Download free
ProcureAI · About
Founder · Martin Bacigal

Open-sourced from one procurement desk — made for every CPO who wants it.

Martin Bacigal — Founder, ProcureAI

Open-sourcing what worked.

Inteva Commodity buying · RFQs
CIE Automotive Global · EU & NA
Henkel 2nd place procurement awards
Nouryon AI operating model · IT category

Built agentic procurement workflows on real category desks — from commodity buying to global IT sourcing. Identified Pactum AI and Coupa before the category had a name for it. The same methodology — governance framework, model routing, capability tiers, board-level KPIs — is what the ProcureAI suite packages for any function serious about running AI as infrastructure, not piloting it.

The field notes — what held in production and what didn't — live under /insights.

  • Day job: Global Category Manager — IT @ Nouryon
  • Delivered: $16M+ cumulative IT / SaaS / cyber / telecom savings
  • Built: ProcurementGPT — RAG sourcing co-pilot, AI governance framework, and capability roadmap at Nouryon
  • Stack: LangChain, CrewAI, vector stores, frontier-model orchestration, enterprise AI policy
  • Trained: 350+ procurement professionals across four continents
Open-source by intention

Every SKILL.md, every template, every formula — fork it, fine-tune it, share it with the team. Override files (assets/playbook.yaml, references/internal-policies.md) make each skill yours. Internal use, modification and redistribution are free. Not for resale.

A suite, not sixteen orphans

xRFP feeds xScore. xContracts feeds xRenew. xSpend feeds xSavings. Every skill writes to the next one's input contract, so the workflow connects — category strategy through bid scoring, contract obligations, renewal radar, all the way to the board readout. The signal is in the seams.

CISO-clean by default

Skills run inside your own Claude Desktop, Claude Code or agent runtime — your data never touches a ProcureAI server. No training on your prompts. Frontier-model routing is yours to configure. Audit trail per invocation. The walkthrough deck is a public artefact — email me and I'll send it.

Want to talk about how to wire the suite into your category? Email [email protected] — I read every one.

Free · 5 pillars · 2 minutes

The CPO AI Readiness Scorecard.

Five pillars, two questions each. Step through them for an instant read on where your team is foundational versus ready to wire agents into the workflow. The full 42-question diagnostic — scored against a peer benchmark, with the specific 90-day moves per pillar — ships inside the free skills download.

Pillar 1 of 5

Data foundations

Spend taxonomy, contract corpus, supplier master

Our spend is classified to a consistent L1–L4 taxonomy — not just raw GL codes.

Our contract corpus is searchable as text — an agent could retrieve a clause today.

Pillar 2 of 5

Workflow agentic-readiness

Sourcing cycles, templates, repeatable steps

We've mapped which sourcing steps are rules-based versus genuine judgement calls.

We have reusable templates — RFP, scorecard, renewal scan — an agent could draft from.

Pillar 3 of 5

Governance & CISO sign-off

Model cards, audit trail, residency, sign-off

We have a signed-off data-handling and residency position for AI tools.

Every AI-assisted output can be traced — who ran it, which model, what inputs.

Pillar 4 of 5

Talent & operating model

Ownership, capacity, build-vs-buy

Someone owns configuring and maintaining AI skills — with real capacity, not goodwill.

We've made an explicit build-internally versus route-to-vendor call.

Pillar 5 of 5

Model-stack literacy

Platform deal vs frontier-routed orchestration

We've looked past a single platform pitch and understand frontier-model routing.

We can match the model to the task — reasoning, long-context, or cheap classification.

Pillar 1 of 5

No email needed · answers are logged anonymously to sharpen the benchmark

Your five-pillar read

0/ 20

Data foundations 0 / 4
Workflow agentic-readiness 0 / 4
Governance & CISO sign-off 0 / 4
Talent & operating model 0 / 4
Model-stack literacy 0 / 4

That's the 10-question read. Here's the full 42.

The full diagnostic scores every pillar against a peer benchmark and returns the specific skills and 90-day moves that close each gap — it ships inside the free skills download.

Get the full diagnostic + 90-day moves

16 SKILL.md files · 5 working files · CC-BY-4.0 · no demo call

ROI Calculator · free · open formulas

Model the savings before the kickoff.

Feed in your spend under management, headcount, contract count and category mix. Every assumption is a named cell — finance can pressure-test it, override it, or rebuild it from scratch.

