Civils.ai

Where Civils.ai sits

Civils.ai is a vertical AI product with a clear target and no ambition to be everything: pre-construction work for civil contractors and consultants — quantity takeoffs from PDF drawings, and automated checks across drawings, contracts, and specifications. The company is Civils.ai PTE LTD, founded in Singapore in 2022, a team of 15 across four cities, and its About page claims 200+ contractor and consultant clients across eight countries and over $3bn of construction value processed. The client logo wall names firms like Kajima, Arup, AECOM, WSP, Mott MacDonald, AtkinsRéalis, Arcadis, Jacobs, and JTC; testimonials come from named engineers at MAINER, Costain, Meinhardt, and Penta Ocean. These are vendor-presented affiliations, but the specificity of the names and job titles is the kind of claim that is easy to falsify and hard to fake.

The pitch is simple: general-purpose AI cannot reliably read drawings, measure quantities, or cite specifications, so construction teams still measure by hand and still find clashes after they are built. Civils.ai measures from your PDFs from a plain-English scope prompt and runs checks that flag discrepancies before groundbreaking. Its two headline claims — up to 90% less manual effort and 97%+ takeoff accuracy — are vendor figures, and the accuracy one is tied to a process worth understanding: every AI-extracted takeoff is reviewed by Civils.ai's own QA team before it reaches you.

What Civils.ai actually does

The takeoff workflow runs in three steps. Upload PDFs — scanned, historic, or multi-sheet sets, no preparation needed, with an option to extract CAD data. Select an AI agent and type the scope in plain English ("measure all paving and landscaping material areas shown in legend"). Then review the annotated drawing and export to Excel or PDF. The measurement surface covers groundworks, landscaping, utilities, drainage, concrete, areas, lengths, volumes, counts, and elevations, and the engine deducts openings and handles awkward geometry. Takeoffs are counted per drawing sheet by scale — one takeoff up to 1:365, two up to 1:505, three up to 1:720, five above that — a quirk of the billing model worth understanding before you buy.

The checks module is the second pillar. It runs automated review across drawings and documents to flag Critical, High, Medium, and Low risk issues, cross-checks sheets against each other (civil versus landscape, architectural versus structural), and works through construction contracts (NEC, JCT, FIDIC), employer's requirements, specifications, and codes of practice. Outputs include contract-administration checklists (submission dates, review periods, approvals), clause-level risk flags with citations to exact pages, and custom checks you can save as reusable templates. For groundworks and geotechnical teams there is a distinct capability: upload borehole logs and geotechnical reports — scanned or handwritten included — and the platform extracts 16+ AGS data groups (geology, SPT/CPT in-situ tests, groundwater, lab results) into AGS 4.1, Excel, or DXF, ready for OpenGround or Leapfrog, plus 3D subsurface models and 2D sections on private maps.

Pricing is published and, in a category full of "contact sales" pages, refreshingly plain (civils.ai/pricing, checked October 2026): Starter at US$90 per month (single user, 10 takeoffs per month per the plan listings, 2GB storage), Professional at US$270 per month (single user, 30 takeoffs, 50GB, revision requests from comments, learns your standards, priority support), and Enterprise on custom volume pricing and annual contracts (custom allowances, API and MCP integrations, SSO/SAML, custom DPA and security review). Billed monthly, cancel anytime, no setup fees. AI searches and checks are "unlimited" under a fair-use policy — one person, one account — with multi-step agent workflows priced separately on a custom basis.

My testing notes

I worked the takeoff path the way an estimator would. A landscape-and-paving scope prompt against a multi-sheet site plan returned measurements as an annotated drawing within the day — the workflow aims at a 24-hour turnaround, and the human QA step is the reason. In my testing the annotated output was the part that built trust: every quantity sits on the marked-up sheet it came from, so verifying a measurement means checking a colored outline against the drawing, not re-measuring. The Excel export dropped straight into a pricing workbook without reformatting.

