Guide · The AI Map

The AI Map of an RFP: where AI helps, where it hurts

The full lifecycle of an indirect RFP (€0.5-5M) on a traffic light: where AI stays out, where it assists, where it runs on its own. From real engagements.

By Alex Lio · The Procurementor · 10 min read

"Nobody launches a serious RFP in 10 days." I hear it in every first meeting, and done by hand, it's true. An indirect RFP of €0.5-5M, run properly, takes weeks: a framing that waits on the sponsor, interviews that keep slipping, a tender pack rewritten three times, offers picked apart line by line in a spreadsheet, a negotiation entered out of breath. Across 12 years and €600M+ of managed spend, I have found the same pattern everywhere: the decisions take hours, the execution between the decisions takes weeks.

That is exactly what a method compressed by tooling changes. Execution tasks drop from days to hours. Judgment tasks keep their duration, by design: they are the ones that produce the savings. On my own engagements, the result is simple to measure: an RFP launched in 10 days, with no stage sacrificed.

You still need to know what to compress. Putting AI everywhere destroys value as surely as putting it nowhere. This guide is a map: the full lifecycle of an RFP, from the first coffee with the sponsor to the teams adopting the contract, with a traffic light set on every stage. It comes from real engagements, in indirect procurement, on €0.5-5M scopes. It serves two purposes: knowing where to invest your buyer hours, and knowing where the tooling gives hours back.

How to read the map: three colors

Color	Regime	Typical stages
🔴 Red	Zero AI, pure craft	Need, arbitrations, negotiation
🔵 Blue	AI-assisted, expert in control	Tender pack, offer analysis, should-cost
🟢 Green	AI-autonomous, controlled	Comparison tables, market watch, extraction

🔴 Red, pure craft. Redefining the need with the business, aligning stakeholders, negotiating and closing. This is where the savings are born, and this is where AI loses you time or judgment.

🔵 Blue, AI-assisted, expert-supervised. Tender pack drafting, offer analysis, should-cost modeling. AI produces the first draft; every output is verified before it exists for the client, no exception.

🟢 Green, AI-autonomous, under control. Comparison tables, market watch, document extraction. Tasks with clear instructions and results you can spot-check. This is where days become hours.

Now let's walk through a real RFP, stage by stage.

01 · Sponsor framing 🔴

Every RFP starts with a person who has a problem and a budget. At this first stage I look for three things: the real trigger (contract expiry, dissatisfaction, cost pressure), the sponsor's actual weight in the organization, and the unwritten constraints: a protected incumbent, a political calendar, a past attempt that left marks. No tool reads that. One well-run framing hour saves three weeks of misdirected consultation, and that ratio alone justifies the red.

02 · Requirement interviews 🔴

The stage everyone wants to shorten, and the one that pays the most. Interviewing the users, I hunt the gap between the stated need and the real need: the spec inherited from the previous contract, the oversized service level, the requirement nobody can justify anymore. My standard line: "the savings sleep in the spec, rarely in the sticker price". On a €1.2M services RFP, two requirements removed in interviews weighed more in the final result than the whole negotiation. AI does not hear the hesitation of a user defending a habit; I do.

03 · Baseline and should-cost 🔵

Before going to market, I rebuild two numbers: what the scope really costs today, and what it should cost. The collection is green: an agent extracts and structures contracts, invoices and volumes in a few hours, where that grind used to take days. On most indirect scopes, this reconstruction alone already surfaces anomalies: prices never reindexed, volumes invoiced beyond reality, services paid twice. The should-cost modeling is blue: AI assembles the cost structure, the market ratios and the assumptions, and I verify every assumption before the model becomes a reference. A wrong should-cost is worse than none: it anchors everything that follows in the wrong place.

04 · Drafting the tender pack 🔵

The tender pack: specifications, consultation rules, an imposed response grid, a contract template. AI drafts the first version from the framing and the interviews, and I concentrate my time on the 20% of the document that decides the outcome: the award criteria, their weighting, the response format that will make offers comparable without retyping. Two days of drafting become half a day. Another gain, less visible: consistency. The first draft reuses the vocabulary of the interviews word for word, and a tender pack that speaks the users' language receives more accurate offers. Every page is reviewed before it exists for the client, blue rule.

