Bid and tender software, supplier and buyer

Qualify. Draft. Prove it.

Every figure traced to evidence you already hold.

The engine cannot clear its own gate. A named person approves.

UK public procurement

The engine that refuses to make things up.

The hardest market to bluff in is the one with a published rulebook, so it is the one the engine is proved against. CrowAgent reads both UK registers, drafts answers cited to the published requirement, to your own past bids and to commitments you have confirmed, and evidences delivery after award.

The same engine runs a private-sector bid. What changes is the rulebook it cites, not the standard of proof.

Refusal gate

A figure with nothing behind it does not survive the draft.

Every pound and every percentage in a CrowAgent answer is computed from a commitment you have confirmed. If the commitment is not there, the claim is taken out before a named human reviews the answer.

Refused

Local supply chain spend. No recorded commitment sits behind the figure.

Statutory figures

Three numbers you are already held to

  1. s.52 £5m contract value threshold
  2. s.71 12 months between published assessments
  3. PPN 002 10% minimum social value weighting

Every figure, sourced

How it runs

Five stages, and one of them is a refusal.

The gate is not a filter bolted on at the end. It sits between the draft and the human who signs it, and it is the only stage whose job is to remove work rather than produce it.

  1. Stage 01

    Find

    Find a Tender and Contracts Finder, read continuously and scored against the evidence you hold.

  2. Stage 02

    Qualify

    Fit, component by component, with the reason for each. Stated as context, never as a probability of award.

  3. Stage 03

    Draft

    Answers grounded in your answer library, the published requirement and your confirmed commitments.

  4. Stage 04

    Refuse

    Any figure without a recorded commitment behind it is taken out of the draft before review.

  5. Stage 05

    Evidence

    After award, each commitment is tracked to the dated evidence that proves it, month by month.

Workstation tour

The rest of the workstation.

Six screens that appear nowhere else on this page: what a buyer publishes, what a supplier answers from, how a commitment is evidenced, how a response is banded, what the audit trail records, and the duties that get carried on a phone.

The CrowMark social value requirement builder for a buying authority. Four figures head the page: the social value weighting set against the PPN 002 minimum, the number of criteria published, the total weight split across quality, price and social value, and the responses received. The award criteria table lists each criterion with its TOMs reference, its weighting and the evidence a supplier must supply.

Buyer · Requirement builder

Criteria, weightings and the evidence each one demands, set against the PPN 002 minimum

The product

One tender, from notice to evidence.

Five screens along the same contract: finding the notice, deciding whether to bid, drafting the answers, evaluating the responses, and proving delivery afterwards. The wording each screen carries is quoted beside it, because the wording is the argument.

app.crowagent.ai/mark/discover
The CrowMark Discover feed. Published notices are listed with the buying authority, the contract value, the sector, the publication date and the register they came from, each carrying a relevance percentage. A filter row offers the two registers, sector, value and relevance, and every notice offers Open in CrowMark, a link out to the source register, and Save.

Notices from Find a Tender and Contracts Finder, scored against what you can evidence today

01 · Supplier · Discover

Notices you could actually win the argument on.

Both UK registers are read: Find a Tender and Contracts Finder. Relevance is scored against the evidence you already hold rather than against the words in the notice, and the register a notice came from is named on the notice, so the trail out to the source is one click and not a search.

Relevance is a sorting aid. It is not a forecast of anything.

app.crowagent.ai/mark/qualify
The CrowMark bid or no-bid screen. A fit score sits beside the contract value, the days to deadline and the number of recorded prior contracts, and a data confidence panel states: “This is FIT context. It is not a probability of award, and CrowMark does not produce one.” Below, the score is broken down component by component, each with a band, a points total and a one-line reason.

Fit broken down component by component, stated as context and never as a probability of award

02 · Supplier · Bid or no bid

Fit, broken into its parts, with the reason attached.

The score is decomposed component by component, each with a band, a points total and a one-line reason you can argue with. What it is not is stated on the screen itself, in the product, where a reader cannot miss it.

“This is FIT context. It is not a probability of award, and CrowMark does not produce one.”

The data confidence panel, on the screen above

app.crowagent.ai/mark/draft
The CrowMark tender questions screen. A tender is broken into its scored questions, each with a category, a word limit, a weighting and a draft action. A banner states the grounding boundary: answers are grounded only in your own answer library, the published requirement and your confirmed commitments; every figure comes from computed commitments and is never invented; and a named human approves every answer before it is submitted.

