Three things make up the AI side of Proposal Connect: the writer that drafts your proposal section by section, the knowledge base it draws on, and the agents that handle the multi-step work behind gate reviews, capture, and research. They are worth understanding together, because the quality of the first depends almost entirely on the second.
5
Writer steps
10
KB categories
6
Agent families
Gemini
Default model
The AI Proposal Writer
A blank page is not the hard part of a proposal — a blank page that has to be compliant is. The writer works from the solicitation you already analyzed and the material your company has already written.
Five gated steps — each one unlocks the next
-
1
Identify Template
Structure from the RFP
-
2
Identify Documents
Map past performance
-
3
Draft Content
Section by section
-
4
Validate Compliance
Check against requirements
-
5
Export
Word, PDF, SharePoint
Identify Template
The AI reads the solicitation and proposes a section structure, following Section L and M where the RFP defines one. Alternatively, start from a template you saved on a previous bid. Reorder, rename, add, and remove — then finalize.
Identify Documents
The references analyzer suggests which past performance and knowledge base documents fit which sections, and you confirm or change the mapping. This is the step that decides whether your draft cites real projects or writes in generalities.
Draft Content
Generate one section at a time, or let it run autonomously through the whole structure. Each section is written against its own instructions, its assigned references, and the requirements from the analysis. Accept, reject, or edit each result.
Validate Compliance
The quality agent reviews compliance, completeness, and quality against the RFP requirements and revises sections that do not hold up. Findings appear in the chat panel so you can see what it changed and why.
Export Proposal
Produce the final Word or PDF document, save the structure as a template, or push it to SharePoint.
Steering the Output
Four controls decide how close the first draft lands to what you would have written yourself.
Writing instructions
Per-section guidance on key points, tone, and requirements to hit. The single highest-leverage input — three specific sentences beat a page of vague direction.
Proposal themes
Reusable win themes, discriminators, and past performance emphasis applied across the proposal so every section argues the same case.
Reference documents
Section-level attachments from your knowledge base. Specific past performance produces specific claims; nothing attached produces generic ones.
Model selection
Gemini by default, with a model selector on agent runs when you want to trade speed against depth.
Changing your mind
Any step can be reset from the workflow timeline, and resetting one clears everything downstream of it. Reset the structure and references, drafting, compliance, and export all clear with it. That is deliberate: it is the only way to guarantee the draft in front of you was written against the outline currently on screen.
What the AI does not do
It does not submit for you, it does not decide your price, and it does not verify that a past performance claim is true. Treat every draft as a first draft from a fast, well-read colleague who has never met your customer. The review pass is still yours.
The Knowledge Base
Every AI feature is only as good as what it knows about your company. This is where that lives — and the context the AI writes from.
| Category | What goes in it |
|---|---|
| Past Performance | Previous project successes and deliverables |
| Corporate Capabilities | Company overview, org charts, capability statements |
| Resume | Team member resumes and professional profiles |
| Certification | Certifications, licenses, and accreditations |
| Teaming Partner | Partnership and joint venture agreements |
| Best in Class Content | Reusable management approach, QA, and staffing plan content |
| Solution Documents | Solutions you have proposed before |
| Diagram | Architecture diagrams, flowcharts, network drawings |
| Company Info | Core company details |
| Other Documents | Case studies, white papers, client references, security and financial documents |
Keeping it organized
Categories are the top level; tags and file groups handle everything below. Tags cut across categories — an agency name, a capability, a contract vehicle. Groups bundle related files so a whole set attaches to a proposal section at once.
Feeding the AI
Categories can be synced, which indexes their contents for AI use: capture fit scoring, the references step of proposal drafting, and knowledge base chat. Sync is per category, so you can keep something for your team without putting it in front of the AI.
Search and chat
Beyond browsing, a chat interface answers questions across your documents and cites what it used. Conversations can be renamed, bookmarked, and pinned, so a useful thread stays reachable instead of getting lost.
What to upload first
Two or three past performance write-ups, your capability statement, and the resumes of your key personnel. That is enough for the AI to stop writing in generalities. PDF for the document categories; Diagram also accepts image files.
Good practice
- 1 Add the win while it is fresh. Write up past performance in the week after award, not the week before the next bid is due.
- 2 Keep resumes current. Key personnel sections are only as strong as the resume the AI has to work from.
- 3 Promote what worked. When a section wins, save it to Best in Class Content so the next proposal starts from it.
AI Agents
Some proposal work is a single question. Most of it is not. Reading a 200-page solicitation, cross-checking it against your past performance, and producing a gate review recommendation takes planning, several passes, and access to different sources.
How an agent works a task
-
1
Gather context
Reads the task and its data
-
2
Plan
Writes the steps it will take
-
3
Call tools
Fetches documents and facts
-
4
Replan
Revises what is left, then reports
Everything is visible
Runs stream to the screen as they happen: the plan, the current step, each tool call with its arguments and result, and the reasoning between them. A recommendation you cannot trace is one you cannot defend in a gate review.
Threads persist
Conversations are saved, so you can leave a long run going, come back to it, and reopen it weeks later to see how a conclusion was reached.
The Agents
Each agent is scoped to a job, with the tools and prompting that job needs. Availability depends on the modules enabled for your account.
Gate Review Analyzer
Evaluates a solicitation against a gate review decision using compliance data, requirements, and bid strategy. A deep variant delegates each section to a focused subagent for long solicitations.
Opportunity Analysis and Explorer
Analyzes a single opportunity from its documents and research, or searches and compares opportunities across your whole account.
Proposal agents
Structure builder, section writer, references analyzer, and quality reviewer — the agents behind the guided AI writer, also usable directly from a proposal.
Capture agents
Fit analyzer, identify agent, and deep analysis agent — the qualification stages behind the capture pipeline.
Knowledge Base agent
Answers questions across your document library and cites the sources it used.
Utility agents
Diagram generator, form auto-filler and parser, vendor analyzer, resume scoring, and solicitation info extractor.
Starting a run
Some agents take a normal chat prompt. Others take structured input — pick the solicitation, pick the gate review decision, optionally pick the model — so a run cannot start missing the context it needs. Where a model selector is offered you can trade speed against depth per run; otherwise each agent uses a sensible default.
Getting good results
- 1 Give it the documents. An agent reading three attachments beats one reading a single merged PDF with no structure.
- 2 Ask for the decision you need. Ask which of these three pursuits has the best fit and why, not tell me about these opportunities.
- 3 Read the tool calls when something looks off. Nine times out of ten the agent was working from a document you did not realize was missing.
- 4 Keep the knowledge base current. Agents reason about your company from what is in it.
What to do next
- 1 Set up your workspace — the Getting Started guide covers signup, setup, and your first proposal end to end.
- 2 See the rest of the platform — the Core Features overview lists every module.
- 3 Add your own Gemini API key under Settings, so AI features run on your own quota.