Skip to content
Novyant
Platforms

Which contracts to bid on, and what to bid.

Federal portals publish thousands of notices a day. This reads them overnight, ranks them against rules drawn from how your business wins, and recommends a price you can open up and change.

In productionRunning today with its first operator.
Book a consultation
contracts · inbox
Last run05:31214 notices read2 to quote
Who it is for

Built for the supplier doing the bidding.

  • 01

    Small and mid-size federal suppliers

    Firms with a CAGE code and a narrow catalogue, facing more notices a day than anyone has time to read.

  • 02

    Whoever decides what to quote

    Usually an owner or an estimator with an hour in the morning, who gets a ranked list with the reasoning and a price attached.

  • 03

    Teams that have to justify the call

    Every bid and every pass keeps the rules and evidence behind it, so the decision is still reviewable a quarter later.

What it does

Read the notice. Read the market. Then rank.

Finding solicitations was never the difficulty. Knowing which are winnable, who you face, and what to charge is the work.

  • 01

    Rules written from your own record

    Setup reads your catalogue, bids and awards, then proposes a weighted rule for each pattern: the codes you win in, the agencies that buy from you, the terms that have cost you money. You confirm or discard each one.

  • 02

    A score you can take apart

    Every notice starts from a common base and moves as rules fire, so a score is always a list of reasons you can read back.

  • 03

    The document read properly

    Item, quantity, agency, deadline, value, delivery terms and the certifications buried in an attachment nobody opens, each with a confidence figure and its source passage.

  • 04

    The field you are bidding into

    Who has won this line item before, how often they win when they show up, and how many are likely to bid this time.

  • 05

    A price with a floor and a ceiling

    A recommended number built from past awards on the item, how rivals price it, your own cost and margin, and the terms on this notice.

  • 06

    Rules that keep improving

    What you bid, pass, win and lose all feed back, and the system proposes the rule change that would have caught the difference.

How the ranking gets built

The rules come from your record. Then they are yours.

Onboarding turns your catalogue, bid history and awards into weighted rules. From there the arithmetic is fixed and visible: switch one off below and watch the number move.

  1. 01

    We read your record

    Your catalogue, what you bid, what you won and at what price. Where your history is thin we start from public award data.

  2. 02

    The model proposes rules

    Each pattern becomes a rule with a name, a condition and a weight. This is the one place a model comes near the ranking, and everything it produces is a proposal.

  3. 03

    You keep the pen

    Accept, reweight or delete. Every edit is logged, and re-scoring the day after a change reads no documents and costs nothing.

SPE7MX-26-T-0418

Pump, Centrifugal, Water · DLA Land and Maritime

50

Composite

Quote

Quote at 90 · Not bid below 51

Switch a rule off and the composite moves. This is the entire ranking mechanism.

  • Base50
  • Classification code you win in+18
  • You have won this line item before+22
  • Agency buys from you regularly+10
  • Named manufacturer you distribute+8
  • First-article testing required-12
  • One competitor takes most awardsSuggested-20
Where the line is

The model reads. The arithmetic decides.

A number a model produced cannot be argued with, tuned, or explained to the person who has to sign the bid. Ask anyone selling scored opportunities where their line sits.

Model

Reads, extracts and researches

Solicitation documents are unstructured, inconsistent between agencies and sometimes scanned. Award records have to be matched to line items and prices normalised across quantities. All of that is reading, which is what it is good at.

  • Every extracted field carries a confidence figure
  • The passage each value came from is kept
  • Below the confidence floor, nothing is cleared to quote
Arithmetic

Ranks, and only it ranks

Weighted rules over extracted facts. The same notice gives the same score every time, and when you disagree you can find the rule that caused it and change it.

  • Every score itemised into the rules that produced it
  • Rule changes logged against who made them
  • Re-score a whole day without re-reading a document

The model proposes rules and it reads documents. It never assigns a score, and it never moves a weight on its own.

The field

How hard this one will be, before you price it.

A crowded line item and an open one look identical on the portal. The difference is in the award record.

Pump, Centrifugal, Water

DLA Land and Maritime

Competition

Heavy

Recent awards on this line item

  1. A
  2. C
  3. A
  4. B
  5. A
  6. D
  7. ·
  8. A

Who turns up

wins when they bid

  • Incumbent A62%
  • National B48%
  • Regional C21%
  • Regional D18%
  • Specialist E9%
7
Likely bidders on this notice
48%
Of recent awards taken by one supplier
2
Bidders that usually win when they bid
1 of 4
Your record on this line item

Competitor names are anonymised here. In the system they are the CAGE codes and company names from the public award record.

The number

A price with a floor under it and a reason above it.

The recommendation is a range built from inputs you can open. Where the award record is thin the range widens and the page says so.

Recommended

per unit, 240 units

$41.20

Your floorWhere you stop winning
Past awardsWhere this item clearsRecommended

What it is built from

  1. 01What the item has awarded at before, and how wide the spread is
  2. 02How rivals price it, and whether they shave margin to win
  3. 03Your landed cost, target margin and walk-away number
  4. 04Quantity breaks and the delivery window on this notice

Nothing is submitted for you. The recommendation is where the quote starts, and the quote is still yours to write.

What it learns

Every decision you make is training data you own.

The useful signal is what you did next, and what happened after that.

  1. 01

    You bid, or you pass

    Both are recorded against the score and the rules behind it. A pass is as informative as a bid, and most tools ignore it.

  2. 02

    The award lands

    Won, lost, or no award at all. Where the price is public, the gap between your number and the winning one is recorded with it.

  3. 03

    It looks for the difference

    What separated the notices you won from the ones you lost, and the passes you regret from the ones you do not.

  4. 04

    It proposes a change

    A new rule, a reweighting or a threshold move, with the cases behind it attached.

Suggested

Raise the first-article testing penalty to 20

You passed on six of the seven notices carrying it last quarter and lost the one you bid. A weight of 12 is not describing your business.

Nothing reweights itself. A ranking that changes on its own is one nobody can defend a quarter later.

The morning run

Done by six, before anyone logs in.

The sequence runs on a schedule, retries itself twice, and alerts a person if it is not done by six.

  1. 05:00

    Fetch

    SAM.gov and DIBBS are queried for the day's notices, filtered against the codes you sell into, and the attachments worth reading come down with them.

  2. 05:05

    Import

    Notices land keyed on solicitation number. Re-running an import never duplicates a row, which makes a failed run safe to run again.

  3. 05:25

    Read

    Anything not already ruled out goes to the model, which returns the fields that matter with a confidence score and the passage each one came from.

  4. 05:35

    Research

    Award history is pulled per line item, competitors resolved to their codes, past prices normalised to this quantity. No model involved, and a failed lookup shows no history at all.

  5. 05:45

    Rank and price

    Every rule is evaluated, the composite lands between zero and one hundred, the thresholds sort it, and the pricing model produces a range against your cost floor.

  6. 06:00

    Alert

    A run that has not finished cleanly by six raises an alert to a person. The one failure mode worth engineering against is a quiet morning that looks like a slow day.

Questions

What suppliers ask.

Platforms