Software hands you a total for outlets, switches, and fixtures on each sheet of a plan set. It hands you the same total with the same confidence whether it’s right or wrong. Call it the confident count: a clean number with none of the doubt a human estimator would attach to it.
In Quebec, that number carries a particular weight. Once your bid is deposited and the withdrawal window has closed, you can’t take it back. An automated count deserves the treatment of a draft: fast, useful, and checked before you sign.
Automatic counting is already inside the tools you know
Takeoff, the step where someone counts every device and measures every run, is where the hours go. A vendor guide (Aginera) puts a mid-size commercial takeoff at two to three full days. Quotr claims it brings that from roughly 20 hours to one or two. The Business Development Bank of Canada (BDC) says AI cuts estimating time by 20 to 40%. All three are vendor or secondary figures, not independent measurements.
The three long-established electrical estimating products already count symbols, and funded newcomers have joined them:
| Tool | Automatic counting | Reported price |
|---|---|---|
| Accubid Anywhere (Trimble) | LiveCount AI recognizes receptacles and switches and sets the drawing scale (Trimble announcement, December 24, 2024) | About $3,000 per seat per year. Other sources say $5,000 to $15,000 or more with implementation |
| ConEst IntelliBid | SureCount, sold separately, a patented automatic count (per an Aginera comparison) | $1,500 to $2,500 per year |
| McCormick | Automatic count listed in vendor comparison guides | About $1,500 per year |
| Drawer AI | Electrical only; its site says 200-plus electrical contractors onboarded | $150 or $300 per user per month; the vendor’s own pricing page shows no figure |
| Quotr | Automated takeoff; raised $4M on September 30, 2026 (Quotr announcement) | Not published |
Prices come from review sites and forums, not from the vendors, and none states a currency. Trimble wrote in November 2025 that its tool detects outlets, switches, and fire alarm systems, with an average of 95% of drawing scales set automatically. That’s the maker’s claim. Drawer AI reportedly raised $5M and Togal $22.65M, according to industry reports. The field is crowded, and the crowd has money.
The Quebec tools, as they describe themselves
Quebec had 3,992 firms holding an electrical licence on July 31, 2025, according to the Corporation des maîtres électriciens du Québec (CMEQ), and membership is mandatory. Four local tools serve part of that market:
| Tool | What it does, per its own page | Published price |
|---|---|---|
| Estim.ai (Montreal) | Automated takeoff, electrical first. It “detects between 80 and 95% of your components,” with CSV or API export. It returns quantities, not prices | No public price |
| SubmitX | Reads PDF plans and specifications; carries the 2026 collective-agreement rates | Free, $99, or $299 a month |
| Estimation.ca (e-Plans) | On-screen takeoff and estimating, “for more than 30 years” | $39, $99, or $299 a month |
| Gestion CMEQ | The CMEQ’s own software, linked to major suppliers’ price lists. It doesn’t read plans; it imports a takeoff done in e-Plans | No price seen |
Estim.ai chose not to price. A La Presse profile from September 29, 2025 quotes the company: “We don’t want to give prices.” Pricing stays with the estimator, by design.
What general AI models can read today
A general-purpose model, with no training specific to electrical work, can now draft a first count from a drawing sheet. The published measurements tell two stories.
In January 2026, AECV-Bench found that models read text and answer questions about construction documents at scores up to 0.95. Counting symbols stayed “unsolved,” at 0.40 to 0.55 for doors and windows.
In September 2026, the consulting firm COXIT published CaseV-Bench: 119 pages from nine public drawing sets, 1,430 hand-marked objects of five types (cabinets, countertops, floor plans, interior elevations, callouts). The best model reached a 92% F1 score (a measure that balances objects found against false alarms) at $0.22 a page and 39 seconds. Across the nine projects its score ran from 73% to 100%. The lowest median cost among the models tested was $0.04 a page. For a 100-page set run page by page, the report puts the bill at about $22 and about an hour. Reading a 30-sheet set twice, 60 page-reads, would cost roughly $2 to $13 at those rates (our arithmetic from the published prices).
Three limits apply. A consulting firm ran the test, and no peer review has taken place. Its objects are large architectural elements, not small electrical symbols. And three searches turned up no public measurement of counting electrical symbols or tracing circuits on real building plans. The report itself concludes that the model still doesn’t replace the person.
Between January and September, one model went from counting rated unsolved to 92% on big objects. The two measurements cover different objects, but the direction is plain: whatever you test today will be out of date soon. That’s one more reason to test on your own drawings and not trust an average.