Award questions with their weightings, above a drafter told where every figure has to come from

03 · Supplier · Draft

The drafter is told where a figure may come from.

The tender is split into its scored questions, each with its category, word limit and weighting. The boundary the drafter works inside runs across the top of the screen and is not a setting anyone can turn off.

“Answers are AI-assisted and grounded only in your own answer library, the published requirement and your confirmed commitments. Every £ and % comes from your engine-computed commitments and is never invented. A named human reviews, edits and approves every answer before you submit it.”

The grounding banner, on the screen above

app.crowagent.ai/mark/evaluate
The CrowMark response review screen, one criterion against one response. A banner states: CrowMark locates the passage that answers this criterion and shows you where it came from; it does not score, rank or recommend; the band is yours to set. A panel quotes the supplier verbatim and cites it to a section and page of the response document. On the right, four bands are offered under the note that there are no half marks and no rounding, with an audit line naming the evaluator.

The passage that answers the criterion, located and cited, with the band left to a named evaluator

04 · Buyer · Evaluate

It finds the passage. It does not set the band.

One criterion against one response. The supplier is quoted verbatim and the quote is cited to a section and a page of the response document, so an evaluator checks a reference rather than trusting a summary. The band, and the audit line under it, belong to a named person.

“CrowMark locates the passage that answers this criterion and shows you where it came from. It does not score, rank or recommend. The band is yours to set.”

The evaluation banner, on the screen above

app.crowagent.ai/mark/delivery
The CrowMark delivery oversight screen, tracking commitments made at award through to the evidence that proves them. Four figures head the page, covering contracts in delivery, measures committed, evidence received this month and section 71 assessments due. A table gives each supplier an evidence percentage, a status of on track, at risk or behind, and its next section 71 date.

What each supplier committed at award, against what they have since evidenced

05 · Buyer · After award

A commitment is only worth the evidence filed against it.

Promises made at award are carried forward as measures, and each measure collects dated evidence against it month by month. The same view carries the next assessment date for each contract, which is the part of the duty that is easiest to lose track of and the hardest to explain later.

Each contract reads on track, at risk or behind against the evidence received, not against a projection. Suppliers see the other half of this: the same commitments, with the upload that proves each one.

Both sides of the table

Read from either end.

The supplier writing the answer and the authority evaluating it are working on the same passage. It should be the same passage, cited the same way, on both screens.

Supplier reads left to right, from above the spine

  1. Requirement

    Supplier: Which measures apply

    Authority: Publish the rubric

  2. Evaluation

    Supplier: Answer from your library

    Authority: Located word for word, or dropped

  3. After award

    Supplier: Deliver what was priced

    Authority: Audit the commitment

Authority reads right to left, from below the spine

For suppliers For buyers

Run the real engine

Watch it refuse a figure.

If a figure cannot be traced, the engine refuses it. A generative tool asked for a number it does not have will produce a plausible one. CrowAgent does the opposite, and it does it visibly.

Worked example

Quantify the community and environmental benefit this contract will create.

A private-sector RFP. The same run answers a public notice.

  1. Step 01: Retrieve

    The published question is read with its wording intact.

  2. Step 02: Ground

    Every figure is matched to a confirmed commitment. One is not, and is dropped here.

    local supply chain spend no recorded commitment dropped here

  3. Step 03: Compute

    What survives is computed from those commitments, never generated.

  4. Step 04: Check

    The draft is counted against what it claimed: four asked for, three traced.

  5. Step 05: Cite

    Each surviving figure carries its source, and a named person approves it.

Claimed
4
Traced
3
Refused
1

4 jobs × £27,000 per year = £108,000 illustrative

The untraced claim is removed before a named approver reads the draft. A person signs off, not the software.

How the drafter works Sources

Plugged in, read only

Your evidence already exists. We read it where it lives.

  • SharePoint Sites.Read.All
  • OneDrive for Business Files.Read.All
  • Confluence Cloud Your own permissions
  • Microsoft Entra ID SAML 2.0, sign-in only
  • Okta SAML 2.0, sign-in only
  • Google Workspace SAML 2.0, sign-in only

Every source above is read only. We do not write to your files, change a setting in your tenant, or create an account.

Slack, Teams and Google Drive are separate outbound connectors, used to send a notification or export a file you have asked for. They are not sources we read from.

How we handle your data

Ready when you are

See it on your own tender.

Bring a live notice and we will run it with you. Access is limited and by request.

Read a tender, free

Access is by request. There is no free plan, and no self-serve signup; a 14-day evaluation trial is available on request.