Construction · Deep dive

Beam AI (by Attentive.ai)

Human-vetted AI takeoff for construction — upload site plans, get quantity takeoffs back in minutes to days, with an in-house QA team signing off before delivery.

emerging

The question that decides it: Beam AI's wedge is a QA-reviewed takeoff service priced as software, with real accuracy discipline and 1,100+ paying contractors — does that human-in-the-loop layer stay defensible as pure-AI takeoff tools race to zero and the platforms that own the drawings (Autodesk Forma / ProEst, Trimble, Procore) bundle 'good enough' takeoffs into what a GC already pays for?

My take ★★ Very interesting ★ Interesting − Maybe ✗ Not interesting

Snapshot

Beam AI is the construction takeoff product built by Attentive.ai — a 2017 India-founded company that spent five years teaching computer vision to read satellite imagery for landscape and outdoor-services estimators, then pointed the same pipeline at architectural and civil drawings and launched Beam AI in 2023. Attentive.ai closed a $30.5M Series B led by Insight Partners on November 12, 2025 (with Vertex, Tenacity, InfoEdge), bringing total capital to ~$48M. Beam AI is used by 1,100-1,200+ contractors and suppliers across the US and Canada and has processed 500,000+ takeoffs. What is distinctive is not the model — half a dozen YC batches are doing OCR-plus-LLM on drawings — it is the operating model: every takeoff is reviewed by an in-house QA team before delivery, priced as an annual license. The open question is whether that human layer is a moat or a liability.

Founding story

Shiva Dhawan and Rishabjit Singh, out of the IIT Delhi ecosystem, started Attentive in 2017 with a co-founding group including Utkarsh Sharma, Sarthak Vijay, Aishwarya Maurya and Arun Singh. The original bet was different: US landscape maintenance and outdoor-services firms — the crews mowing office parks and re-striping parking lots — that were driving to properties just to measure them. Attentive’s pitch was that satellite, aerial and drone imagery plus deep-learning object detection could measure a property to within a couple of percent from the office. Sequoia India’s Surge and InfoEdge Ventures led a $5M seed in July 2021.

Two things pushed the construction pivot: the industry had the same manual-measurement problem on a much larger addressable spend, and the CV/ML muscle built for satellite imagery retrained onto architectural and civil PDFs. Beam AI launched in 2023; by late 2024, blended ARR reportedly reached ~$12M and construction was the line investors funded. Both the February 2024 Vertex-led Series A (extended by Tenacity) and the November 2025 Insight-led Series B were about the construction business. The landscape/roofing origin is why Beam AI’s trade coverage is unusually broad in low-glamour outdoor verticals — paving, landscaping, roofing, flooring — that most competitors ignore.

How it works

Follow a bid. A subcontractor — say a mechanical firm bidding an HVAC package — uploads the drawing set and confirms scope: which trades, floors, systems. Then the customer picks the workflow. DIY returns a full AI takeoff in ~10 minutes at a claimed ~90% feature-capture accuracy; the estimator owns the last-mile review. Done-for-you routes the job through Beam AI’s in-house QA team, which runs cross-sheet validation, catches the corner cases the model missed (revision clouds, buried notes, non-standard symbols), and delivers a QA-reviewed Excel takeoff in 24-72 hours, marketed as within plus or minus 1% of in-house benchmarks. A hybrid path gives AI-first output with a human sign-off on high-stakes bids.

The AI-plus-QA loop does two jobs. The visible one is quality — a delivered artifact rather than a probabilistic output. The less visible one is data collection: every QA correction becomes labelled training data on the exact conventions estimators flag. In theory, the human layer thins as the model closes the gap. Whether it thins fast enough to defend the price point is the whole ballgame.

Product and business overview

The product surface has widened since the 2023 launch:

• AI Takeoffs (core) — automated quantity takeoffs from drawings with QA sign-off. DIY, done-for-you (24-72 hour SLA), and hybrid modes.
• Trade coverage — launched with HVAC/mechanical, then expanded to concrete, rebar, utility/earthwork, civil, structural steel, electrical, plumbing, masonry, demolition, painting, roofing, landscaping, flooring and paving. Each added trade opens a new subcontractor segment and creates cross-sell lanes into GCs.
• BIM CoPilot (2026) — human-vetted BIM management for the build phase, extending the “AI + QA team” pattern beyond preconstruction. Beam AI’s first serious step past takeoffs.
• Estimating outputs — Excel deliverables designed to drop into existing bid workflows, plus structured cost calculations for supplier-side quoting.

The buyer split spans GCs, trade subs and suppliers — suppliers sizing and quoting quickly are a different budget and sales motion than subs buying to lift bid volume.

Business model and pricing

Beam AI charges an annual, license-based subscription tailored to trade mix and expected bid volume — not per project, not per takeoff. The company explicitly positions this as predictability: a contractor commits to a fixed annual line item and gets unlimited estimating capacity within it. Pricing tiers and dollar figures are not published on the site; the company’s pricing page routes buyers into a sales conversation. Publicly available reviews and comparison write-ups indicate that Beam AI sits at the premium end of the AI-takeoff category, with price cited by some buyers as the primary purchase barrier — a sign the QA labour is real cost that is being passed through.

