AI Take Off Software: Trial Three Tools on One Real Plan Set

Pick one finished job whose quantities you counted by hand and trust, run that same plan set through each tool's trial, and score three things: counts against your manual numbers, whether the tool found the scale correctly, and which items it missed. Choose on those scores first and the price second.

Prices and trial terms below come from the vendors' own pages, read on 29 October 2026, for Kreo, Togal.AI and BuildVision AI. Those three publish prices and give you a route to run your own drawings. The method works for any tool, including ones that only offer a demo. The plan set, scores and timings in the examples are made up to show the method, which is why the tools in them are called A, B and C.

Follow me on Instagram@sagnikteaches

Choose a plan set you already know the answers to

The test only works if you know the right answer, so the plan set has to come from a finished job. Pick a job that was built, where the final account, the as-built drawings or your own site counts confirm the quantities. Look for fifteen to thirty sheets that include floor plans, at least one enlarged plan, elevations, a door and window schedule, and a revision. Include one messy sheet, such as a scanned page, a crooked photocopy or a drawing with hand notes. You are going to bid on messy sets, so test on one.

Connect on LinkedInSagnik Bhattacharya

Avoid three traps. A job you lost has no ground truth. A bid estimate is not a key, because it carries your own counting errors. And a vendor's sample plans are not a test, a point the sample-plan section further down comes back to. Your trade shapes the choice too: a drywall or flooring contractor cares most about areas and lengths, an electrician about symbol counts, and a general builder about all of them.

Subscribe on YouTube@sagnikteaches

The vendors' pages make claims you can test. Togal's says up to 98% accuracy on floor plans. Kreo's says Auto Measure delivers a full sheet take-off in minutes that the user only checks. BuildVision's promises quantities in minutes and shows a "98% conf." figure, which is the software's confidence score, not accuracy measured against known counts. None of the pages I read states how accuracy was measured, which is the gap your trial fills. For a broader look at what the technology can and cannot do on drawings, can AI do a construction take-off from drawings covers the limits.

Build the answer key before you open any tool

Write the key first and keep it out of sight while you test, so the tools cannot nudge your numbers. List the quantities you would normally take off, the units, a tolerance and a weight for how much each one matters. The tolerances are mine: 2% for areas, 5% for lengths, exact or within one or two items for small counts and 5% for large ones. Tighten them for work with thin margins.

Here is a filled-in key for a single-storey light-commercial building of 3,200 square feet over 22 sheets. Use metres if that is your unit; only the percentages matter.

The weights add to 100 and reflect two things: the money at stake and how easily a miss slips through review. Partitions carry 25 because they are the biggest dollar line. Sockets carry 15 although each is cheap, because dozens of small symbols are the easiest thing to miscount and the hardest to spot when checking. Fixtures carry 5 because there are only nine and a reviewer sees them at a glance. Writing "two double doors" into the key matters: a tool that counts each leaf will say 26, and you want to know whether it did that or simply miscounted. Spend an hour or two on the key. It is the one part of the trial the vendors cannot help with.

Turn the key into a scoring sheet with four formulas

Put the key in a spreadsheet before any tool runs, so scoring on the day means typing numbers, not doing arithmetic. Use one row per item and these columns: A item, B key quantity, C tolerance type ("pct" or "count"), D tolerance (0.02 for 2%, 1 for "within 1", 0 for exact), E weight and F unit price. Tool A's quantity goes in G, and row 2 then needs four formulas.

Pass?             H2  =IF(C2="pct", ABS(G2-B2)/B2<=D2, ABS(G2-B2)<=D2)
Weighted points   I2  =IF(H2, E2, 0)
Money left out    J2  =IF(G2<B2, (B2-G2)*F2, 0)
Money added       K2  =IF(G2>B2, (G2-B2)*F2, 0)

Totals: =SUM(I2:I10) is the weighted pass rate;
=SUM(J2:J10) and =SUM(K2:K10) are the dollars left out and added.

Copy the block of columns for tools B and C. Then write the pass mark into the sheet before you see a single result, so a polished demo cannot move it. A workable mark for fixed-price work is a weighted pass rate of 70 or more and money left out below 1% of the tested value. For the key above, the tested value is $73,613 (each quantity times its unit price), so the 1% bar is about $736. Raise the pass rate if your margins are thin; never lower it after the results are in.

