Bluebeam: annual |
| Spreadsheet AI | Copilot Business add-on $21 per user a month on annual billing ($25.20 monthly; $18 promotional to 31 December 2026); Claude Pro $20 | Monthly or annual |
| Chat assistant | ChatGPT Plus about $20; Claude Pro $20; business plans $20 to $25 per seat with a two-seat minimum | Monthly or annual |
| Contract and payment review | Mastt free to start, Professional $165 a month (its average cost per project); Lexilio Starter $29, Professional $299, Enterprise $1,999 a month | Mastt month to month. Lexilio: 7-day trial, card required, plan starts automatically after it |
| Phone and site capture | Polycam Basic $150 a year, Business $400 per user a year; Apple's built-in Measure app | Polycam monthly or yearly |
Type 1: AI takeoff from drawings
This is the type most quantity surveyors mean by AI. Kreo's pricing page splits it cleanly: the Essential plan at $100 per user a month is takeoff and estimating without AI, and the Pro plan at $175 adds one-click area and line measurements, automatic symbol counting, AI takeoff for rooms, walls, doors and windows, AI scale detection and drawing naming, and batch takeoff across sheets. Kreo bills annually only and offers a 7-day trial. Togal.AI's Growth plan is $299 per user a month billed yearly with unlimited automated takeoffs, and you book a demo first. Bluebeam Revu is the markup and measurement tool many practices already own, at $260 to $590 per user a year; its Max plan lists AI drawing reviews and drawing comparisons, which suit checking what changed between drawing revisions, but the page does not mention AI takeoff.
Vendors' accuracy claims are the weak point. One pricing page I read calls its product the most accurate on the market without describing a method, another publishes no accuracy figure, and published figures elsewhere are measured on the vendor's own test plans. So run your own. Here is an invented test on a finished two-storey rear extension, where the final-account quantities are the benchmark. Tolerances are set in advance: areas within 3%, lengths within 5%, counts exact.
| Element | Unit | Final account | AI takeoff | Variance | Within tolerance? |
|---|---|---|---|---|---|
| External wall, net of openings | m2 | 186.4 | 181.9 | -2.4% | Yes |
| Ground-floor slab | m2 | 64.0 | 64.8 | +1.3% | Yes |
| Roof area on the slope | m2 | 78.5 | 77.9 | -0.8% | Yes |
| Floor finishes | m2 | 118.0 | 119.2 | +1.0% | Yes |
| Internal doors | nr | 14 | 14 | 0 | Yes |
| Internal partitions | m | 74.2 | 70.1 | -5.5% | No |
| Windows | nr | 11 | 12 | +9.1% | No |
| Skirting | m | 132 | 118 | -10.6% | No |
Five lines of eight pass, and the pattern is the useful part. Areas and counts came out close. The failures are the derived measures: partitions and skirting depend on room-by-room perimeters, door deductions and alcoves, and the window count included a rooflight. The practical rule is to trust a tool per element type, not overall. Areas and door counts might go straight into the estimate after a spot check; lengths get measured by hand until a second and a third test say otherwise.
The time sum, with its assumptions in view: a manual takeoff of this job took 6 hours. The AI-assisted version took 1 hour to run and 2 hours to check and correct the three failed lines, 3 hours in all. The saving is 3 hours, worth $210 at $70 an hour. At one such job a month, Kreo Pro at $175 nets $35. At two a month it nets $245. Break-even is 175 / 210, about 0.83 jobs a month. Togal.AI at $299 needs about 1.4. The same test also catches a realistic mistake: a takeoff run on the superseded drawing revision, because nobody told the tool which sheet was current. Check the revision on every sheet before you run it, and see a small firm's test of AI takeoff from drawings for a fuller method.
Watch for three other limits. Takeoff tools measure geometry, so they do not apply your measurement rules for deductions or item descriptions. Finishes, preliminaries and anything not drawn will not appear. And upload terms matter, because drawings are the client's confidential material.
