Mark HollandSenior AI Solutions Engineer

RFP Assist

ProductionBuilt at work for a national healthcare data company

Proposal teams answer the same questions again and again. RFP Assist drafts each answer from the approved Loopio library, shows where the wording came from, and leaves every decision to a reviewer.

RFP Assist dashboard: median hours per request, 21.5, split into 0.9 hours of measured review time and 20.6 hours reported, and a running total of hours saved, 304.0.

What it does

  • Drafts RFP responses from the company's Loopio answer library, with reviewer approval on every answer and built-in checks that flag claims against registered figures, catch answers still naming another health plan, and block export while placeholders remain.

From my resume.

Who it is for

The proposal team answering health plan RFPs.

The outcome

From 3 to 4 months to 2 to 3 weeks

Average RFP response time, before and after.

How it works

RFP Assist, step by step

  1. Load

    The client's own questionnaire comes in as it is, and every question keeps its cell.

  2. Draft

    Each question is matched against the Loopio library, and the closest current answers become the draft.

  3. Check

    A claims register flags figures that disagree with registered values, and answers that still name another health plan.

  4. Decide

    A reviewer accepts, edits, or rejects every answer.

  5. Export

    Export stays blocked while placeholders remain, and answers are written back into the client's file.

The real screens

RFP Assist dashboard: median hours per request, 21.5, split into 0.9 hours of measured review time and 20.6 hours reported, and a running total of hours saved, 304.0.

  1. Measured and reported. Review time comes from the event log. The rest is what the team reported at close.

  2. Hours saved. A running total, counted only for requests that have both figures.

The dashboard keeps what the tool measured apart from what the team reported, so no hour is counted twice. Shown with demo data.
RFP Assist review screen: a question about provider reporting levels, with four facts above the draft: closest match 0.74, source age Current, drawn from 1 entry, fact checks Clear. The answer was already reviewed and accepted.

  1. Four facts. Source match, source age, how many entries it drew from, and the fact check, above every draft.

Each draft shows how close its source match is, how old the source is, and whether a fact check flagged anything. A person decides every answer. Shown with demo data.
RFP Assist export screen: one unresolved draft that needs a person, a list of the 21 cells that will be written blank and why, and a box asking the reviewer to type 21 before the export is written.

  1. Blank cells, listed. Every cell that would ship blank is listed with the reason.

  2. Typed confirmation. The reviewer types the count before anything is written.

Export waits for people. Before a file is written, a reviewer types the number of cells that would ship blank. Shown with demo data.
RFP Assist retrieval coverage for 31 questions against a quality floor of 0.60: 30 covered, none legacy only, 1 with no good match, none with no matches, shown as a table and bars.
Coverage is read against a stated quality floor, so the team knows in week one which parts of a file will need a person. Shown with demo data.

The decision that matters

A person confirms every figure

When a draft disagrees with the figures register, RFP Assist flags it where the reviewer reads it instead of quietly changing the number. A person keeps the register, and a figure nobody has confirmed should reach a person before it reaches a client.

Built with

  • Loopio answer library
  • Excel questionnaires in and out
  • Retrieval with ranked sources
  • Figures register