Skip to main content
For AI agents: a documentation index is available at https://docs.coverbase.com/llms.txt. This page is also available in markdown by appending .md to the URL.
This guide is part of the User Guides collection. It follows one sourcing event from a sentence of intent to an awarded vendor with a decision memo. For what RFP Platform is and where it fits, see RFP Platform. For what happens after the award, see How to run an assessment.

​
What’s in this guide

Each part stands on its own, so you can jump to the stage you need.

Part 1: Draft the RFP

Eight wizard steps, from one sentence to a complete RFP.

Part 2: Find your way around

Seven sections, in the order you work through them.

Part 3: Run it without outreach

Compare the field before contacting anyone.

Part 4: Score the responses

Rubric scoring, evidence, and must-have failures.

Part 5: Compare and decide

Cross-vendor comparison, pricing, and the decision memo.

Part 6: Onboard the winner

Hand the vendor to intake without re-keying anything.

​
Before you start

You will move faster with these to hand:
  • A sentence about what you need. Not a specification. “We need to replace our applicant tracking system for about 12 recruiters, integrated with our HRIS and Microsoft 365” is enough for the wizard to work from.
  • Your hard constraints. The things a vendor cannot talk you out of: an access scope you will not grant, a certification you require, a jurisdiction your data cannot leave. These become must-have requirements.
  • A rough timeline. When responses are due and when you need to decide. The wizard proposes dates from those two, and you can move any of them.
  • Permission to create RFPs and vendors. If a button described here is missing, check your role with an administrator.

​
Two terms worth pinning down

Rubric vs questionnaire

The rubric is what you are evaluating: categories, the criteria under each, their weights, and the ideal answer for each criterion. It is how you decide.The questionnaire is what you ask vendors. Each question is tied back to a rubric criterion, so an answer arrives already attached to the thing it is evidence for. The wizard builds both, and links them.
A silent RFP researches how each shortlisted vendor would most plausibly answer your questionnaire, from public information about their product. Nobody is contacted. It exists so you can narrow a field of ten to a field of three before you ask anyone to spend a week writing a response.Its output is Coverbase’s research, never a vendor’s own words, and every screen that shows it says so.

​
Part 1: Draft the RFP

From RFPs, click New RFP. You are asked how you want to build it.
Choosing between the AI wizard and manual creation

Start with the AI wizard, or fill the form in yourself.

Build it myself opens a short form: a name, a category, and dates. Everything else you add later from the RFP page. Use it when you are recreating an RFP you already have. Draft it with AI is the eight-step wizard below. Nothing it produces is final; every step is editable, and nothing is written to your organization until the last one.

​
Step 1: What you need

Describe what you are buying in a few sentences, in your own words. Include the constraints, because they carry through to the requirements.
The brief step of the RFP wizard

A few sentences is enough. The example chips fill the box if you want to see the shape of a good brief.

​
Step 2: Clarify

The wizard reads your brief and asks the handful of questions that would change the RFP: how many users, which systems it must integrate with, what data it will touch, what your budget shape is.
Generated intake questions

Questions are generated from your brief, so they differ every time.

Answer what you know. Anything optional you skip is treated as unstated rather than guessed at. A few questions are marked required, because the rest of the draft leans on them; leaving one blank keeps Continue disabled and scrolls you to the one that needs an answer.

​
Step 3: Description

The wizard drafts the RFP’s name, its short description, and the longer narrative that vendors read.
The drafted description and proposal narrative

Edit any of it directly, or describe the change you want and let the wizard redraft.

The Refine box is the faster path when something is off in tone or emphasis. “Make the security section firmer, and say we will not grant tenant-wide mailbox access” produces a better redraft than editing the paragraph by hand.

​
Step 4: Timeline

Dates are proposed from your decision deadline and the size of the questionnaire, with the reasoning shown. Move any of them.
The proposed timeline and lifecycle stages

The lifecycle stages you select here are the ones this RFP will move through.

Three settings live here:

​
Step 5: Vendors

The wizard researches the market and proposes five to ten vendors, each with what it is known for, why it fits your brief, what to watch out for, and a contact address where one is public.
Researched vendor suggestions

Each suggestion carries its reasoning, so you can disagree with it on the evidence.