Type 2: Spreadsheet AI for tender and rate checks
This is where most practices get value fastest, because the data already lives in a sheet. Microsoft's FAQ lists a Copilot Chat-eligible business plan among the subscriptions that work with Copilot in Excel, as basic access, with the paid Copilot licence adding priority access. Two settings trip people up: desktop Excel has to meet Microsoft's app-version requirements, so check the update channel if the button is missing, and Copilot's editing works only when Calculation Options are set to Automatic, a setting often switched to manual on heavy cost plans. The old =COPILOT() worksheet function was retired on 14 September 2026, so the side pane is the route. Claude for Excel is the alternative; its documentation says inputs and outputs are deleted on Anthropic's side within 30 days and carries a prompt-injection warning, which means you should not open a spreadsheet from an unknown source with it. Google Sheets has an equivalent AI function on eligible plans.
The job to test is a tender comparison. Put the returns in columns C to E and use this prompt.
The table in A1:E14 has a line number in column A, an item in column B and
three tender prices in columns C, D and E. For each line:
1. Flag any tender more than 25% above or below the median of the three.
2. List every cell that is blank or contains text such as "incl." or "excl.".
3. Do not change any figure. Put your findings on a new sheet called Checks
and use formulas where possible so I can audit them.An invented set of returns and the sort of findings that come back:
| Line | Tender A | Tender B | Tender C | Finding |
|---|---|---|---|---|
| 5 Brickwork, 186 m2 | $27,900 | $29,300 | $38,400 | C is 31% above the median of $29,300 |
| 8 Roof covering, 78 m2 | $11,200 | $10,900 | $11,500 | No flag |
| 11 Windows, 12 nr | $14,400 | incl. | $15,100 | B contains text, not a price |
What to fix: the 31% on line 5 is real (38,400 / 29,300 = 1.31), and the next step is a clarification to Tender C about specification, not a conclusion that it is wrong. The finding on line 11 is correct, but assistants sometimes read "incl." as zero and report a bargain, so check that the sheet's totals did not treat it as $0. A median of three tenders is crude, so use it to find lines to read, not to rank bidders. And ask for the formula, for example =IF(ABS(E5/MEDIAN(C5:E5)-1)>0.25,"check",""), in a visible column. A model that does the sums silently can be wrong silently; a formula can be audited. Comparing subcontractor quotes with AI works the same job end to end.
Type 3: Chat assistants for cost reports and tender letters
A chat assistant is the right tool for words and the wrong tool for measuring. Its jobs are cost report commentary, clarification letters to tenderers, summaries of qualifications and exclusions, and meeting notes. Mind the file limits: Claude takes PDFs up to 1,000 pages, but for a PDF of 101 pages or more it reads the text only, so every chart, sketch and drawing in a long tender pack goes unseen; split the pack so the drawings sit in a file under 100 pages. ChatGPT accepts files up to 512 MB or about 2 million tokens. Use a business plan for client material. RICS's FAQ says insurers increasingly want business versions of AI tools with stronger data protections, and the business plans from ChatGPT and Claude both cost $20 to $25 per seat with a two-seat minimum.
You are helping a quantity surveyor write the cost commentary for a monthly
client report. Use only the figures in the table below. Do not calculate new
totals; if a figure is missing, write "not provided". Write about 120 words:
the current forecast final cost, the change since last month, the two largest
movements and their causes, and one risk we are monitoring.
Table: [paste]An illustrative result: "The forecast final cost is $1,284,000, up $18,000 on last month. The two largest movements are variation 7, additional drainage (+$14,500), and an electrical provisional sum adjustment (-$9,000). Contingency remains at $42,000. We are monitoring ground conditions at the rear of the site." Check it line by line. The figures match the table, which is the point of the "use only" rule. But "ground conditions" appears nowhere in the table, so the assistant invented a risk. Strike it and write your own. Three habits keep this type safe: paste figures in, never ask it to recall them; ban new totals; and read every number against the source before it goes out under your name.