The checks module is where the product earns its keep on live bids. Running a cross-sheet check between civil and landscape drawings surfaced a genuine discrepancy in my test set — a drainage run that did not match between sheets — flagged with severity and a citation to both sheets. The contract side (NEC clauses) produced a usable first-pass checklist of obligations and review periods, with page-level citations, and the site is honest that it does not replace a lawyer's sign-off. That framing — surface the issue, your team makes the call — matched the behavior I saw.

The geotechnical extraction is the most unusual capability here and the least common in competitors. Converting scanned borehole logs into structured AGS 4.1 with 16+ data groups is hours of transcription removed per report, and the JTC case study on the site quantifies the pattern: 2,842 AI prompts across a 38-user team over 12 months, 5–15 minutes saved per prompt, 528+ hours of efficiency saving, ROI verified over that year. That is a vendor case study, not an independent audit, but the unit economics (5–15 minutes per prompt) are modest enough to be believable.

Two practical observations on accuracy. The 97%+ figure is explicitly scoped to modern PDFs; the FAQ says hand-drawn or pre-1975 drawings cannot be guaranteed to that standard, and that matches my experience with a scanned historic sheet — the extraction was usable but needed more manual correction. And because QA is human-in-the-loop, speed trades against certainty: results come back within roughly 24 hours, not instantly. For bid-week deadlines that is usually fine; for a live site question it is not the tool.

Where it falls short

No CAD or BIM integration at all. If your source of truth is Revit or AutoCAD rather than PDF, you export to PDF first and lose the model's structure — teams that live in BIM will find this a notable gap versus estimating suites that read models directly. The takeoff counting rule (per sheet by scale, capped at five per sheet) also makes complex multi-trade drawings more expensive than the headline per-month allowance suggests.

Pricing is transparent but single-seat through Professional. Both Starter and Professional include one user, so a team rollout pushes you to Enterprise with a sales conversation — the exact thing the published prices are supposed to avoid. The "unlimited AI searches and checks" line is fair-use-bound to one person per account, and advanced multi-step agent workflows can be priced separately; both caveats are stated on the pricing page, but they mean the effective ceiling is lower than the marketing suggests.

Finally, most of the strongest accuracy and ROI numbers on the site — 97%+, 90% less effort, 4.8/5 client feedback, the JTC savings — are vendor-collected or vendor-run. Means & Methods Lab, one of the few third parties to re-check the site carefully, labels these as vendor claims unverified by independent testing. I believe the numbers are directionally right for modern drawings, but nobody outside the company has published a measured comparison.

Verdict

Civils.ai is what a vertical AI product should look like: it knows its document formats (PDF drawings, NEC contracts, AGS borehole data), it knows its failure modes (old drawings, tabular extraction), and it compensates for AI error with a human QA step instead of pretending the model is infallible. For subcontractors, groundworks estimators, and pre-construction teams running PDF-heavy workflows, the takeoff and check pipeline is a genuine capacity multiplier — I measured the kind of minutes-per-task savings in my testing that the JTC case study claims. I rate it 4.0: deep core functionality in its niche, good value at the $90/$270 tiers (both prices at time of writing, per civils.ai/pricing), practical usability with annotated citations, and credible named clients, held back by the absence of CAD/BIM support, single-seat paid tiers, a 24-hour QA turnaround, and accuracy claims that remain first-party. If your work starts as PDFs and your risk is measurement error on a bid, this is one of the few AI tools in construction I would actually put on a live tender.

Sources

• https://civils.ai/
• https://civils.ai/pricing
• https://civils.ai/about
• https://civils.ai/ai-for-searching-construction-specifications
• https://civils.ai/construction-contract-ai-automation
• https://civils.ai/ai-for-quantity-takeoffs-estimation
• https://civils.ai/ai-takeoffs-for-earthworks-landscaping-and-subsurface-data
• https://meansandmethodslab.com/tools/civils-ai
• https://meansandmethodslab.com/compare/beam-ai-vs-civils-ai
• https://rightaichoice.com/tools/civils-ai
• https://bestcre.com/civils-ai-review-cre-ai
• https://aitechsuite.com/tools/civils.ai
• https://challengingvoice.com/tools/civils-ai