What each trial gives you and what carrying on costs

The three vendors start you off differently, and the timing is part of the test. Kreo gives seven days, BuildVision fourteen with a card on file, and Togal a booked demo. Start the clocks only when the key is finished.

Three details change how you run the test. Kreo's Essential plan has no AI, and its pricing table puts Auto Measure and AI scale detection on Pro and above, so trial Pro or you will be testing a manual tool. Pro includes Auto Measure with no per-run billing inside the app, so you can run a sheet repeatedly. BuildVision meters work in tokens, at 500 per page and 20 per detection, and its page says model-assisted steps such as quote generation use more. A 22-sheet set with 600 detections is 22 × 500 + 600 × 20 = 23,000 tokens, so two runs fit inside the 50,000 trial allowance. Its own example, a 50-page set with 1,000 detections, is about 45,000 tokens, which allows one run. Pick a set that allows two so you can run once on defaults and once after fixing the scale.

BuildVision also offers to run one bid package for you and send back the quantities with a match against your numbers. That is useful as a second opinion, but it is the vendor's run on the vendor's settings, so do your own as well. Beam AI, another name that comes up in searches, sells a do-it-yourself licence from $8,000 a year per trade, for HVAC, plumbing and steel only, and a done-for-you service from $25,000 a year. That is a different budget, and nothing here is a trial.

Run all three the same way, from one PDF

Fairness matters more than speed here. Use the same PDF, exported once at full resolution, the same person at the keyboard and the same clock. Do it in this order:

1. Upload the full set unchanged, without trimming sheets to flatter the tool.
2. Record the minutes to the first result, with a stopwatch rather than a guess.
3. Let the tool run on its default settings and save its quantities unedited, as a screenshot or export.
4. Then correct what you can: set any scale it missed, point it at the legend, remove duplicates. Record the minutes to a take-off you would be willing to price from.
5. If the allowance permits, run again from scratch with your corrections and note whether the result improved.
6. Fill in the scoring sheet the same day, while you remember which results came from which run.

A demo-only tool gets the same steps as requests. Send the PDF two days ahead, then ask the presenter to upload it unchanged while you watch, show the default count before any clean-up, set the scale on the enlarged plan live, and export the quantities to a file you keep. Ask too how long you keep access afterwards and whether a second run is allowed. If a vendor will only demonstrate its own sample set, record the tool as "no result" on every row. A tool that has not counted your drawings has not passed.

Confidentiality deserves a minute. The pricing pages I read say nothing about whether uploaded drawings train a model. Ask in writing before uploading a client's set, and read any confidentiality clause in the contract. If you can, pick a job with no sensitive site details.

Score the counts: pass or fail by tolerance, then weight

For each item, compare the tool's number with the key. An item passes if it is inside the tolerance. The weighted pass rate adds the weights of the items that passed. A percentage-error average is a poor measure, because a 10% error on a small item looks mild and still costs money. Here are the results for tools A, B and C against the key above.

Tools A and B are almost tied on 72 and 73, and C trails at 60. The tie is the interesting part, and it is why one number is not enough. B's three misses are two over-counts (carpet, and two extra doors that suggest a double-leaf count) and one fixture left out. A's misses include a window and eight sockets it never saw. The next section separates those two kinds of error.

Score the scale: one known dimension on every sheet type

A wrong scale makes every quantity on a sheet wrong by the same factor, so check it before trusting any number from that sheet. On each sheet type, pick one dimension printed on the drawing, measure it with the tool's own measuring function, and compare. Kreo's page says its AI finds the scale on a sheet and asks you to confirm it, which is the step to watch rather than skip.

Here C halved every length on the enlarged plan because it read the wrong scale, which is the sort of fault that can explain a short partition total, and A and C both gave up on the scanned page while B read it 2.5% short. You then ask the right follow-up questions. Can the scale be set by hand in thirty seconds? Does the tool warn you? Does it remember the setting after you reload? A tool that fails safely, by flagging "no scale found", is better than one that quietly guesses.

Score what was missed, in dollars, not percentages

Under-counts and over-counts are different risks. An over-count is visible on review: you see 26 doors on the sheet and know there are 24. A missed item is invisible, because nobody sees what the tool never listed. Under-counts also hurt the margin on a fixed-price job, while over-counts risk the bid. So price each error with your unit prices from the key and add the two kinds separately.