Type 4: Contract and payment review tools
These tools read contracts and applications and flag issues. Mastt's pricing page shows a free plan with unlimited users and tracking of up to 10 payments across a project, and a Professional plan at $165 a month, given as the average cost per project, that adds an AI Assistant and "AI Contract & Payment Review", month to month by card. Lexilio's plans run from Starter at $29 a month through Professional at $299 to Enterprise at $1,999, with cross-document conflict detection on Enterprise only. Its 7-day trial needs a card and rolls into the paid plan unless you cancel, so diary the end date. Lexilio's site would not load when I checked it, so confirm seat and contract limits, and which standard contract forms it reads, on its pricing page before you trial it.
The vendor pages do not publish accuracy figures, so test with planted errors. Take a closed interim application, copy it, and seed three mistakes: an overclaim on one line, a variation claimed without an instruction on file, and retention at 3% where the contract says 5%. Then see how many the tool finds. Output from a tool like this might read as follows, and the right-hand column is what you still do.
| Tool flags (illustrative) | What the surveyor still does |
|---|---|
| Line 14 claimed at 90%, previous valuation 70% | Visit or check progress evidence before agreeing a figure |
| Variation 6 claimed with no matching instruction | Check the instruction register and the contract's valuation rules |
| Retention deducted at 3%, contract says 5% | Confirm against the contract data, then correct the certificate |
The tool's flags are leads. Deciding what to certify is professional judgement, and contract interpretation stays with you or the contract administrator. See how to test whether AI can check an interim valuation for the longer version.
Type 5: Phone and site capture for existing buildings
Phone scanning helps when a job starts from an existing building: a refurbishment budget, a variation on a house with no current drawings. Polycam's pricing page shows a Free plan, Basic at $30 a month or $150 a year, and Business at $400 per user a year, which is the lowest tier whose description lists 2D and 3D floor plans and advanced measure tools. Space scanning works on compatible iOS devices, and the page makes no accuracy claim. Apple describes its built-in Measure app as giving approximate measurements, so treat it as a sense check.
Test it with five dimensions: measure each with a tape or laser measure, scan the same room, and record the difference. An invented result: wall lengths of 4.20 m, 3.85 m and 5.10 m by tape come back at 4.17 m, 3.88 m and 5.02 m, so errors of 3 cm, 3 cm and 8 cm. That is fine for a budget estimate of a room's floor area, and not fine for a final account where an 8 cm error on a long wall multiplies through every finish. Set the tolerance by use before you scan, not after.
A test every type should pass, with errors planted
The same protocol works for all five types, and it takes less than a day.
- Pick a closed project, so you know the right answer.
- Write the tolerances first: areas 3%, lengths 5%, counts exact, money to the dollar.
- Run the tool exactly as you would on live work, with the same inputs and settings.
- Compare line by line, and mark each line pass or fail.
- Plant two or three known errors in a copy and see whether the tool finds them.
- Record the pass rate by element or job type, the time taken, and the checks it needed.
- Decide per job type: use after a spot check, use only with full checking, or do not use.
Repeat after major software updates and every six months. A takeoff tool that passed on walls last spring may behave differently after a model change, and nobody will tell you it has.
The RICS standard: what to document for each tool
RICS's page on responsible use of AI in surveying practice describes the September 2025 edition of the standard, published on 17 November 2025 and in effect since 9 March 2026, for members and RICS-regulated firms. Its FAQ says it applies only to AI use with a material impact on the delivery of surveying services, and that deciding what counts as material will often be a matter of judgement. The page lists what the standard covers: knowledge requirements for using AI; practice management, including data governance, system governance and risk management; AI procurement and due diligence; output reliability and assurance; client communication and transparency; and guidance for firms building their own AI. For outputs with a material impact, the FAQ says a written decision on reliability must be prepared by, or under the supervision of, an appropriately qualified and named surveyor, and that regulated firms stay accountable for every output even when each one is not scrutinised. On clients, it says the standard sets reasonable requirements for what goes into terms of engagement, aiming for transparency without lengthy disclosures, and it notes that insurers increasingly want business versions of AI tools. Read the standard itself for the exact wording; the pages I read do not set out a tool register or a separate written client notice.
