Partly, and in a different way from a human estimator. A 2026 benchmark published by a takeoff vendor tested its own system, general AI models and professional estimators against expert takeoffs of ten residential plan sets. On framing, the vendor's system scored 82 on the benchmark's composite, the estimators 71 and the best general model 54. Across all nine trades, the vendor's system found more of the items a job needed, and the estimators got closer on the quantities they listed. A quantity counted as right if it fell within 25% of the expert figure, which is far looser than a lumber order. The result has not been reproduced independently, so it is best read as what is possible, not what a dealer should expect.
What the usual answer says
Takeoff vendors sell speed and accuracy. One vendor's guide to lumber takeoffs says that "Using AI takeoff software improves accuracy, reduces bid turnaround time", and gives no test set or tolerance for the accuracy. The claim of speed is fair. The claim of accuracy needs the measurement behind it.
What the benchmarks measured
General-purpose models struggle with drawings. AECV-Bench, published in January 2026 by Aleksei Kondratenko and colleagues, tested current models on floor plans and drawing questions. They found text reading strong, "up to 0.95 accuracy". The same models were weak at symbols: "reliable counting of doors and windows - remains unsolved (often 0.40-0.55 accuracy) with substantial proportional errors".
A purpose-built system does better, on its maker's test. Handoff-H1, published in August 2026 by researchers at Handoff, which sells the system, was tested on "10 real residential blueprint sets paired with consensus-validated expert takeoffs". Matching predicted items to the expert list was done by an AI judge, OpenAI's gpt-5.5, not by people. Across all nine trades, general models scored between 35 and 61 on the composite. Independent professional estimators scored 77.6%, with 65.5% coverage and 87.9% of quantities within 25%. Handoff-H1 scored 81.6%, with 86.1% coverage and 78.8% within 25%. The blueprint sets are available only on request.
The framing row is the one a lumber dealer needs: 82 for the system, 71 for the estimators, 54 for the best general model. The authors add that "Trades measured in linear or square feet (framing, siding, roofing) show consistently higher error rates than discrete-count trades (doors, windows, fixtures), in Handoff-H1 and in every baseline".
Two different kinds of error
Coverage and quantity precision are separate failures. On the benchmark's coverage measure, the estimators missed about a third of the items the experts listed, but were closer on the ones they kept. The system missed fewer items and was further out on more quantities. For a lumber order, a missed item means a second delivery, and a 25% tolerance on a wall's studs is the difference between enough lumber and a second trip.
What the drawing doesn't hold
A takeoff gives quantities in the plan's terms. The order is written in the dealer's: stock lengths, grades and species from its own catalogue, and allowances that depend on how a particular builder frames. None of that is in the drawing, so an AI reading only the plans can't produce the order. It needs the catalogue and the builder's past orders as well.
How to use it
Let AI build the first complete list from the drawings, since coverage is its strength on this benchmark. Have an estimator check the quantities for the items that cost most, since that is where precision matters. Then convert the checked quantities to stock sizes and lengths using the dealer's catalogue. Keep the quoted and delivered quantities for finished jobs side by side. That record becomes the local test the benchmark can't be, and it shows which item types the AI misses most often.
When it doesn't apply
A small, simple job that a skilled estimator can take off in minutes gains little from AI. Some plans are too incomplete for any takeoff, human or machine, to be reliable. And the benchmark covers residential plans only. Commercial and engineered-wood jobs may behave differently.
Quarri for wood products is built around how a wood products plant runs, where the order book meets real capacity.
Sources
- Chicelli et al., "Handoff-H1: An Orchestrated Vision-Agent System for Material Quantity Takeoff from Construction Blueprints", arXiv 2608.15032, 15 August 2026: arxiv.org
- Beam AI, "Everything You Need to Know About Lumber Takeoffs": ibeam.ai
- Kondratenko et al., "AECV-Bench: Benchmarking Multimodal Models on Architectural and Engineering Drawings Understanding", arXiv 2601.04819, 8 January 2026: arxiv.org
Quarri is an AI-native data platform for the timber supply chain. It connects buying, production, sales and inventory for forest management, sawmill, wood products and pulp, paper and packaging operators.