Five live savings drivers

1
Sourcing-cycle savings% of addressable spend, adoption-ramped
2
Renewal recaptureauto-renewals caught before notice closes
3
Tail-spend consolidationfragmented indirect into framework deals
4
FTE productivity recapturehours/week × FTEs × loaded cost
5
RFP cycle-time reductionfirst draft from weeks to an afternoon
Download the XLSX

Free · no email required · < 50 KB · bonus 38-row L1–L4 spend-taxonomy starter tab

Default scenario · $50M spend · 35 FTEs · 6 categories live

$9.2M

3-year cumulative
savings modelled

~6.1%

of 3-year spend
under management

71%

first-draft RFP
cycle-time reduction

~1.5mo

payback period
vs $250k Y1 invest

Numbers from the workbook's default inputs. Plug in your spend, headcount and category mix — the formulas update live.

FAQ

Questions CPOs actually ask.

Business case first — or jump straight to install, routing and fine-tuning.

No — and it shouldn't try to. The suite sits above your system of record: drafting RFPs, scoring bids, summarising MSAs, classifying spend, surfacing renewals. Your ERP, CLM and P2P stay the ledger. Think co-pilot layer, not rip-and-replace. Most teams wire one skill into an existing workflow first — usually xRFP or xRenew — and expand from there.
Same day for a single skill: unzip, drop into Claude Desktop or Claude Code, run one real task — an RFP first draft, a renewal scan, a spend taxonomy pass. A connected workflow across two or three skills typically lands in two to four weeks once someone owns configuration and your override files (playbook.yaml, internal policies) are in place. The readiness scorecard above tells you which pillar will slow you down.
Someone with real capacity, not a side project. Usually a category lead or a procurement-ops person who already owns templates and tooling — with CISO sign-off on model routing and data residency. IT doesn't need to build anything; they approve which runtime and which model API you point at. The suite is markdown on disk — no ProcureAI server, no new platform to operate.
Free for internal use inside your organisation — including modification, customisation and redistribution to colleagues. Fork the repo, drop the suite on your team's shared drive, build a private version on top. The one prohibition is resale or repackaging into a competing service. Attribution to procureai.tech in any forked SKILL.md is appreciated but not required.
Anywhere a SKILL.md (YAML frontmatter + markdown body) is read. That means Claude Desktop via its Skills folder, Claude Code via ~/.claude/skills/, the Claude Agent SDK via its skill loader, the Anthropic API as the system prompt of a tool-using agent, plus n8n / LangChain / CrewAI / OpenAI Assistants by pasting the body into the system prompt. Same files, every runtime.
Unzip the bundle, open Claude Desktop → Settings → Capabilities → Skills, and upload each SKILL.md (or drag the whole skills/ folder). Claude reads the frontmatter and auto-activates the right skill based on what you ask — type "draft an RFP for managed cloud" and xRFP fires; "scan contracts for renewals" pulls in xRenew. No slash commands to memorise.
Copy the skills/ tree into ~/.claude/skills/ (user-wide) or .claude/skills/ in your project (repo-scoped). Restart Claude Code and the skills load automatically — confirm with /skills. Same applies to the Claude Agent SDK's skill loader; point it at the folder, it indexes the frontmatter.
Yes. The body of every SKILL.md is a self-contained system prompt — drop it into a Set node in n8n, a PromptTemplate in LangChain, an Agent role in CrewAI, or the system message of an OpenAI Assistant. The skills are deliberately framework-agnostic markdown so they survive any runtime move. If your platform supports tool routing, the frontmatter's description field doubles as the routing hint.
That's the design. Each skill folder ships with override files — assets/playbook.yaml (your negotiation playbook, scoring weights, sourcing rules), references/internal-policies.md (procurement policies, residency requirements, approval matrices), templates/ (your house RFP / contract templates). Edit those, leave the SKILL.md alone, and the skill becomes yours without forking it. Quarterly suite refreshes won't overwrite your overrides.
It's your runtime, you pick. Sensible defaults: reasoning-heavy skills (xRFP, xClause, xNegotiate, xScore) to Claude Opus 4.7 or GPT-5.5; long-context contract ingestion (xContracts, xRenew) to Gemini 3.1 Pro or Llama 4 Scout; bulk classification (xSpend) to GPT-4o-mini; privacy-critical workloads on-prem with DeepSeek V4. Full benchmark with prompts and accuracy-per-dollar in /insights.
Only as far as the model API you point your runtime at — and on the model-provider terms you've already signed. The skills themselves are static markdown on your disk; ProcureAI has no server in the loop, no telemetry, no callback. If you route to Claude Desktop, you're on Anthropic's no-training-on-inputs terms. If you run DeepSeek V4 locally with Ollama, nothing leaves your laptop at all.

Got the answers. Now get the skills.

Sixteen SKILL.md files — drop into Claude Code or any agent runtime and ship the work this quarter.

Download the skills