How It Works
Nine steps, in the order the job is done. At the steps where a reviewer would ask a question, the answer is on the record.
Find the opportunity
Opportunities arrive from named sources, or you upload the RFP you already hold.
BidAngel reads opportunities, amendments and their documents from SAM.gov and AusTender on a schedule, and keeps each version. If the RFP reached you another way, such as a buyer portal or an email, upload it. It is handled the same way and is visible only to your workspace.
Understand the solicitation
The RFP is shredded into requirements and evaluation criteria, each linked to its place in the source.
BidAngel reads the solicitation documents and lists the mandatory requirements, compliance obligations and evaluation criteria it finds. Each item points back to the passage it came from, so a reviewer can check it against the source.
“Who checked the extraction?” Extracted items are suggestions until a person approves them.
Decide whether to bid
Fit scoring and a recorded go/no-go (bid/no-bid) decision.
Each opportunity is scored against your capabilities, past performance and capacity, and the reasons for the score are shown. The score is an input. A person makes the go/no-go (bid/no-bid) decision.
“Who decided, and why?” The decision is recorded with its reasons and the name of the person who made it.
Plan the response
Compliance matrix, tasks, owners and deadlines.
Deciding to bid opens a pursuit: a compliance matrix built from the approved requirements, and a plan of tasks with owners and deadlines. For each evaluation factor you can record how strong your position is and why, before any drafting starts.
Reuse what you have proved
The content library: past performance, résumés and boilerplate, each with an owner and a freshness state.
Your content library holds what you have already proved: past performance, résumés, certifications and boilerplate. Each item has an owner and a freshness state. Rows in the compliance matrix are mapped to the library items that answer them.
“Is this still true?” Stale and conflicting content is flagged.
Draft
Drafts are written from approved library content only. Where nothing covers a requirement, you see a gap.
Each section is drafted from approved library content, with a citation beside every claim, and written against the buyer’s evaluation criteria as well as the requirements. Where the library does not support a full answer, the draft answers what it can and names the gap. It does not fill the space with unsupported text.
“Did the AI make this up?” Every claim is checked against the evidence it cites.
Review and approve
Reviewers are assigned. Flagged claims must be accepted, edited or removed before a draft can be approved.
Reviewers are assigned to sections. A draft is a suggestion until a proposal lead approves it, and a draft with flagged claims cannot be approved until a person has accepted, edited or removed each one. Any two versions of a section can be compared side by side.
“Who signed this off?” The approval record names the approver and cannot be edited.
Assemble the submission
A versioned package of approved items. You submit it to the buyer’s portal.
Before the package is assembled, a coverage report lists every mandatory requirement and flags any without an approved response. The package is versioned and holds approved items only. You submit it to the buyer’s portal yourself.
Learn from the result
Win/loss reasons and debrief scores feed the next qualification.
Record the outcome, the reasons, and the scores from the buyer’s debrief against each evaluation factor. They are kept as structured records, and they inform how the next opportunity is qualified.
Where a person decides
What AI produces is a suggestion. It becomes part of the bid only when a named person approves it.
- A go/no-go decision is made by a person, not by the fit score
- Extracted requirements and criteria are suggestions until a person approves them
- Library content is approved before a draft may use it, and flagged when it goes stale
- A flagged claim is accepted, edited or removed by a person before the draft is approved
- Only approved items go into the submission package
Where AI agents fit
BidAngel’s own drafting agent works inside the steps above. You can also connect your own: a workspace admin issues a credential with a role, a list of permitted tools and a usage limit. Either way the same three rules apply.
What an agent may read
Opportunities, requirements, the compliance matrix and approved library content, within the role its credential was given.
What an agent may change
It can propose: extract requirements, map evidence, write a draft. What it writes is saved as a suggestion and labelled with the agent that wrote it.
What always needs a person
The go/no-go decision, approving a requirement, approving library content and approving a draft. No agent can approve its own output.