Inside Bid Reasoner: how it levels 5–15 sub bids per package.
The full workflow, every feature, and the trust model — the depth behind the homepage, for estimators who want to see how it actually works. ← Back to home
Bid leveling is still done by hand.
Heavy-civil GCs receive 5–15 subcontractor bids per package. Estimators normalize apples-to-oranges scope by hand, chase missing line items, and second-guess the low number. And then they pick a winner from gut feel — with no audit trail when the choice gets questioned six months later.
Storing bids isn't the same as reasoning over them.
A folder of PDFs keeps the bids. It doesn't line them up against your scope, flag the unpriced items, or write down why you picked the winner.
Documentation
- PDFs upload and sit in folders
- Your estimator still normalizes scope manually
- No risk flags, no leveled totals
- The reasons behind the award live in email and memory
Reasoning
- Extracts line items from PDF, Excel, Word, CSV and image bids
- Scores bidders across six dimensions; saved AI findings cite the document and page
- Recommends a winner — with a paragraph of reasoning and a confidence score
- Marks an override when you go against the recommendation and asks for a category and a reason
From sub bids to a defensible award.
Five steps, from sub invitations to the award paperwork.
Invite or upload
Invite subs by email; they get a link to a portal where they accept and upload — no password to create. Or upload a bid that reached you by email yourself. PDF, Excel (.xlsx, .xls), Word (.docx), CSV, ZIP and images are accepted.
Read & normalize
Line items are extracted from each bid, mapped to your scope of work, and normalized so bidders are comparable. Scope-coverage gaps surface automatically — items in your SOW the bidder didn't quote, items the bidder bid that weren't in your SOW. For scope items a bidder didn't quote, Bid Reasoner plugs a labeled price at the median the other bidders quoted. Works in any state. Scanned image-only PDFs may not be read — ask the sub for the native file.
Score & recommend
Bidders are scored on six dimensions: price, scope alignment, schedule, compliance, past performance, and risk. Analyze Package recommends a winner with a confidence score and a paragraph of reasoning — every saved finding names the bidder, the source document and the page.
Challenge & accept
Re-run Analyze Package under another decision mode (Lowest Responsible Bid, Best Value, Schedule Priority, etc.) to see the trade-offs. Ask a what-if question — "what if the GC absorbs site welding?" — and get a written answer that says whether it would change the recommendation. Accept or reject each finding; your verdicts stay on the record.
Award & defend
Pick a winner. If you agree with the AI, just capture rationale. If you disagree, the award page marks it as an override and asks for a category and a reason, and both print in the award memo. Bid Reasoner generates the Reasoned Award Memo, Evidence Report, Bidder Scorecard and four other decision documents in Word, alongside the award letter, LOI, draft subcontract, non-award notice, four exhibits, three Excel reports and a PDF packet. Winner and decline notices are a separate step you review first.
Built for the way estimators actually think.
Every adjustment is labeled. Every leveled total shows its work. Your senior estimator can still argue with it — that's the point.
Decision Workbench
Per-package recommendation page with AI bidder evaluations across six dimensions, findings that cite the bidder's own document and page, a confidence score, and a what-if challenge box. Re-run Analyze Package under another of the seven decision modes to see how price-priority vs risk-priority changes the winner. The single most important page in the product.
- Custom Weighting — tailor each of the six weights (price, scope, compliance, schedule, performance, risk) to your project's priorities.
- Mode chips preview, don't run — switch modes to see the weights, re-run only when you're ready. No AI burn.
Override capture & audit trail
When you award against the AI's recommendation, the award page marks the decision as an override and asks for a category and a reason, and both print in the award memo. An append-only audit trail records who selected the winner, edited the decision, approved or revoked it, and when.
- Decision Review Needed alerts list what to clear before award, such as findings waiting for your review, a low-confidence run or an addendum issued after the run — on the dashboard AND inside the package.
Auto-generated decision documents
Seven Word document types — Reasoned Award Memo, Bid Reasoning Report, Evidence Report, Bidder Scorecard, Executive Summary, Clarification Recommendations, Project Handoff Rationale — populated from the underlying data. Real .docx, not screenshots. Pull a fresh copy any time until you lock the final set.
Scope-coverage gap analysis
For every bidder, surfaces the line items in your SOW that the bidder didn't quote — and the items the bidder bid that weren't in your SOW. Find scope gaps and scope creep before award, not after. Works in any state: bid line items are matched to your scope by pay-item code.
Deterministic risk findings
Not opaque AI: unbalanced unit prices (at or below $1.00); peer outliers (2× or more, or half or less, of the median for that pay item, once three or more bids price it); total-bid outliers (20% or more above or below the median total, once three or more bids have totals); and front-loaded mobilization (10% or more of the bid total, rated high at 15% or more). Click any chip to filter the matrix to those flagged items.
Pay-item normalization & baseline pricing
Every bidder's line items are matched to your pay items and normalized against the peer median of your own bids — no government data required, and it works the same way in every state. When the package is leveled, a scope item a bidder didn't quote is priced into that bidder's leveled total at the median the other bidders quoted, matched by pay-item code. A bid with no pay-item codes on its lines gets no plug.
Tracked clarifications, no email chains
Every Q&A with a sub is a tracked thread, not a buried email. Subs answer through their portal. A response can be marked as revising pricing, so you can update the leveling. The clarifications you include become an award exhibit.
Sub solicitation & portal
Invite subs by email — they get a link to a portal where they accept or decline, upload bids through a 4-step wizard, and acknowledge addenda, with no password to create.
Built for an industry where decisions get questioned.
When the project owner asks "why this bidder?", the answer needs to hold up to legal review.
Tenant isolation
Each company's projects, bids and awards are kept in its own workspace. Automated tests try every page with another company's IDs to confirm nothing leaks between workspaces. Your bids, scopes, and awards are never visible to another GC.
Audit log on staff actions
Feature changes, plan changes, budget changes, manual retries, email resends. Every super-admin action is logged with a timestamp; plan, budget and feature changes also record the before/after values and any reason given.
Read-only support mode
When our team needs to look into your tenant for support, we enter Support Mode with a required reason (a ticket reference is optional). All writes are blocked at the middleware layer for the duration. 60-minute hard auto-expiry. Every blocked write is logged.
Explainable by design
Every saved AI finding names the bidder, the document and the page. Every leveling adjustment is labeled with its basis, and you can remove it or add your own. Override anything; the override is part of the record.
Bring a live bid package. We'll level it with you.
Bid Reasoner is in private beta and accepting its first founding partners: heavy-civil general contractors who get founder-direct onboarding on a live package. Bid Reasoner doesn't train or fine-tune any model on your bids.