Now the tie breaks. B's errors are mostly over-counts you would catch in review, and it left out only $410, so it is the safest to price from. A left $2,375 out, mainly in partitions and sockets, which a reviewer would not notice. C is the riskiest, with $4,532 left out. Against the pass mark written down before the trial, only B clears both bars: 73 against 70, and $410 left out against the $736 limit. A passes on the rate but leaves out more than three times the limit. The percentages hid that: all three tools pass most items, yet their costly errors are of different kinds. For each miss, write down why it happened: a symbol drawn unlike the legend, a partition on a thin line, a window behind a note. The pattern tells you whether a better setup could fix it or the tool simply does not read your kind of drawing.

One more column is worth adding if a tool marks its own doubts. BuildVision's page says every take-off carries confidence scores and flags ambiguous items for review. Note whether the flags sat on the items the tool got wrong. A flag on the right item turns an invisible miss into a visible one, while a confident score on a wrong count is the worst result in the sheet.

Time and price: what the subscription costs per take-off

Accuracy comes first, but time pays for the subscription, so measure both. Assume a manual take-off of this set takes 6.5 hours, which is 390 minutes. The three tools' times are below.

Then spread the monthly price over your take-offs. At four a month, Kreo Pro at $175 works out at $43.75 a take-off, and Togal or BuildVision Pro at $299 at $74.75. At $45 an hour of estimator time, that means each tool must save about 0.97 hours (Kreo) or 1.66 hours (the others) per take-off to cover its fee. At two a month the figures double, and at eight they halve. If a tool really saves four hours a take-off, as A and B did in the table, the fee is not the issue. The $124 a month between Kreo Pro and the $299 plans is smaller than one $1,040 error.

That is the reason to test accuracy before price. Billing terms still matter: Kreo bills once a year, Togal bills yearly, and BuildVision's monthly plan is the cheapest to leave. If you are weighing several tools beyond take-off, one platform or point tools sets out the wider choice, and price bands for estimating software gives context for these numbers.

Reading the scores: three outcomes and what to do

With scores in hand, you land in one of three places.

• One tool is clearly safest. It clears the pass mark you wrote down, leaves out little money and fails safely on scale. Buy it on the shortest billing term available and run it beside a manual check on your next two live bids before relying on it.
• Two tools tie. Choose on the type of error, then on workflow and price. Prefer the one whose errors are over-counts you can see. Check how each exports to your estimating sheet, and see bid checks after the take-off for what to review next.
• None is within tolerance on your drawings. Do not buy. Note which sheet types failed, try again in six months, and meanwhile keep your manual process, perhaps with a spreadsheet template and a second person's check.

Whatever the outcome, run the revision test before deciding. Take addendum 1, upload the revised sheets and see whether the tool shows what changed. Togal's page claims one-click comparison of drawing sets. If a tool cannot show you changes, you will redo the check by hand on every addendum, which erases some of the time saved. Catching changes between drawing revisions goes into that separately.

The sample-plan trap: why a clean demo set misleads

An estimator books three demos, and each vendor opens a clean, vector-based sample set: crisp lines, a clear legend, everything at one scale. Each tool performs well. She buys the one with the friendliest interface, then runs her first live bid, a scanned set from an architect with a smudged legend and three scales on one sheet. The tool misses a block of partitions, and she only catches it when a subcontractor's quote does not reconcile with her numbers.

The sample set was the problem, not the tool. It showed best-case behaviour on drawings she would never receive. A trial on a real, slightly awkward set would have shown the weakness before she paid. That is why the key above includes a scanned sheet, an enlarged plan with a different scale and a revision.

A trial calendar built around the three clocks

This calendar respects each trial clock. Day numbers count from the day the key is finished.

Day 0   Key finished (items, units, tolerances, weights, unit prices).
Day 1   Book the Togal demo (scheduling takes longest); confirm run-your-own-set.
Day 2   Start BuildVision trial (14 days, card needed; first charge on its day 15).
        Run 1 on defaults. Save unedited quantities. Fix scale. Run 2.
Day 3   Start Kreo trial on Pro (7 days, so it ends on Day 10). Run, correct, re-run.
Day 5   Togal demo: upload the same PDF, record the session, save quantities.
Day 6   Score all three: pass/fail, weighted pass rate, dollars left out, dollars added,
        scale table, minutes to checked quantities.
Day 7   Revision test with addendum 1 on each tool.
Day 8   Decide: buy one, buy none, or run a second round on a different job type.
        Cancel anything you are not buying the same day.

