Use casePublic sector RFPs
Four weeks to answer a 142-page RFP. The facts that decide the bid are scattered and must be quoted exactly, so the first days go to reading.
The gist in minutes, each sentence with its pageEvery answer points back to its passageThe reader checks, then decides
Document management services for four hospitals, in two lots, for four years. Price weighs 40%, technical value 50%.
source · Specifications p. 6, p. 7, p. 112Two passages differ · the AI shows both pages
Miss the site visit on page 14 and the whole response can be thrown out. The trouble with this RFP is its length: scope on page 6, penalties on page 87, scoring on page 112. Sophie has to read all of it before she can decide anything.
Sophie asks the open document instead. The assistant answers from the RFP in the viewer, and every sentence cites its page. She checks each answer on the passage it points to, and the go or no-go decision stays hers.
These screens are simple on their own. Their order is the method: the gist first, then the clauses that decide, then every fact checked on its page before it goes into the note. Sophie starts on the Monday the RFP arrived.
The RFP arrived on Monday. Sophie wants the shape of it before reading page by page.
Sophie Lane, bid manager at Norvia Document Services, opens the specifications in the viewer from the bid case. The document opens on page 1. She asks the assistant for the gist: what is bought, how it is split, how long, how it is scored.
Sophie wants to check the weighting before she writes it in her note. She follows the citation to page 112 and reads section 14.2, the award criteria table. The sentence in the summary matches the page.
Sophie asks for every date in the document. The assistant lists six, each with its page, including a site visit that is mandatory for lot 2. She adds them to the bid plan.
The gist is clear. Now Sophie and her technical lead need exact answers, quoted.
Service levels weigh 20% and carry penalties. Sophie asks what happens if the platform is unavailable. The assistant quotes clause 11.4: the availability target, the amount per hour below it, and the exclusion for planned maintenance.
Sophie asks the assistant to check the document for inconsistencies. It finds one: the timetable on page 3 gives December 11 for submission, and section 3.6 on page 15 gives December 10. She adds the point to the written questions for the buyer.
Ahmed Benali, the technical lead, opens Exhibit 4, the security requirements. He asks where patient data must be hosted and who may access it. The answer quotes section 3.2 and section 5.1, with their pages.
Sophie turns what she read into a compliance matrix and a go or no-go note. The choices are hers.
Sophie asks for every requirement written as must or shall in the specifications and Exhibit 4. The assistant returns 23, each with its page. She sets a status on each: 19 met, two with a partner, one open for Ahmed, one her team cannot meet as written.
Sophie drops a note on each page she checked: the weighting on page 112, the penalties on page 87, the date question on page 15. The writers who join next week open the same pages and read her notes.
Each call to the assistant keeps its question, its answer, the model and the time. When the director asks where the penalty figure came from, Sophie opens the exchange and its page.
An answer is checked on its page, flagged because two pages disagree, or marks a requirement the team cannot meet. Each sends the bid a different way.
The answer cites its page, and the reader has read the passage.
What follows: the fact goes into the bid plan or the note.A date on page 3, another on page 15. The AI shows both.
What follows: a written question to the buyer, before the questions close.Found in the list of musts, with its page and its exact wording.
What follows: the bid manager weighs it in the go or no-go note.It decides nothing about the bid. It sets no status in the matrix. Across the nine steps on this page, every fact is checked by a person on its page before it is used.
The AI summarizes, finds dates, answers on clauses and flags what disagrees. It reads only what the user may read, and every exchange stays on record.
Pulled from the Pinecrest County bid. A different RFP and other questions will give other counts.
Where these numbers come from. They are counted on the scenario described on this page, with example data. They are not a customer result.
Bid managers, the experts they lean on, and IT all touch this one.
The first week goes into reading, before any decision.
Asked about one exhibit, you still have to find the page yourself.
Staff paste documents into tools you do not control.
With one long document type. Take the RFPs, contracts or reports your teams read in full every week, and ask them their first three questions.
The assistant works inside the viewer, over the bid cases and documents you keep today.
Ready to try
Bring a long document your team read recently, anonymized, and the questions they had. We will ask them in the viewer and check every answer on its page.