The gross-margin question is the interesting one. If Beam AI’s economics look like SaaS (80%+ GM), the QA team is small relative to the license base and the moat compounds. If they look like KPO/services (30-50% GM), QA cost scales with volume and the exit multiple is much lower. Attentive.ai has not disclosed the split, and Insight priced the November 2025 round without publishing a valuation.

Traction over time

Headcount sits at ~628 (LeadIQ / ZoomInfo, 2026), ~388 in India — a labour arbitrage that is both a QA cost advantage and a hiring lever most US-headquartered peers lack.

Market analysis

Takeoff software was ~$2.1B in 2025, forecast to ~$4.9B by 2034 (Dataintelo, 9.8% CAGR). The broader estimating bucket — where Beam AI is expanding with cost calculations — was $2.73B in 2025 and forecast at $3.07B for 2026, reaching $5.58B by 2031 (Mordor, ~12.7% CAGR). North America is 38%+ of estimating spend (Grand View). Adoption is mainstream: a BuiltWorlds survey found nearly 90% of contractors have implemented some estimating solution, 64% on every project.

Three forces are moving the market: a labour shortage in preconstruction estimating, material-price volatility since 2022, and the AI step-change since 2023 that made “read a PDF drawing set” solvable for teams with far less capital than Autodesk. The last one is double-edged: it grew the market and invited every YC batch in.

Competitive intel

The set breaks into three layers, and Beam AI is fighting all three. AI-native pure-plays (Togal.AI, Kreo, Bild AI, Ediphi, the YC pipeline) push price toward zero; Togal.AI is the closest direct comp at 4.8/5 on G2 with ~60 reviews vs. Beam AI’s 4.9/5 with ~30, marketing pure-software speed rather than QA. Legacy takeoff and estimating SaaS (ConstructConnect On-Screen Takeoff, PlanSwift, STACK, Bluebeam Revu) own the incumbent seat, cost less per license, and are shipping their own AI features in 2025-26. The drawing-hosting platforms — Autodesk (Construction Cloud, Forma for Preconstruction, ProEst) and Trimble (Assemble Systems, Trimble Construction One, plus the April 2026 Document Crunch acquisition) — are the bundling risk; they need only build an adequate takeoff inside software the GC already licenses.

The Trunk Tools posture is the most instructive datapoint. Trunk Tools raised a $40M Series B from Insight in July 2025 — the same lead as Beam AI’s Series B four months later — and explicitly refuses to sell a takeoff product. In a mid-2026 Bricks & Bytes piece, CEO Sarah Buchner argued per-trade takeoff economics do not justify the R&D cost and the move is to make drawing-reading a substrate for higher-value agents. Whether Insight sees the two as complements or a hedge is worth asking.

History and evolution

Timeline is in the Traction table above; funding sequence in the capital table. Two things matter beyond the numbers. The pivot: Attentive.ai spent five years selling to landscapers and three selling to contractors without swapping CEOs — an unusual execution signal and the reason trade coverage skews to outdoor verticals no one else prioritizes. And BIM CoPilot in 2026 pushed the pattern past takeoffs into build-phase model management. The stumble the company will not name is the two-year gap between the 2021 seed and the 2024 Series A: the landscape business alone did not attract Series A capital; the Series A was priced on the Beam AI thesis.

What people say

The case for. G2 (4.9/5, ~30 verified reviews mid-2026), Capterra and Software Advice cluster on three themes: turnaround time (takeoff cycle from days to hours or minutes), QA and support quality (the phrase “top notch” recurs), and bid volume lift (~2x bids per quarter after adoption). Insight’s own memo leans on the same numbers — 90% time savings and 3x bid capacity for select customers. Third-party comparison pieces (Ruh AI, Dan Cumberland Labs) consistently rank Beam AI at the top of the multi-trade AI takeoff category and note that it handles complex multi-trade jobs well rather than optimizing for a single commoditized trade.

The complaints. Recurring negatives across G2 and Capterra are specific. The Excel output does not always align with a customer’s bid form, forcing double entry. Reporting granularity in utility/civil is criticized — sanitary sewer mixed with water main and storm sewer, requiring manual regrouping. Several reviewers note real onboarding time. Price is the primary barrier for buyers who choose a cheaper AI-native pure-play. And the loudest strategic critique — from the Bricks & Bytes construction-AI beat — is that per-trade AI takeoff accuracy is being commoditized weekly and a QA-heavy service model may not hold pricing power as the underlying capability turns free.

Outlook: the open question

The question that decides this company: does the QA-reviewed takeoff service, priced as annual software, stay defensible as pure-AI competitors race to zero and drawing-hosting platforms bundle ‘good enough’ takeoffs — or does Beam AI have to become something structurally larger before the middle collapses?

Bull case: 1,100+ paying contractors and 500K+ takeoffs is a live network of QA labels retraining the model daily — the flywheel needed to survive commoditization. Trade breadth in outdoor verticals is a defensible corner Trimble and Autodesk do not obviously care about. The India engineering-plus-QA base is a cost advantage US-headquartered pure-plays cannot easily replicate. Insight underwriting the round four months after leading Trunk Tools’ Series B suggests either category conviction or a portfolio hedge across opposite theses.