Two habits protect you. Save every tool's unedited output before you correct it, because a tool's defaults are what you would get on a rushed bid. And repeat the test on a second job of a different type if you are close to buying, since one set can flatter or punish a tool by accident. The trial costs a few hours of estimator time. A wrong purchase costs a year of subscription and a quantity error nobody sees until a subcontractor's quote disagrees.

The first two live bids after you buy

A trial proves a tool on one finished job; the next live bids prove it on your normal work. For the first two bids after buying, run the tool and a quick manual count of the three highest-weight items side by side, which for the key above means partitions, sockets and carpet. Log three numbers per bid in the same sheet: minutes to checked quantities, items outside tolerance, and dollars left out on those three items.

If either bid leaves out more than 1% of the value of the items you checked, look at the scale and legend settings before blaming the tool, then run it again. If both bids pass, drop the side-by-side to a monthly spot check of one high-value item, and put the renewal date in the diary with the scores attached. A renewal decided on twelve months of spot checks is a much easier call than one made from memory of a good demo, and if the scores slip after a software update, you will see it in the log before a subcontractor's quote shows it.

Questions estimators ask before running a trial

How many sheets should the test plan set have?

Fifteen to thirty is enough to include floor plans, elevations, schedules and one messy sheet. Size matters for BuildVision's trial: its page counts 500 tokens a page and 20 a detection, so a 22-sheet set with 600 detections uses about 23,000 of the 50,000 trial tokens. That leaves room for a second run after you fix the scale.

Should I test on a job I won or one I lost?

Use a finished job you won and built, because you know the real quantities from the final account and site. A lost job has no ground truth, and a bid estimate may contain your own counting errors. If you must use a bid, mark every number you are unsure of and leave those lines out of the score.

Is it safe to upload a client's drawings to a trial?

The pricing pages I read for Kreo, Togal and BuildVision do not say whether uploads are used to train a model. Ask in writing before uploading, check any confidentiality clause in your client's contract, and choose a job with no sensitive site details if you can. A vendor that answers vaguely has told you something.

What if a tool offers only a demo, not a trial?

Ask to run your own plan set live in the demo, record the session with permission, and score the numbers afterwards exactly as you would in a trial. Togal's page routes you through a booked demo and states no free-trial length, so ask how long you keep access and whether a second run is allowed.

Further reads

• AI Drywall Estimating Software: Plan or Photo Takeoff for a Small Crew? — Plan or photo take-off for a drywall crew.
• AI Estimating Software for Flooring: Test Accuracy on Your Own Plans — The same own-plans test applied to flooring.
• AI Estimating Software for HVAC: Replacement Quotes and Duct Takeoffs — What HVAC take-off tools measure and miss.
• AI Estimator for Contractors or a Human Estimator? How to Choose — When a person still beats the software.
• Asphalt Paving Estimating Software: What AI Adds and What It Costs — AI in paving estimating means measuring and takeoff, not bidding. Four setups priced from vendor pages, plus a tonnage sum and a five-lot test.
• Best AI App for General Contractors: Site and Office Tools Compared — Site and office AI tools priced per month for a six-person firm, each with a worked example, plus four stacks from $189 a month.
• Best AI Software for Construction Companies: Audit What You Own First — A subscription audit for a small construction company: what AI QuickBooks, Microsoft 365 and CompanyCam include, and which gaps justify buying.
• Best AI for Quantity Surveyors: Five Tool Types Compared by Job — Takeoff software, spreadsheet AI, chat assistants, contract review and site capture, scored per quantity surveying job with prices and checks.
• Blinds Quoting Software: Measure-and-Quote Tools and What AI Adds — Five blinds quoting tools compared on price lists, on-site measuring and cost, plus what their AI really does and a 20-cell audit to run before you trust it.
• Bill of Quantities Template: A Worked Example Checked With AI — A worked bill of quantities with Excel check formulas and AI prompts for codes, descriptions, units and totals, with measurement left to the QS.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Kreo pricing page; Togal pricing and home pages; BuildVision AI pricing and home pages; Beam AI pricing page. All read on 29 October 2026. Example plan set, tool results and unit prices are invented for illustration.