Judge it on three things: whether it reads your drawings in the format they actually arrive, whether the cost catalogue is yours to maintain, and whether the output fits the contract you sign. The AI features come fourth.
Key takeaways
Most software comparisons for construction estimating are feature checklists, which is why they are useless. Every product lists takeoff, catalogues, assemblies and reporting. The differences that decide whether you keep using it in a year are not on the list, and they come down to four questions you can answer during a trial.
Not as they should arrive. As they do. If half your work comes from architects on Revit-generated vector PDFs and half comes from a property manager sending a photograph of a 1987 mechanical drawing, those are two different software problems.
Vector drawings carry real geometry, which is what makes on-screen measuring possible. Scanned drawings are pictures until something processes them. Bluebeam's documentation on OCR describes the step precisely: OCR transforms scanned documents into searchable PDFs by detecting text on the page and translating those characters into searchable data. That makes text findable. It does not make the linework measurable, and you still calibrate the scale by hand.
The practical test during a trial: take your five worst incoming drawing sets, not your best, and run them through. If the tool handles those, the good sets take care of themselves.
This is where a lot of buyers are casual and later regret it. Procore's Estimating documentation describes a customizable database of parts, assemblies, equipment and services, with material prices, labour units and profit margins you adjust. That is the right shape — the numbers are yours.
The questions to ask are ownership and exit. Who maintains the catalogue, you or the vendor? If a vendor supplies starting prices, are they Canadian, and how often are they updated? Can you export the whole catalogue, with assemblies intact, in a format you could load somewhere else? Two years of catalogue maintenance is a genuine asset and you should not build it inside a box you cannot open.
There is a discipline point here as well. The same Procore documentation that describes catalogue browsing during auto-count warns that when you select a material from the catalogue, you should always check that it matches the client's specification and unit prices. That check is a habit, not a feature, and software will not install it for you.
The CCDC standard forms are the reference point for most Canadian commercial work, and they need different things from an estimate.
Match the output to the form
If you sign CCDC 2, the stipulated price contract — a single pre-determined fixed price for the project — you need a total you can defend and an exclusions list that survives contact with the consultant.
If you sign CCDC 4, the unit price contract, where the total is the unit price multiplied by the actual measured quantity performed, you need quantities organized by pay item, and you need to be able to re-measure and reconcile as the work proceeds.
If you sign CCDC 3, cost plus — actual cost plus a percentage or fixed fee — the estimate is a budget, and what matters is that its cost codes match the codes you will actually report against.
The last of those is the one to check hardest. If the estimating tool's cost codes do not map onto your accounting system's job cost codes, you have bought a quoting tool, not an estimating system, and you will never close the loop between what you priced and what it cost.
You will be uploading client drawings, tender documents and sometimes subtrade pricing given to you in confidence. The Privacy Commissioner's principles for generative AI technologies set out what responsible handling looks like — transparency about how information is used, explainability, safeguards, limiting the sharing of sensitive or confidential information, and privacy by design. Canada's Voluntary Code of Conduct on advanced generative AI systems sets out parallel commitments for developers.
Translated into questions for a vendor: where is the data stored, is any of our content used to train models, who at your company can see an uploaded drawing set, and what happens to our data if we cancel. Get the answers in the agreement, not in an email from a sales rep.
You will be shown percentages — time saved, accuracy achieved, hours per bid. Ask where each one is published, because vendor documentation is generally far more restrained than a sales deck: Procore's automated area takeoff page says the feature uses machine learning to detect areas and that results improve over time, which is a description of behaviour rather than a performance guarantee.
So generate your own number. Pick three completed jobs where you know the final installed quantities. Run their drawings through the trial. Compare. That number is worth every vendor benchmark combined, because it is measured on your drawings, by your people, on the kind of work you actually do.
There is also a legal backstop behind the skepticism, not just a business one. The Competition Act makes it reviewable conduct to make “a representation to the public in the form of a statement, warranty or guarantee of the performance, efficacy or length of life of a product that is not based on an adequate and proper test thereof” — and the Act puts the burden of proving the test was done on the person making the claim, not on you to disprove it. A vendor quoting an accuracy percentage should be able to say what the test was. If they cannot, that alone is worth more than the number.
Before you sign anything
Run the trial on your five worst drawing sets, not the vendor's sample project.
Confirm you can export the full cost catalogue, including assemblies, in a usable format.
Confirm the cost codes map to your accounting system's job cost codes.
Validate quantities against three closed-out jobs and write down the variance.
Get data storage, training use and post-cancellation deletion in writing.
Check what the renewal price is, not just the first-year price.
If your estimating and your job costing live in different systems that do not talk, you will keep re-keying. That is usually the argument for the platform. If your volume is low, a standalone takeoff tool feeding a spreadsheet you already trust is a legitimate answer.
Long enough to price a real job end to end, including a revision. A tool that is pleasant on a fresh drawing set and painful when addendum 3 arrives is a tool you will abandon.
Your estimate becomes contractually meaningful once it turns into a price and a scope. Payment timing, holdbacks and lien rights are governed by provincial legislation — our sister firm covers Ontario's regime in these notes on prompt payment under the Construction Act and what owners must hold back.
A 30-minute call is enough to tell you whether AI pays for itself here.
