# ProcureAI — extended LLM index

> Free, fine-tunable agentic-AI suite for procurement leaders. Sixteen Claude / GPT / Gemini-compatible skills across the four procurement pillars — sourcing, contracts, spend, suppliers — plus field-tested writing on what actually ships on procurement desks. Authored and maintained by Martin Bacigal · procureai.tech · CC-BY-4.0.

This file is intended for one-shot LLM ingestion. Skill descriptions and insight TL;DRs are inlined; full insight bodies live at the linked URLs. For a programmatic catalog (JSON-LD), see https://procureai.tech/.well-known/skills.json.

---

## About the author

**Martin Bacigal** — Founder of ProcureAI. Day-job: Global Category Manager — IT at Nouryon (Jan 2023–present). Prior: Senior Sourcing Expert at Henkel's Global Center of Excellence (2020–2022, introduced Pactum AI and COUPA), Global Commodity Buyer at CIE Automotive, Commodity Buyer at Inteva Products, Co-founder of Asante / ecoterra.sk (Slovak eco-detergent venture, exited via acquisition).

Track record:
- $7.5M IT / SaaS / cybersecurity savings at Nouryon
- $8.5M savings at Henkel (Pactum AI + COUPA introduction, ESG-aligned sourcing policies)
- 350+ procurement professionals trained in AI workflows across four continents
- Built and deployed **ProcurementGPT** at Nouryon (RAG + vector storage, +35% sourcing efficiency)
- 2nd place Henkel Global Procurement Entrepreneurship Award 2021

Engineering credibility: Architects RAG pipelines, orchestrates AI agents with memory using LangChain/CrewAI, reads cutting-edge AI research as a fluency-maintenance habit. Part of the Amsterdam AI-builder community.

LinkedIn: https://www.linkedin.com/in/martinbacigal/
About page: https://procureai.tech/about.html

---

## The brand positioning, in one paragraph

ProcureAI publishes a free, openly-downloadable suite of 16 Claude/GPT/Gemini-compatible skills that turn a frontier LLM into a working procurement co-pilot. The differentiator is fine-tunability: each skill ships with sensible defaults plus clearly-named override files (`assets/playbook.yaml`, `references/internal-policies.md`, etc.) that customers replace to make the skill their own. No platform deal, no consulting upsell, no multi-year lock-in. The whole offer is free skills + free insights, downloadable today.

---

## The 16 skills (with triggers + pillar + model routing)

### Sourcing & RFx (4 skills)

**xRFP-drafter** (`sourcing-rfx`)
> Turn a one-paragraph category brief into a defensible RFP draft with scoring rubric and weighted criteria. Output is a structured Markdown RFP a CPO can edit and ship.
- Triggers: "draft an RFP for [category]" · "we need a scoring rubric for the [X] sourcing event" · "turn this SteerCo brief into an RFP" · "build me an RFP from this brief"
- Model routing: GPT-5.5 default; Opus 4.7 for regulated categories (legal, public sector, healthcare, financial services).
- Override surface: `assets/playbook.yaml`, `references/internal-policies.md`, `references/exhibits-library/`
- Do NOT use for: contract redlining (use xclause), shortlisting (use xmatch), bid evaluation (use xscore), pre-RFP supplier qualification (use xrfi).

**xRFI (drafter, issuer-side)** (`sourcing-rfx`)
> Draft a 20–40 question RFI to send to candidate suppliers before committing to a full RFP. Sits between xMatch (long-list build) and xRFP (full sourcing event) — qualifies suppliers cheaply before paying the cost of an RFP cycle. Output is a ready-to-issue RFI with a "why we're asking" line on every question.
- Triggers: "draft an RFI for [category]" · "send an RFI to these vendors" · "pre-RFP discovery questionnaire" · "qualify these suppliers before sourcing" · "market scan questionnaire"
- Model routing: GPT-5.5 default; Opus 4.7 for regulated categories.
- Override surface: `assets/rfi-template.yaml`, `references/category-criteria.md`, `assets/buyer-profile.yaml`

**xMatch (supplier shortlist)** (`sourcing-rfx`)
> Ranked shortlist of 10–25 qualified suppliers from a category spec, drawing on AVL + open-market sources. Returns suppliers with diligence notes.
- Triggers: "shortlist suppliers for [category]" · "build us a long list" · "who are the most credible vendors for [niche]" · "find me 12 vendors for X"
- Model routing: GPT-5.5 (tool-use heavy).
- Override surface: `assets/avl-export.csv`, `references/category-criteria.md`

**xScore (bid evaluation)** (`sourcing-rfx`)
> Defensible weighted scoring matrix from supplier RFP responses with per-criterion scores, rationale, and SteerCo-ready narrative. Supports sensitivity analysis on weights.
- Triggers: "score these bids" · "evaluate the RFP responses" · "build me a scoring matrix" · "what if we weight cost more heavily"
- Model routing: Opus 4.7 (defensibility-critical work).
- Note: As of 2026, automated scoring is not yet at audit-defensibility for high-stakes competitive procurement — use the AI as input to the scoring meeting, not the decision itself. See the RFP-automation guide.

