Bid Leveling
Bid Leveling is the engine that compares two to ten vendor quotes for the same scope and recommends which vendor to award. It normalizes units and currency, aligns every line item across the bids into one table, flags outliers and unbalanced-bid tactics, and gives reasons for the recommendation.
What is Bid Leveling?
Bid Leveling is the workhorse engine for procurement. Feed it multiple vendor responses to the same RFQ (or the same scope described differently across proposals), and it produces a side-by-side comparison: every line item from every bid aligned into one table, prices normalized to a common currency and unit system, outliers flagged, coverage gaps identified, and a narrative recommendation at the top.
The engine also runs three unbalanced-bid detectors (per-line peer-outlier, front-loading, concentration risk) — the same patterns an experienced estimator scans for manually. These show up as cautions in the narrative and as explicit flags in the structured output.
When should I use Bid Leveling?
Use Bid Leveling once you have two or more responses to the same RFQ and need to choose between vendors with a documented recommendation, or to catch the vendor who is lowest only because they left something out.
- You've solicited bids (via an RFQ or direct outreach) and have 2 or more responses in hand.
- You're choosing between vendors and want a defensible, documented recommendation.
- You want to catch the vendor who's "lowest" only because they forgot to include something.
Bid Leveling works best when all vendors bid against the same RFQ. If you have an RFQ Generator output, chain into Bid Leveling and the RFQ becomes the reference spec — line items align automatically.
What do I upload to Bid Leveling?
Between 2 and 10 vendor quotes. Supported formats: PDF, DOCX, XLSX/XLS, CSV. Scanned PDFs do not work — there is no OCR on this path, so a scan comes through empty and the run is refused. Use the text-first original, or paste the quote text.
Optional — upload the original RFQ or spec as the reference spec. If provided, the engine matches every vendor's response against the reference so coverage gaps (vendor A missing line item 7) are explicit rather than inferred.
How do I run Bid Leveling?
To run Bid Leveling, attach two to ten vendor bids, optionally add the original RFQ as the reference spec, confirm currency and units, and click Run analysis, then read the recommendation, the leveled table and the cautions before you decide.
Open the engine
Sidebar → Engines → Bid Leveling card.
Attach vendor bids
Drop 2–10 vendor documents into the bid upload slots. Each slot corresponds to one vendor. Name each slot with the vendor name (or let the engine auto-detect the name from the document).
Optionally attach the RFQ
If you have the original RFQ/spec, drop it into the reference-spec slot. Highly recommended — improves coverage detection and makes "did vendor X miss item Y" an explicit flag.
Confirm currency / units
If vendors quoted in mixed currencies or unit systems (metric + imperial), the engine converts them to one common basis. Default is USD + imperial; change in Settings if your project uses different defaults.
Run analysis
Most bid tabs level in under a minute; bigger packages with more quotes take longer. Leveling is the most prompt-intensive engine in the platform.
Read the recommendation
Top of the result: recommended vendor, win-probability, reasons-to-bid, reasons-not-to-bid, clarifications needed. This is the AI's synthesis.
Scan the leveled table
Below the narrative: every line item across every vendor. Price columns per vendor, true-cost column (includes alternates, exclusions, allowances), deviation column (± vs peer average), risk flag column. Click any cell to see its source excerpt.
Check the cautions
Below the table: unbalanced-bid flags and any coverage gaps. Take these seriously — they're the kind of thing that bites you mid-project.
Decide and chain forward
Click Chain to Contract Review on the winning vendor to upload their contract and start the review. Or Chain to Negotiation to generate talking points against the recommended vendor.
How do I read the Bid Leveling results?
Start with the AI Recommendation card at the top of the result, then use the financial summary, coverage chart, leveled table, unbalanced-bid flags, reliability scores and cautions panel to check it.
Top of the result. Contains recommended vendor, confidence score, win probability (LOW/MEDIUM/HIGH), reasons for and against, and any clarifications to request before awarding.
