1. What the study did and what it found
The paper "Testing AI Accuracy in Quantity Takeoff" by Hooman Sadeh, Dominick Geloso and Dimitar Todorov was published on June 2, 2026 in the proceedings of the Associated Schools of Construction conference. They ran a commercial drawing set through Togal AI and compared 29 lines with the contractor's own takeoff: exterior finishes, floor and ceiling finishes, windows, doors and base trim.
The numbers from their Table 1. All ten window types and the glass doors matched the contractor exactly, 0.0% off. Square-footage lines ran low: carpet −4.2%, concrete floor −8.1%, one ceiling tile type −11.3%; the paper calls this "a consistent underestimation for area-based quantities". The façade went the other way: stone was counted 13.4% high. Two lines were simply wrong. Door type G: the contractor counted 6, the AI 10. Base trim: the contractor measured 12,978 ft, the AI 2,699 ft, which is 79% short. Arithmetic: on that line the AI missed more than 10,000 feet of material and labor.
The authors' conclusion in their words: AI "can accelerate quantity takeoff but should be used to augment, not replace, estimator judgment", and "estimator oversight remains essential for accuracy assurance in complex building elements". It performed best on repetitive, clearly bounded items and worst on irregular geometry and façades. One project, one tool, one study; that is all the independent evidence there is today.
2. What it costs and whether it helps a small contractor
Togal's pricing page, read October 10, 2026: the Growth plan is $299 a month per user, billed yearly, so $3,588 a year for one seat; larger teams get a custom quote. The same site says "up to 98% accuracy" and "5x faster" takeoffs. Those are the vendor's claims. The university study measured accuracy on one job and did not measure speed.
A Dodge Construction Network survey of 235 contractors with CMiC, published December 10, 2025, found 57% name "lack of reliability or accuracy in AI output" as a chief concern, and 49% of smaller firms see the cost of AI as a greater issue, against 26% of large ones. The study above explains both numbers.
Does it help? If you bid from full drawing sets every week, a tool that nails every door and window and gets floors within a few percent saves real hours, and the baseboard line is a known trap you check by hand. If you are a remodeler who measures the house with a tape and a phone, you do not have a drawing set to feed it, and $3,588 a year buys nothing.
Verdict: not worth the money yet for a residential contractor who works from site measurements. Worth trying cheaply only if you already bid from plans; then run one past job through the trial and compare every line with your own count before you trust a single number.
3. This week, if you let any AI read a plan
- 01Check every linear-foot line by hand: baseboard, casing, crown, gutters. That is where the test found the big miss.
- 02Recount every door type. The AI counted 10 of a door the contractor counted 6 times.
- 03Expect square footage to come in a few percent low, and stone or siding on a complex façade to come in high.
- 04Keep one past job as your test set. Run every new tool against it before it touches a client estimate.
Where WorkTime CRM fits, and where it does not
The AI Estimator in WorkTime CRM reads a plan too, at the scale of a house, not a drawing set: upload a PDF or a photo of a floor plan with dimensions under "Add floor plan, sketch or scan", or a Polycam room scan exported as "Blueprint + measurements", and it fills the rooms and the lines. After "Analyze" the panel "How AI read your plan" marks every size green (read from the plan), yellow (check) or red (guessed), which is exactly the check the study says a human must make. There is no separate charge: an AI estimate costs one AI call inside the subscription. What it does not do: it does not measure façades from elevation drawings, it does not do commercial drawing sets, and if the plan has no printed sizes it asks you for one real measurement instead of guessing. The quantities on the lines are yours to confirm before you press Send.