Where it fails
Vendor accuracy figures are the vendors’ own, unaudited. Quotr publishes 95 to 99% on clean vector PDFs and 80 to 88% on low-resolution scans. Estim.ai claims 80 to 95%. At Quotr alone, the gap between a clean PDF and a scan runs from 7 to 19 points, and your plan set lands somewhere inside it.
Practitioners on forums name the same weak spots: poor scans, lines crossing a symbol, dense sheets. One estimator on Reddit summed up a trial of one tool as “about 50% right, 20% missed, and 30% confidently wrong.” That anecdote can’t be verified, so treat it as a story and weigh it accordingly. It still shows the shape of the risk: the costliest error arrives looking exactly like a good result.
Drawer AI states on its own site that AI won’t be able to fully automate electrical estimating in the foreseeable future. A specialist vendor says so itself.
A deposited bid doesn’t come back
The Bureau des soumissions déposées du Québec (BSDQ), the province’s bid depository, covers bids of $20,000 and up in a covered specialty: electronic filing only, and a 10% bid security from $100,000. A bid can be withdrawn until eight working hours before closing. After that, in the BSDQ’s words, the bid “remains irrevocable.” Withdrawing costs 1% of the pre-tax bid amount, up to $5,000, and a bidder who withdraws can’t win the contract. A 1997 Court of Québec decision upheld that fee as a penalty clause.
Margins make the stakes concrete. Electrical contractors average a net margin of about 6% (US tax data from 2022, as reported by a software vendor). On a fixed-price bid, a takeoff that misses 5% of the job’s cost eats most of that margin. A count that runs high loses the contract on price. Either way, nobody calls you back about the number once it’s filed.
Plan-based takeoff mostly concerns commercial and institutional work. According to Soumissions Québec, residential work is priced from flat rates per job type, which leaves little takeoff to automate.
Six questions before you trust a count
1. Measured on what, and by whom? A claimed accuracy comes from a plan set you haven’t seen. Ask for the test set, its size, and the symbol types. No public benchmark covers electrical symbols, so every figure is the vendor’s.
2. What’s the rate per device type, not the average? A 92% average hid projects from 73% to 100%. For a bid, what matters is the rate on receptacles, on emergency fixtures, on panels, one by one. An average can hide a whole category that was missed.
3. What about my own drawings? Are yours vector PDFs or scans? Are the sheets dense? At Quotr, by its own test, the figure falls from 95 to 99% on one side to 80 to 88% on the other.
4. What does the tool do when it’s unsure? A good tool flags doubt: an unreadable symbol, a missing legend, two readings that disagree. A poor one prints a clean number. Ask to see what it does with a sheet it reads badly.
5. Who answers for the error, by contract? The terms of three takeoff tools reviewed for this article put verification on the user and disclaim any guarantee of accuracy. Five US estimating services cap their liability at the fee paid and make the client responsible for checking quantities. Read yours. If a clause caps liability, whether it holds under Quebec law wasn’t checked here; that’s a question for a lawyer or a broker.
6. How long does the checking take? If verifying a count takes as long as making it, the tool saves nothing. Time it.
The cheap test
Run a few past bids through the tool, bids whose takeoff you already hold. Compare by device type, using your own count as the reference. Note where it diverges, how long the checking took, and what that would cost per bid. A vendor that refuses the test has answered question 1.
A count that passes still keeps its role: the tool proposes, your estimator decides. The number that goes to the BSDQ is the one someone checked.
If you’d like to talk it through for your own business, the consultation is free.
FAQ
Can AI count outlets and switches on electrical drawings?
Yes, as a draft. Trimble, ConEst, and companies such as Quotr and Drawer AI already sell automatic counting. General models are improving fast: a benchmark published in September 2026 gave its best model a 92% F1 score, but on large architectural objects, not electrical symbols. No public benchmark covers electrical symbols.
How accurate is an automated electrical count?
Vendors claim 95 to 99% on clean vector PDFs (Quotr) and 80 to 95% of components detected (Estim.ai). Quotr’s own figure for low-resolution scans is 80 to 88%. These are vendor numbers, unaudited. Only a test on your own drawings gives you yours.
Can you withdraw a deposited bid if the takeoff was wrong?
Until eight working hours before closing, through the electronic system. After that the bid stays irrevocable. Withdrawing costs 1% of the pre-tax amount up to $5,000, and a bidder who withdraws can’t win the contract.
Who is liable for a counting error made by software?
Under the terms we reviewed, you are. Three takeoff tools put verification on the user and disclaim accuracy. Five US estimating services cap their liability at the fee paid. Whether such a cap holds under Quebec law wasn’t checked here; a lawyer or a broker can answer.