Total bids, recommended award amount, potential savings from splitting the award across vendors (cherry-picking cheapest line item per vendor), spread between highest and lowest.
The summary tile shows the total of the vendor ranked first. That is not always the lowest total: vendors are ranked by the mandatory-requirements check, then by whether the quote matches its own stated total, then by the fewest lines left unquoted, then by the lowest total. A note under the tiles says which vendor the total belongs to, any vendor that quoted less and why it ranks below, and any vendor left out for having no usable total. The same note is in the PDF and the emailed report.
Donut chart per vendor showing percentage of RFQ line items they covered fully, partially, or missed entirely.
The main artifact. Rows = line items, columns = vendors. Highlight colors: green = lowest price for that line, amber = partial quote, red = missing or flagged.
Three signals surfaced when triggered: per-line peer outliers (≥25% deviation, ≥3 lines, or 1–2 at ≥40%), front-loading (first-N items disproportionately high), concentration risk (single line ≥40% of total).
A single 0–100 score per vendor, computed from this bid's own coverage, quantity mismatches, lead-time status, price relative to peers, and compliance. It reflects only this bid — it is not stored on the vendor's record and does not factor in performance from other projects.
1–3 short bullets on risks worth discussing before awarding. Includes unbalanced-bid callouts, quantity mismatches, and scope gaps.
Every control, explained
Add vendor slotAdds a new upload slot (up to 10 max).
Name vendorSet the vendor name for this slot. If left blank, the engine tries to auto-detect from the document.
Reference spec uploadOptional slot for the RFQ or spec. Turns coverage-gap detection from inference into ground-truth matching.
Run analysisConsumes one run. Most bid tabs level in under a minute.
Filter: All / Covered / Partial / Missing / ExtrasNarrows the leveled table to one status. Useful when reviewing coverage gaps across many line items.
Sort by varianceOrders the table by how much vendors disagreed on each line. Top rows = the most controversial items.
Chain to Contract ReviewOpens Contract Review pre-configured with the recommended vendor. Upload the signed contract.
Chain to NegotiationOpens the Negotiation Intelligence tool with the leveled data pre-loaded — generates talking points against any vendor.
Export PDFFull leveled report with recommendation, table, and cautions. Standard format for handing to a review committee.
Export XLSXPreserves the leveled table structure — each vendor is a column, each line item is a row. Useful for further analysis in Excel.
Chains
Chains into
Chained from
Exports
- PDF — formatted committee-review report. Executive summary, recommendation with reasons, leveled table across pages, cautions appendix.
- XLSX — raw leveled data. Columns per vendor. Useful when you need to run your own sensitivity analysis.
- CSV — flat line-item list with vendor columns. For import to procurement systems.
Troubleshooting
This is normal — vendors describe the same work differently. The engine does its best to match by description, quantity, and unit. If a match is wrong, click the row to see the source excerpts and use the chat to request a correction: "Line 7 from Acme is actually line 12 from Ashby."
Engine auto-converts using the rate at run time. If you need historical rates, set them in Settings > Currency before running.
Upload the reference RFQ/spec if you haven't. Without it, coverage is inferred from what other vendors quoted, which can miss items that every vendor forgot.
The detectors require ≥3 vendors quoting a line to compute a peer average. With only 2 bids, front-loading and concentration still fire, but per-line outliers won't. That's a limitation of the statistical approach.
Very rare — only happens when all vendor PDFs are >100 pages each. Workaround: upload only the priced line-item sheet from each vendor, not the full proposal package.
Sources
- FAR 15.404-1: Proposal Analysis TechniquesAcquisition.gov, Federal Acquisition RegulationThe federal price analysis techniques, including the rule on unbalanced pricing.
- Summary: A701-2018, Instructions to BiddersAIA Contract DocumentsHow bidders get documents, ask questions, submit bids and bid security, and how performance and payment bonds are handled.
- MasterFormatConstruction Specifications InstituteThe standard list of numbers and titles used to organize construction specifications, scopes and bid items by division.