Bear case: the Trunk Tools refusal-to-sell-takeoff posture is a competitor with a stronger drawing-reading claim explicitly saying standalone-takeoff unit economics do not work. If they are right, Beam AI is sandwiched — pure-AI YC entrants pushing price toward zero from below, Autodesk/Trimble/Procore bundling adequate takeoffs from above. The QA moat depends on the AI staying imperfect enough that customers need human review, but Beam AI’s own roadmap is disarming that gap. If gross margin looks like KPO with a software wrapper, the exit multiples are not what the round was priced against.

Bull-case conditions: (1) QA-corrected training data thins the human review layer from days to minutes over 24 months without losing the plus-or-minus 1% accuracy, so gross margin climbs toward SaaS norms; (2) BIM CoPilot and downstream extensions turn Beam AI into a preconstruction-to-build-phase platform before the incumbents bundle it away; (3) outdoor and long-tail verticals prove defensible. Bear-case trigger: Autodesk Forma, Trimble Construction One or Procore ships an in-platform takeoff that is 80% as good and free with an existing license — because 80% and free beats 100% and priced, on the buyer side, every time.

How a challenger would attack it

Squeeze the QA layer from both ends and hit the output where reviewers already complain. Beam AI’s premium is paying for human review of AI output — a cost structure a challenger can attack by shipping model-only takeoffs at Togal.AI-style pricing while accuracy on commoditized trades (concrete, masonry, sitework) closes the gap, because 80%-as-good and cheap beats plus-or-minus-1% and premium for the SMB sub whose G2-documented top barrier is already price. The sharper wedge is Beam AI’s own delivery friction: reviewers complain that Excel outputs don’t align with bid forms (forcing double entry) and that utility/civil reporting lumps sanitary, storm and water main together. A challenger that integrates directly into the customer’s estimating stack — native bid-form mapping, trade-correct groupings, no re-keying — wins the workflow argument Beam AI’s generic Excel deliverable loses. Structurally, the attack Trunk Tools has already articulated is the most dangerous: don’t sell takeoff at all, make drawing-reading a free substrate under higher-value agents, collapsing the standalone category’s pricing umbrella. And Beam AI’s 24-72 hour done-for-you SLA is a service-window vulnerability — a challenger promising QA-grade output in under an hour, using LLM cross-sheet validation instead of a 388-person India review bench, attacks both the speed and the gross-margin question Insight priced the round without answering.

Same playbook, new buyer

AI-plus-QA-team, priced as software, transfers to any document-heavy estimation trade. The pattern Beam AI runs — probabilistic model output hardened by human review into a warranted deliverable — maps directly onto insurance property claims estimation (reading damage documentation against Xactimate line items), facilities and property condition assessments, MEP engineering design review, and permit-set compliance checking, all markets where the buyer pays for a signed-off artifact, not a probability. Within construction, the underexploited buyer is the owner/developer side: independent cost validation of GC bids is bought today from consultants at consulting prices, and a QA-backed AI takeoff sold as bid-check rather than bid-production monetizes the same engine against a wealthier, less price-sensitive customer with zero channel conflict against Beam AI’s contractor base — which is precisely why Beam AI itself can’t easily sell there. Geographically, Beam AI’s US/Canada focus leaves the UK, Australia and Gulf markets — quantity-surveyor cultures that already pay humans for exactly this deliverable — open to a challenger with the same India-cost QA bench. Attentive.ai’s roadmap is committed to deepening one buyer (BIM CoPilot, more trades for contractors), and its landscape-origins DNA means adjacent-vertical discipline, not adjacent-buyer expansion.

Sources and further reading

• Attentive.ai Secures $30.5 Million Series B to Accelerate AI Innovation in Construction — BusinessWire, Nov 12, 2025
• Attentive.ai Secures $30.5 Million Series B — Insight Partners, Nov 2025
• Attentive.ai snags $7M to boost automation in landscaping, construction services — TechCrunch, Feb 7, 2024
• Attentive.ai raises $5M venture funding from Sequoia India’s Surge & InfoEdge Ventures — Attentive.ai blog, Jul 2021
• Shiva Dhawan, Co-founder and CEO of Attentive.ai — Interview Series, Unite.ai
• A New AI Takeoff Tool Every Week: Why the Best Drawing AI in Construction Won’t Sell You One — Bricks & Bytes, 2026
• Beam AI Reviews — G2, 2026
• Takeoff Tool Comparison 2026: Bluebeam vs. Togal.AI vs. Kreo vs. Beam AI — Ruh AI Blog
• Beam AI Takeoff Pricing Explained — ibeam.ai
• Attentive.ai Expands Beam AI to Accelerate Construction Estimating — For Construction Pros, 2026
• Construction Takeoff Software Market Research Report — Dataintelo, 2025
• Estimating Software Dominates Preconstruction Tech Adoption — BuiltWorlds via ConstructionOwners, 2025