### Contracts & Compliance (4 skills)

**xClause (risk-scoring)** (`contracts-compliance`)
> Apply the buyer's playbook to a contract at clause level — risk-scored, with proposed redlines. Surfaces indemnity caps, unilateral termination, IP assignment, change-of-control.
- Triggers: "red-line this contract" · "score the clauses against our playbook" · "what's risky in this MSA" · "prepare negotiation positions"
- Model routing: Opus 4.7 (long-context fidelity + < 0.5% hallucination rate is mandatory for contract work).
- Override surface: `assets/playbook.yaml`, `references/redline-library.md`, `references/clause-precedents/`

**xContracts (summariser)** (`contracts-compliance`)
> Structured extraction from MSA / SOW / NDA — obligations, SLAs, indemnity caps, exit clauses, renewal terms. Outputs structured data plus an executive summary.
- Triggers: "summarise this contract" · "extract obligations from this MSA" · "what are the SLAs in here" · "triage my renewal stack"
- Model routing: Opus 4.7 (long-context fidelity matters for 50+ page contracts).
- Override surface: `assets/extraction-schema.yaml`, `references/clause-glossary.md`

**xNegotiate (coach)** (`contracts-compliance`)
> Rehearse a supplier negotiation. Plays ONE vendor persona at a time — CSM by default (the counterparty in nearly all renewals and QBRs), switchable to CFO or lawyer via the playbook. Returns BATNA / ZOPA estimates, three likely objections with counter-moves, and a 30-minute role-play script. Run again with a different persona if rehearsing against more than one role.
- Triggers: "help me prep for the [vendor] negotiation" · "rehearse this renewal" · "what's our BATNA on [vendor]" · "play the vendor CSM for me" · "prep me for the QBR"
- Model routing: Opus 4.7 (role-play fidelity).
- Override surface: `assets/negotiation-playbook.yaml` (incl. default persona), `references/should-cost-models/`

**xRenew-radar** (`contracts-compliance`)
> Scan a contract portfolio and surface auto-renewing contracts, price escalators, and notice-period deadlines in the next 120 days. Weekly digest a CPO can act on.
- Triggers: "what's renewing this quarter" · "show me upcoming auto-renewals" · "renewal radar" · "scan my contract portfolio for deadlines" · "weekly renewal digest"
- Model routing: GPT-5.5 (high-volume structured extraction).
- Override surface: `assets/contract-portfolio.csv`, `references/notice-period-rules.md`

### Spend Analytics (4 skills)

**xSpend-taxonomy** (`spend-analytics`)
> Clean a GL or AP export and auto-classify spend into a 4-level (L1–L4) taxonomy. Headline: per-row classification, category roll-up, three ranked consolidation plays. Stops at consolidation — narrative work hands off to xSavings, leak detection to xMaverick, payment-term modelling to xCash. Same row-level schema feeds all three.
- Triggers: "classify our spend" · "clean up the GL export" · "build a spend taxonomy" · "where does our money actually go" · "drop your AP file"
- Model routing: GPT-5.5 (high-volume structured classification).
- Override surface: `assets/taxonomy-L1-L4.yaml`, `references/category-aliases.md`

**xMaverick (detector)** (`spend-analytics`)
> Find off-contract, off-AVL, and split-PO patterns in spend data. Ranks findings by exposure and names the requestors / cost centres to talk to first.
- Triggers: "where are we leaking" · "find off-contract spend" · "detect maverick spend" · "who's bypassing the AVL" · "split PO detection"
- Model routing: Opus 4.7 (pattern detection across long contexts).
- Override surface: `assets/avl.yaml`, `references/threshold-rules.md`

