Skip to content

Evidence, not slogans

Methodology

Every percentage on this site should be traceable. This page defines how we measured 28 minutes, 70% reduction, and 99.6% item capture — and where those numbers do not apply.

Plain language: These are benchmark results on internal plan sets, not guarantees on your plans. Your mileage varies with scan quality, sheet count, and spec clarity. Book a live demo on your own set for a real measurement.

What we measured

Plan to bid package

28 min

Median wall-clock time from upload to evidence-backed estimate ready for review.

Time reduction

~70%

Median reduction vs manual takeoff + estimate on the same sets.

Item capture

99.6%

Quantities correctly extracted vs ground-truth count.

Sample

  • —n = 47 heavy-civil plan sets sourced from past bid archives (de-identified). No customer production data.
  • —20–140 sheets per set, median 62 sheets. Vector PDF primary; 9 sets included DWG/DXF overlays.
  • —Trades: grading/earthwork (100%), underground utilities (72%), paving (55%), structures/bridges (21%).
  • —Specs: CSI MasterFormat Div 31–33 typical; DOT specs where present.
  • —Exclusions: hand-sketched as-builts, raster scans <150 dpi, and plan sets requiring >3 RFIs to disambiguate were excluded.

Baseline (what we compared against)

  • —Human baseline: 2 experienced estimators (8+ years heavy-civil) working independently with PlanSwift + Excel + manual spec lookup. Time measured from clean upload to first complete bid tab ready for senior review.
  • —BidSync AI path: Same upload → 8-agent run (or 1-agent on Starter) → evidence layer with trace to Plan/Detail/Spec → human approval gate. No manual cleanup counted inside the 28 min; review time separate.
  • —Both paths used identical plan revision and addenda.
  • —Hardware/network: Cloudflare Edge, no local SQL/RDP per claim.

Definitions

Item capture 99.6%
Correctly extracted quantity line items ÷ total ground-truth line items, per set, then averaged. An item is a takeoff quantity tied to a CSI MasterFormat line (e.g., 4,820 LF 12" waterline). Off-by->2% quantity or wrong CSI counts as miss. Evidence link (Plan C-302 / Detail W-14) must resolve.
28 minutes median
Median across 47 sets. p50 = 28 min, p90 = 46 min. Manual median on same sets: ~11 days elapsed (2 weeks calendar with interruptions) / ~34 hours active estimator time. We report elapsed 28 min vs active 34 hrs as ~70% when comparing active time; calendar reduction is ~95% but we do not headline it to avoid conflating elapsed vs active.
What is not included
Senior review, owner clarifications, bonding/insurance, and final bid strategy are outside the timed window. Poor scans, missing details, or spec conflicts increase time and reduce capture.

Reproduce it

Bring your own plan set to a live demo — we will run it side-by-side and share the trace sheet. Anonymized benchmark details available on request for peer review.

Last updated: Sep 2026 · Next review when n reaches 75 sets. Claims will be revised if median shifts >10%.

Related