Workflow
Plans in. Calibrated bid out.
One workflow, nine steps, every line source-linked. Drop the plan set, watch the takeoff fill in row by row, ratify the confidence-gated edits, and ship a branded proposal — all inside the same workspace.
120
Sheets / minute
Customer-validated outcome
Recognizer accuracy after active learning
< 1 hr
Avg pipeline time
1+
Source links / line
The pipeline
Nine steps. Each one auditable.
Every step has its own confidence threshold; below it the run pauses for review, above it it auto-progresses. You ratify what you want to ratify; the rest just runs.
01
Intake
Drop the plan set, addendums, spec book, and any RFP attachments. We classify each document by trade, division, and revision.
02
Plan review
Read every sheet and every spec section into a structured scope of work. Every claim ties back to a sheet + paragraph.
03
Quantity takeoff
Per-trade symbol library + schedule parser produces editable line items with bidirectional source links.
04
Resource schedule
Crew composition, equipment hours, durations — generated from the takeoff with org-level overrides honored.
05
Direct cost
Material + labor + equipment loaded against your cost database; RSMeans fallback for missing SKUs.
06
Indirect / general conditions
Project-aware general conditions template, pulled from past jobs of the same project class.
07
Validation
Compliance checks against the bid documents — signed boilerplate, bond requirements, certification list.
08
Risk envelope
Monte Carlo distribution over the bid total based on per-line variance; you see 10/50/90 percentiles.
09
Markup + peer review
Markup waterfall by trade and bucket; AI peer reviewer grades the bid, flags scope gaps and pricing anomalies.
Confidence gating
The estimator stays in charge.
Per-step thresholds control auto-progression. Below the threshold, the run pauses on a review queue; above it, it advances. Senior estimators bulk-confirm trusted recognizers; junior estimators see every line.
- Per-trade confidence thresholds — electrical can be 0.85, demolition 0.95, your call
- Bulk-confirm above N% for senior estimators who've seen the project class before
- Reject a row → the recognizer's next bid learns from the negative example
- Confirm a row → it becomes ground-truth for the symbol library
- Audit log: every state transition is timestamped with the reviewer's ID
- Reopen any line at any time without losing the original belief score
What you ship at the end
A bid packet that reviews itself.
Editable takeoff
Spreadsheet-style line items with quantity, unit, labor hours, cost, source links, and confidence per row. Edits propagate.
Cost rollup
Material + labor + equipment broken down by CSI division with markup waterfall and Monte Carlo risk envelope.
Branded proposal
Cover letter, scope of work, schedule, exclusions, terms — all editable, your logo, your color palette, your boilerplate.
Audit packet
Per-line evidence graph, reviewer history, and the model versions used. Hand it to the senior PM and the bid runs without questions.
Frequently asked
What estimators ask before signing on.
- How long does a 1,000-sheet plan take?
- Roughly 8 minutes for the first read, plus your review time. The system streams results so the first sheet is reviewable while the rest finishes.
- Can I run only some steps?
- Yes — each step is opt-in per project. Stop after takeoff, run the full pipeline, or anything in between.
- What happens to my edits if the takeoff is regenerated?
- Edited rows are pinned (reviewed_by_user=true) and survive re-extraction. The recognizer learns from your edits without overwriting them.
- Can I bring my own labor units / cost database?
- Yes. Org-level overrides win over the platform defaults across labor, material, and equipment lines.