Select the ones you want to invite, or take the whole set with Add all and pare it back from there. You are not committing to a shortlist size here. More can be added from the RFP’s Vendors section at any time.
Vendor research is a starting point, not a market survey. It reflects what is publicly documented, which favors vendors who publish. Add the ones you already know about.

​
Step 6: Requirements

The wizard proposes the rubric: weighted categories, the criteria under each, and the ideal answer for every criterion.
The proposed rubric with weighted categories

Weights within a category total 100, and the categories themselves total 100.

Take time on this step. The rubric is what every later screen scores against, so a criterion that does not matter to you will pull the ranking around.
Mark the criteria you will not compromise on as must-have. A response that does not clearly clear a must-have is flagged as failing it, and ambiguity counts as a failure rather than a pass.

​
Step 7: Questions

The wizard writes the vendor-facing questionnaire, spread across the rubric so every criterion has evidence behind it. Choose how many questions you want.
The generated vendor questionnaire

Each question names the criterion it is evidence for.

​
Step 8: Review

Everything the wizard is about to create, on one screen.
The wizard review step

Nothing has been written to your organization until you confirm here.

Confirming creates the RFP, its rubric, its questionnaire, the vendor records, and the invitations, all as one transaction. It lands as a draft. No vendor is contacted until you publish it.

​
Part 2: Find your way around

The RFP overview section

Seven sections, in the order you work through them.

Everything on the Details card is editable at any stage, draft or not. Category and scoring scale offer their alternatives directly; the owner is a person picker; feature toggles flip in place.
The RFP vendors section

Vendors, how each was sourced, and where their portal link goes.

​
Part 3: Run it without outreach

A silent RFP researches how each invited vendor would most plausibly answer your questionnaire, so you can compare the field before asking anyone for real work.
The submissions section with the silent RFP control

Run it from the Submissions section. Researched responses sit alongside real ones and are always labeled.

Three things to know:

These are not vendor responses

They are Coverbase’s research, modeled from public information about each product. Nobody was contacted, and no vendor has seen or agreed to any of it. Do not present a researched response to a vendor, an auditor, or a committee as something the vendor said.
Vendors who have already submitted are skipped. Replace real submissions exists and is off by default; even when on, it only ever adds a researched estimate alongside the real answer. Re-running a silent RFP never deletes a vendor’s own words.
Each answer carries the model’s confidence that it reflects the vendor’s real position, plus the basis it used. A fluent answer with low confidence is a plausible guess.

​
Part 4: Score the responses

Open a submission and click Score with AI. Each criterion in the rubric gets a suggested score, the reasoning behind it, and the verbatim span of the response the score rests on. The suggestion sits next to the reviewer’s own score and never replaces it. Re-running the scorer leaves anything a person entered exactly as it was. Three rules apply to the suggestion:
  • A score must cite the response. A criterion the response never addresses scores 0 with no quote, rather than an inferred number.
  • A must-have fails on ambiguity. Must-have criteria pass only when the response clearly clears the bar. Failures are listed above the scoring table.
  • A renamed criterion is surfaced, not dropped. Scores returned for criteria that are no longer on the rubric are reported rather than silently discarded.
Scoring also reads the commercial terms out of the response, which is where the pricing table in the next part comes from.

​
Scoring a demo, together

Each requirement is its own card, and everyone evaluating the submission writes on it. Your assessment sits at the top; below it are your colleagues’ notes, each under their name. This is built for the demo itself:
  • The score is optional. Pick No score and write only what you saw. An observation without a number reaches the team and the re-evaluation.
  • The score and the note save separately. Type a note twenty minutes after entering a score and the score stays where it was.
  • Turn on “Saw this first-hand” for anything you watched in the demo or heard on a reference call rather than read in the response.
When the session is over, click Re-evaluate with team input. Coverbase re-reads the response together with every note the team recorded. A note marked as seen first-hand is treated as evidence the written response does not contain: it can move a score, the rationale says whose observation moved it, and where the observation contradicts the response, the observation wins. Re-evaluating never touches anyone’s own score or notes. It rewrites the suggestions only.
Different stakeholder groups see different things in the same demo. Security watches the integration scopes, the hiring team watches the workflow. Both writing on the same rubric row turns a demo into recorded evidence.
The RFP requirements section

The rubric, with weights and ideal answers, is what every score is measured against.

