About & methodology

Means & Methods Lab (formerly PreconBench) is an independent lab for AI across construction technology, covering the whole project lifecycle: design and engineering, preconstruction, project management and field, safety, equipment and fleet, robotics, materials and supply chain, finance and payments, sustainability, workforce, and facilities/operations. It started from a heavy-civil estimating perspective, which still shapes the first benchmark. The name is a nod to "means and methods," the contractor's own choice of how work gets built. The brand is the face of the project. There is no founder photo or bio; results carry the weight.

Scope: the first planned benchmark covers AI takeoff, drawing review & QA, and contract & spec review. The directory is much wider. It spans every category on the categories page, each labeled AI-first, AI features added to an established product, or not AI-first.

CURRENT STATUS

No tool has been tested yet. Directory pages are desk research: every fact comes from the vendor's public website, the source URLs are listed on each page, and anything not publicly stated is marked "Not publicly listed". Vendor accuracy or speed claims are quoted as claims, not verified.

PLANNED TEST PROTOCOL

• Every tool runs on the same public NJDOT / municipal bid package, linked to its source and never rehosted (short excerpts only). Package A: takeoff (site/civil plan set + bid-item schedule). Package B: drawing review (same set + addenda; recall, precision, citation accuracy). Package C: contract & specs (~40 critical estimator clauses, e.g. notice, LDs, retainage, escalation, indemnity, DBE).
• An independent estimator builds a verified answer key from the bid-item schedule.
• Metrics: overall error %, missed items, false counts, minutes to first result, minutes to human correction — each with tool version and test date.
• Runs are screen-recorded and logged. Anthony signs off on every final result. Vendors get 7 calendar days to reply, and replies are published alongside results.
• Every result shows the tool version and test date. Re-tests after major releases or at least every 6 months.

INDEPENDENCE & DATA

• Rankings are never for sale.
• Vendor diagnostics stay private unless published with a "Commissioned" badge.
• Affiliate status never changes a rank or a recommendation.
• No fake or synthetic reviews.
• Client files are deleted after 30 days.