The practical answer is one page per tool. Here is one, filled in for a takeoff tool.
| Field | Entry |
|---|---|
| Tool and plan | Takeoff software, Pro plan, one user, annual billing |
| Job it is used for | Quantities for small refurbishment tenders |
| Material impact? | Yes: quantities feed the estimate sent to the client. Decision and reasoning recorded 3 November |
| What we upload | Drawings only; no client names or personal data |
| Provider information obtained | Data retention, use of our files for training, accuracy method; requested in writing 3 November, reply filed 10 November |
| Accuracy test | Completed project, 8 lines, 5 within tolerance; partitions, windows and skirting measured by hand |
| Output sign-off | Named surveyor writes "quantities checked" on every takeoff |
| Re-test | After each major update and every six months |
And a short paragraph for the terms of engagement:
We may use AI tools to assist with measurement, cost checking and report drafting. A named surveyor checks and takes responsibility for every output, and our advice rests on professional judgement, not on the tool. On request we will explain how an output was produced and checked.
The record does not need to be long to be useful: the point is that someone can see what you checked and when. A small firm's walk-through of the RICS AI standard goes further on keeping these records.
What a three-surveyor practice pays, and the break-even
A practice of three surveyors might run two Kreo Pro seats ($350), three Copilot Business seats ($63), Mastt's free Starter plan ($0) and Lexilio Starter ($29): $442 a month, or $5,304 a year. At $70 an hour of surveyor time, break-even is 5,304 / 70, about 76 hours a year, or 6.3 hours a month across the practice. About two hours per person per month covers it. That will not happen if the tools sit unused, which is why the test protocol and the one-page record matter more than the shopping list. Run the sum again with your own seats and rates, and keep the tools that clear it.
The bill-of-quantities side of this is covered in a bill of quantities template checked with AI. If you are weighing a takeoff tool against another type, buy the one whose job costs you the most hours now, and leave the others until the first has passed its test.
Questions quantity surveyors ask about AI tools
Can ChatGPT or Claude take off quantities from a PDF drawing?
They can read schedules, notes and dimensions printed on a drawing and answer questions about them, but measuring geometry to scale is the job of takeoff software, and chat assistants are weaker at arithmetic. Treat any quantity a chat assistant reports from a scaled drawing as unverified until you have measured it yourself or had a takeoff tool produce it.
Do I have to tell clients that AI was used?
RICS's FAQ says the standard sets reasonable requirements for what your terms of engagement say about AI, aiming for transparency without lengthy disclosures. The page I read sets out no separate written-notice rule, but it says regulated firms remain accountable for every output. Read the standard for the exact wording, then put a short paragraph in your terms of engagement, like the example in the body, before a client asks.
Which type should a one-person practice buy first?
Start with what you may already have: a business-plan chat assistant and Copilot or Claude in Excel for cost checks, about $20 to $25 a month. Add takeoff software only after a trial on a finished project shows a saving larger than its price. At $175 a month and $70 an hour, a tool has to save about two and a half hours a month.
Does the RICS standard apply if I am not a member?
It is aimed at RICS members and RICS-regulated firms. If you belong to another professional body, check whether it has published equivalent guidance, and ask your insurer what it expects. Keeping the one-page record for each tool is good practice whatever your membership, and it takes an afternoon to set up.
Further reads
- Responsible Use of AI in Surveying Practice: A Small Firm's RICS Checklist — A small firm's checklist for the RICS responsible-AI standard.
- AI Take Off Software: Trial Three Tools on One Real Plan Set — Trial three takeoff tools on one real plan set.
- How Much Does Estimating Software Cost? Price Bands With and Without AI — Estimating software price bands with and without AI.
- AI Tools for Construction Estimating: Bid Checks After the Takeoff — Takeoff tools make quantities; checking the bid is a different job. A comparison with list prices and a five-gate bid check on a worked example.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Kreo, Togal.AI, Bluebeam, Mastt, Lexilio and Polycam pricing pages; Microsoft 365 Copilot Business page and Microsoft's Copilot in Excel FAQ; Anthropic's help article on uploading documents to Claude; RICS page and FAQ on the standard for responsible use of AI in surveying practice.