​
Part 5: Compare and decide

​
Cross-vendor comparison

The comparison reads every submission together. What separates two bidders only shows in the contrast between them.
The comparison section

Stored with its timestamp rather than regenerated on view, so the version somebody circulated stays the version they see.

When any submission in the set is researched rather than real, the summary says so and weighs it lower. Ranks are written back onto the submissions, and the leaderboard shows each vendor’s position against where it sat before, so a re-run makes its own effect visible. You are not bound by the ranking it produces. Drag any row by its handle to put the field in the order you believe is right, and the board renumbers from the top. A manual reorder is recorded in Activity, so the override is visible next to the comparison it replaced.

​
Pricing and total cost

Year-one cost and three-year total cost sit side by side, each with a badge for how firm the number is: A response that never stated a figure shows blank rather than zero, because “contact us” is not free. Watch-outs such as uncapped uplift, implementation billed separately and seat minimums are read out of the response and shown next to the numbers.

​
Q&A

Vendor questions and the answers you publish back to all of them. Draft with AI writes an answer from what the RFP establishes, for you to edit before publishing.
The Q&A section

A draft never mentions other bidders, and says so plainly when the RFP does not settle the question.

If the RFP does not answer the question, the draft says so and suggests what the RFP should add.

​
Decision memo

After you award, Write the memo produces the record of why: the recommendation, the reasoning, each runner-up with the specific reason it lost, the risks you are knowingly accepting, and the conditions the winner must meet before signature. Where a researched response stood in for a real one, the memo says so. Somebody reading it in a year needs to know which part of the decision rested on research.
The RFP activity feed

Every event on the RFP, including which AI steps ran and who ran them.

​
Part 6: Onboard the winner

From the Vendors section, Onboard opens the vendor form prefilled with everything the RFP already established: the research that shortlisted them, the invitation contact, the RFP’s description as the use case, and anything flagged during sourcing.
Onboarding a vendor from an RFP

The vendor record already exists, created by the RFP when the vendor was invited, so this updates it rather than creating a duplicate.

From there the vendor enters the normal program: Autonomous Intake for due diligence, then Assessment Copilot, reusing the responses and evidence the RFP already collected.

​
Running an RFP from chat

Everything above is available through the MCP server, so an RFP can be drafted and compared in a conversation. Reads go through query_rfps; writes go through mutate(entity="rfp", ...) and ask you to confirm before anything is created. A typical exchange:
  1. “Draft an RFP to replace our applicant tracking system for 12 recruiters, integrated with our HRIS and Microsoft 365. No tenant-wide mailbox access.”
  2. “Research vendors for it, and suggest requirements.”
  3. “Create it.” You confirm, and the RFP, rubric, questionnaire and invitations are created together.
  4. “Run it silently and compare them.” Researched responses, then the cross-vendor comparison.
The same rules apply: researched responses are labeled as research everywhere they appear, and every write asks for confirmation first.

​
Common questions

Yes. Every field on the Details card, the timeline, and the rubric stay editable after publication. Rubric changes are versioned, so you can see what the rubric looked like when a given submission was scored.
New suggestions are added to the ones you already have. Vendors you have already invited are matched by name rather than duplicated, and re-inviting a vendor who declined puts them back in the running.
No. Build it myself skips the wizard, and every AI action on the RFP page is something you click rather than something that happens. The rubric, the questionnaire and the scores can all be entered by hand.
Anyone with access to the RFP, and it is labeled as research everywhere it appears. Researched responses are never sent to vendors and never appear in a vendor’s portal.
Scoring is its own permission. A role with RFP: read plus RFP evaluation: create and update can score submissions and write demo notes, but cannot edit the RFP, change its vendors, or award it. Siloed Members have exactly that by default; for Auxiliary Members or Guests, an admin grants RFP evaluation through a custom role.
Yes, that is the intent. Each person’s score and notes are stored separately against the requirement and shown under their name. Nobody overwrites anybody.

RFP Platform

What RFP Platform is and where it fits.

How to run an assessment

What happens after the award.

Requesting a vendor

The intake path the winner is handed to.