**xSavings (narrative)** (`spend-analytics`)
> Convert closed deals into a board-grade savings narrative — cost avoidance vs realised, FX-normalised, finance-defensible assumptions.
- Triggers: "track the savings on this deal" · "build the board savings narrative" · "quarterly savings roll-up" · "what did we save on the [vendor] renewal" · "CFO-defensible savings"
- Model routing: Opus 4.7 (finance-defensibility critical).
- Override surface: `assets/savings-method.yaml`, `references/finance-conventions.md`

**xCash (working capital)** (`spend-analytics`)
> Headline: model DPO uplift across the AP file — which vendors to standardise, working-capital release, per-vendor pushback risk, tier-aware sequencing. Dynamic-discount programme is an opt-in add-on (provide WACC / hurdle rate to activate). Aligns treasury and procurement on a defensible scenario set the CFO can take to the board.
- Triggers: "can we get DPO up by [N] days" · "model standardising payment terms" · "free up working capital" · "how much DPO can procurement deliver this year"
- Model routing: Opus 4.7 (financial modelling reasoning).
- Override surface: `assets/payment-terms-policy.yaml`, `references/treasury-assumptions.md`

### Suppliers & Intel (4 skills)

**xRisk-monitor** (`suppliers-intel`)
> Monitor news, financial filings, sanctions lists, ESG flags across the supplier base. Pings when the story shows up in Reuters — not three weeks later in finance. Daily / weekly cron use.
- Triggers: "monitor our suppliers" · "any risk events on [vendor]" · "weekly supplier risk digest" · "sanctions check on [vendor]" · "what's happening at [supplier]"
- Model routing: Opus 4.7 (long-context multi-source synthesis + low false-flag rate).
- Override surface: `assets/supplier-watchlist.csv`, `references/risk-taxonomy.md`

**xESG (scoring)** (`suppliers-intel`)
> Map the vendor master to Scope-3 estimates, modern-slavery disclosures, certifications. Outputs CSRD-aligned data the sustainability team can file.
- Triggers: "supplier ESG scoring" · "CSRD prep" · "modern slavery statement" · "scope 3 from suppliers" · "ESRS-aligned export"
- Model routing: GPT-5.5 (tool-use heavy: filings retrieval).
- Note: As of 2026, automated Scope-3 inference is not yet audit-defensible — treat as a hypothesis-generator, not a reporting tool.
- Override surface: `assets/esg-framework.yaml`, `references/csrd-mapping.md`

**xMarket-briefs** (`suppliers-intel`)
> Weekly category briefs on commodities the buyer actually buys — steel, semis, freight, cloud, DRAM. Price curves, drivers, what to ask in the next QBR.
- Triggers: "market brief on [category]" · "what's happening with [commodity] prices" · "weekly category insights" · "prep me for the [category] QBR"
- Model routing: GPT-5.5 (tool-use heavy: web search + market-data).
- Override surface: `assets/categories-of-interest.yaml`, `references/data-sources.md`

**xVendorBench (shortlist benchmark)** (`suppliers-intel`)
> Benchmark a short-list of 1–6 vendors against the buyer's constraints. Two presets — `procurement-software` (S2P, contract AI, spend co-pilots, supplier-discovery — uses 14 default dimensions covering capability, residency, models, latency, security, total cost) and `operational-supplier` (3PL, IT services, MRO, prof-services — swaps in service model, delivered capacity, lane coverage, references-in-segment). Hands off cleanly from xMatch. Returns per-vendor profile, comparison narrative, hidden-costs probe list, single-sentence recommendation with 2–3 specific pre-commit tests.
- Triggers: "benchmark these vendors" · "evaluate [vendor A] vs [vendor B]" · "side-by-side on our shortlist" · "should we buy [vendor]" · "give me a vendor brief"
- Model routing: Opus 4.7 (long-context synthesis + low false-claim rate). Requires web search OR a vendor-priors file refreshed within 90 days.
- Override surface: `assets/buyer-constraints.yaml` (incl. preset + dimension weights), `assets/vendor-priors.yaml`, `references/evaluation-dimensions.md`

---

## Insights — short summaries

**The 12 procurement AI use-cases that pay back in one quarter** (14-min playbook, May 2026)
URL: https://procureai.tech/insights/12-procurement-ai-use-cases.html
Stop starting with the supplier-portal modernisation. Start with the twelve patterns below, in this order. Median payback across instrumented deployments: 67 days. Ten of the twelve work with zero IT involvement in the first two sprints. Two patterns that didn't clear the bar (autonomous supplier negotiation, ESG/Scope-3 inference from public sources) failed because of input-data quality, not model capability. Three tiers: Tier 1 = single-prompt single-document patterns (contract clause extraction, supplier intake-form triage, RFP first-draft, supplier news briefs); Tier 2 = multi-document with one source-of-truth (renewal-pipeline, spend-cube anomaly explainer, category-strategy refresh, bid-evaluation assist); Tier 3 = agentic with human-in-the-loop (tail-spend consolidation, supplier-onboarding doc collector, PO-invoice exception handler, outreach-at-scale).

