From five messy department docs to a polished RFP, before lunch

This is the problem the FreeITSM RFP Builder exists to solve. Eight years of wrestling with procurement spreadsheets distilled into one AI-driven workflow inside the open-source FreeITSM platform. By the end of the afternoon — not the end of the week — you should have a polished, deduplicated, conflict-aware RFP document, plus the infrastructure to score every supplier reply when they come in.

Here's how it works, end to end.

Phase 1: Drop the docs in

You create a new RFP inside the Contracts module and upload each .docx in turn, tagging it with the contributing department. The colour-coded department list is configurable, and those colours flow through every later step — so you'll always know which voice each requirement came from.

You can re-extract or delete any document at any point. There's a raw-text preview if you suspect the parser dropped something weird. You're at the foundation, not yet doing any AI work. This bit takes about three minutes.

Phase 2: Watch the AI read every word

Click Extract requirements on any document. The AI reads it and produces a structured list — every requirement, every pain point ("we currently struggle with…"), every challenge ("we're worried about migration because…") — each tagged with the original verbatim quote.

This matters more than it sounds. Your stakeholders will eventually ask "where did that requirement come from?" and you'll click straight to the source quote in the IT director's email from three weeks ago. Political cover is built in.

One document might produce fifteen items. The next, thirty. By the time you've extracted from all five, you'll typically have between fifty and a hundred and fifty raw items scattered across departments. They overlap. They conflict. They use different words for the same thing. This is the chaos that the next phase untangles.

Phase 3: The magic moment — consolidation

This is the step that separates the RFP Builder from a glorified document concatenator.

You click Run consolidation, and a modal opens showing the AI working live. Within two seconds, the first category appears — "Identity and access management". Then "Reporting and dashboards". Then "Mobile experience". The AI is proposing a structure for your specific RFP, based on what your departments actually asked for. Not a generic template. Not a pre-baked list.

A progress tracker shows three pulsing tasks: Categorising, Consolidating requirements, Detecting conflicts. Each ticks over to green as the AI finishes that part. You watch the tokens stream past. After roughly two minutes, you have:

A week of manual deduplication, replaced by one AI call streaming live to your screen.

Each consolidated requirement keeps its source attribution. Click "Show sources" on any row and you'll see the exact extracts from each contributing department, with their original verbatim quotes. Five different departments saying "we need single sign-on" become one consolidated requirement called Single sign-on integration with the existing identity provider, with five sources cited and the priority calibrated against how many people asked and how strongly.

Your judgment trumps the AI's

The AI is good. It is not perfect. The RFP Builder gives you full editing tools to override any decision it made:

• Edit any consolidated row's text, priority, category, or rationale.
• Split a row the AI over-merged — pick which sources go to each new row.
• Merge rows the AI failed to dedupe, by ticking checkboxes.
• Add custom for things the AI missed entirely.

For each conflict the AI flagged, you choose how to resolve it: pick Side A, pick Side B, merge them into one position, or dismiss as not actually a conflict. Your reasoning is captured to the audit trail.

When you're happy, you click Lock for generation. The consolidated set is now frozen — no more drift, no more accidental edits underneath the document layer. Phase 4 unlocks.

Phase 4: The document writes itself

This is where Anthropic's prompt caching pays off.

You click Generate all. The AI starts at the top: an introduction grounded in the optional context you supplied ("we're replacing our legacy ITSM tool, end of life next quarter, hybrid workforce…"). Then a scope statement. Then response instructions, complete with placeholders for the dates you'll fill in. Then it works through the 14 category sections one by one — overview paragraph, requirements as a polished bullet list, current pain points, challenges and concerns, department-specific notes.

Each call is streamed live to your browser — you see the section taking shape as the model writes it. The pattern matters: prompt caching means later sections in a batch reuse the cached system prompt, so a full generation pass over fourteen categories is significantly cheaper than fourteen independent calls. The AI activity panel keeps a running tally of token usage and cache hits so you can see what's happening under the hood.

Every section is editable in TinyMCE. Every regeneration snapshots the prior version into history, so you can restore at will. The Restyle button lets you re-apply the style guide if you've heavily edited a section and want it tightened back to consistency. Or change your style guide entirely — sentence case, "shall" not "must", British English — and restyle the whole document.

When everything reads right, you click Preview document to open the print-ready single-page view. Ctrl+P, save as PDF. That's the artefact you send to suppliers.

Phase 5: Score the replies

Suppliers reply, in their own time, through whatever channel your procurement team uses — email, supplier portal, secure file transfer. The RFP Builder doesn't try to handle that part; suppliers don't get logins. You read their responses yourself.

For each supplier, you open the scoring page. Click any of six colour-graded boxes per requirement (red 0 through deep green 5). Notes textarea below for evidence and caveats. The page autosaves as you go. The left sidebar shows your running average per category, updating live as you score. The bottom bar shows your overall, the team aggregate, and a Spider button that opens a full-screen radar chart of your category averages.

Multiple analysts can score independently. Calibration stays single-blind — you see the count and average from your colleagues but not their individual scores or names. So you score honestly, without being unduly influenced by what someone else thought.

Phase 6: Decide

The Compare page is the decision-making view. Big-number cards rank every supplier by overall score, with gold/silver/bronze ribbons for the top three. A multi-supplier radar overlays everyone on one set of category spokes, each in their own colour. A category winners table marks the leading supplier per category and the gap to second place. If two suppliers tie within hundredths, the table says so.

You take this to your decision-making committee with confidence. Every score is traceable to a requirement. Every requirement is traceable to the original supplier response. Every requirement is traceable, in turn, to the original department quote it came from. Defensibility is baked in from the first .docx you uploaded.

What changes for you

Without this tool, the procurement journey is weeks of spreadsheet wrestling, document concatenation, and political horse-trading. Conflicts get buried. Departments feel unheard. The final document is shaped by whoever stayed up latest, not by what the organisation actually needs.

With it, the wrestle is over by the end of Tuesday. Every voice is preserved. Every conflict is surfaced. Every supplier is evaluated against the same scale, by the same people, at the same time. Your boss asks why you chose Vendor B; you bring up the Compare page and the answer is right there.

The kind of analysis that used to take a week now takes an afternoon.

That's the deep dive on the FreeITSM RFP Builder. It's free, open source, and self-hostable — try it on your next procurement.