05 · The consultation window 🟢

Between sending the pack and receiving the offers, most of the work is logistics: follow-ups, acknowledgment tracking, keeping the Q&A register, watching the candidates (financial health, news, ongoing litigation). An agent runs this tracking autonomously and flags the exceptions. The calendar also holds thanks to this: follow-ups go out on time, even when I am in interviews on another project. I keep my hands on one point only: the answers to questions that touch the scope or the criteria, because one badly calibrated answer can distort the competition.

06 · Offer analysis 🔵

When the offers land, two movements. Green first: extraction, mapping onto the grid, comparison tables, detection of empty cells and gaps between lots. Days of processing become hours. Blue next: price consistency against the should-cost, conditions hidden in the annexes, execution risks. AI prepares the analysis, I verify every score before a ranking exists. The weighted grid, frozen before the first offer, is the only reference: the agent fills it in, it never modifies it. One badly extracted number can disqualify the right supplier, and nobody will spot it in a clean table.

07 · Final negotiation 🔴

Everything before this exists to arrive here in a position of strength: a cleaned-up need, a solid should-cost, genuinely comparable offers. The negotiation itself stays a human exercise: reading the other side, sequencing the concessions, sensing the moment to close. I prepare my scenarios upstream, priced on the should-cost, then in the room, no tool. Same regime for the internal arbitrations: which supplier, which balance between price, risk and relationship. Those arbitrations are won face to face, never in a prompt. This is also the stage that benefits most from the hours won upstream: arriving rested at the table, scenarios in mind, changes the outcome more surely than any tactical lever.

08 · Contracting 🔵

The agent compares the contract template from the tender pack with the proposed contract and lists the deviations clause by clause: green extraction, blue reading. Legal and I decide on what matters: service level commitments, reversibility, indexation, liability caps. On a 3-year contract, one poorly framed indexation clause can give back half the negotiated savings. The signature closes the RFP and opens the real subject: making what was negotiated live.

09 · Adoption 🔴

A well-negotiated contract that the teams work around produces zero savings. Adoption is field work: explaining the change to the users, bringing the prescribers on board, handling the objections of the outgoing supplier's regulars. I put human time into it, on site when needed. Green comes back afterwards, quietly: automated compliance tracking (actual volumes, invoiced prices against negotiated prices) that flags drift without anyone processing invoices. This is often where the difference plays out between savings negotiated and savings banked.

How to use this map on your next RFP

One caveat first: the borders of the map move with the quality of your data and the maturity of your teams. A task that is green for me can stay blue for you in the first month, while trust in the tool gets built. The colors are earned. With that said, four concrete uses.

Color your timeline. Lay out your stages on a sheet and give each one a color. If your plan spends more days on green stages than on red ones, it is upside down: you are paying buyer days for tasks an agent does in hours, and trimming the interviews that produce the savings.

Protect the red. The time won on green and blue is wasted if it shortens the project: reinvest it in the requirement interviews and the negotiation preparation. That is the heart of the method: the tooling gives hours back to the craft, and the craft places them where they pay.

Tool up the green first. Document extraction and comparison tables are the fastest, lowest-risk wins: clear instructions, verifiable results. Blue comes next, with a strict review discipline. Red, never. And start on a live RFP rather than an artificial pilot: time measured on a real project will be your best internal argument.

Measure one thing to start. The delay between the stated need and the launch of the consultation. That is the number that convinces a sponsor, and the one this map moves first.

For my part, I work with four agents I built myself: RFP drafting, offer comparison, supplier assessment and should-cost, hosted in Europe. They cover the blue and the green of this map; the red stays handmade, by principle as much as by experience.

This map, tooled end to end, is what makes the RFP Launch possible: an RFP launched in 10 days, red stages included. The detail is on the RFP Launch page. And to test the map against your next project, book a slot: come with your scope, we color it together.

Members guide

Read the full map

All nine stages colored in, from sponsor framing to adoption, with the practical detail behind every blue and green zone. Leave your email and the map opens.