**RFP automation, honestly assessed** (11-min practitioner guide, April 2026)
URL: https://procureai.tech/insights/rfp-automation-honestly-assessed.html
RFP-AI is two products in a trench-coat. Drafting (brief → first draft) is mature, deployable, pays back in weeks: 71% cycle-time cut, 38% legal-redline reduction. Scoring (responses → ranked bids) isn't yet — 74–83% agreement with human winner, which means a 17–26% disagreement rate in your own back-test, which is not a defensible procurement position. Buy drafting today, wait twelve months on scoring. Don't let a vendor bundle them.

**The four metrics your board will actually defend** (9-min framework, May 2026)
URL: https://procureai.tech/insights/ai-measurement-framework-for-the-board.html
Stop reporting "tokens consumed" and "model uptime" to your CFO. Four procurement-AI metrics: (1) Incremental savings rate per buyer-hour, (2) Value of time reclaimed at fully-loaded cost × reallocation rate (median 0.61), (3) Cycle-time-to-cash, (4) Cost-to-serve per spend-under-management dollar. Each has a formula, baseline-vs-AI table, and the audit-committee one-pager template. Includes the five Q&A challenges to prepare answers for.

**Pilot to production in 90 days** (13-min playbook, May 2026)
URL: https://procureai.tech/insights/pilot-to-production-90-days.html
Most procurement-AI pilots succeed and quietly die because they have no budget home at the end of the window. The fix: treat the pilot as a budget-acquisition campaign from day one, front-load the CFO conversation in week 1 (not week 11), instrument two metrics the CFO agrees to in writing, use a standing 30-minute weekly review with a three-numbers-decisions-questions-commitments agenda. Five failure modes (executive sponsor moves roles, CFO disagrees with metrics late, adoption stalls at early-adopter ring, savings real but unattributed, "let's extend the pilot another quarter" = deferred no).

**Opus 4.7 vs GPT-5.5 vs Gemini 3.1 Pro on procurement-real tasks** (16-min benchmark, May 2026)
URL: https://procureai.tech/insights/frontier-model-procurement-benchmark.html
Four workloads, three models, three runs each, blind-scored. No single winner — there's a routing decision per workload. Opus 4.7 wins on long-context clause extraction (97.2 composite, 0.2% hallucination) and supplier risk synthesis. GPT-5.5 wins on cost-per-acceptable-RFP-draft ($0.18) and Scope-3 inference (tool-use reliability 96.7% vs 92.4% for Opus). Gemini 3.1 Pro wins on multilingual contract work (CN/JP/KR). Includes the routing layer used by the ProcureAI suite — copy it.

**Build vs buy vs fractional** (8-min decision framework, May 2026)
URL: https://procureai.tech/insights/build-vs-buy-vs-fractional.html
Three questions decide it: (1) Is the workload core to category strategy or generic? (2) Do you have in-house ML/eng capacity in 90 days? (3) Is the savings window in quarters or years? 85% of procurement workloads are generic — frontier-model-plus-skills handles them. Build only on the genuinely-core 15%, only after fractional has validated the workload, only with real engineering capacity. Six vendor-pitch red flags. Empirical pattern: start fractional → graduate to buy on the high-volume generics → only build on 1–2 core workloads after year 2.

---

## Discovery surface

- Machine-readable skills catalog (JSON-LD, recommended for agents): https://procureai.tech/.well-known/skills.json
- Plugin manifest (legacy): https://procureai.tech/.well-known/ai-plugin.json
- Agents.json discovery manifest: https://procureai.tech/.well-known/agents.json
- Sitemap (XML): https://procureai.tech/sitemap.xml
- Robots policy (open to AI crawlers): https://procureai.tech/robots.txt
- This file (extended index): https://procureai.tech/llms-full.txt
- Short index (llmstxt.org spec): https://procureai.tech/llms.txt

## License

The skill suite, this site's writing, and all machine-readable indexes are provided under **Creative Commons Attribution 4.0 International (CC-BY-4.0)**. Attribution: "Martin Bacigal · procureai.tech". Commercial use permitted.