Procurement

Pilot wave 2 (M7). Impact: high. See the hub.

Current state (assumptions to validate)

Procurement buys across radically different lines: fresh and frozen provisions (short lead times, cold chain, ISO 22000 supplier requirements), deck/engine/cabin stores and marine coatings (thousands of SKUs, mixed local/import sourcing), safety equipment and PPE (certified products, e-commerce stock), CBRN/hazmat items (restricted, specification-heavy), and tender-driven purchases with back-to-back terms against defense/public contracts. A large share of buying is reactive: a vessel order arrives, part of the list is not in stock, and procurement must source, price, and secure delivery before the sailing deadline — sometimes same-day.

Assumptions — to validate in discovery: team ~6–10; POs in SoftOne; supplier quotes arrive by email in every imaginable format (PDF, Excel, email body); comparisons done manually in Excel; supplier performance tracked informally; import shipments involve customs paperwork coordinated with logistics; no structured supplier master beyond SoftOne records. Pain points: quote-comparison latency on urgent sourcing, no price-history visibility at the moment of buying, supplier certificates (ISO 22000, product certs) chased manually, tender purchases requiring careful spec compliance.

Target AI-first operating model

By M12 an out-of-stock line on a vessel order triggers an assisted sourcing loop: the assistant proposes suppliers from history, drafts the RFQ emails, and when quotes return, an agent extracts and normalizes them into a comparison table with landed-cost estimates and last-paid prices from SoftOne. The buyer decides; the assistant drafts the PO in SoftOne for approval. Supplier certificates are tracked with expiry alerts. Spec-heavy purchases (CBRN, tender back-to-back) get automatic spec-vs-offer compliance checks with human verification. Humans own supplier selection, negotiation, and all commitments.

AI use cases

Use case	Pain addressed	Data needed	Complexity	Impact	Pilot
Supplier quote extraction & comparison (any-format quotes → normalized table with price history)	Manual Excel comparisons under deadline	Quote emails/PDFs, SoftOne purchase history, supplier master	M	H	Y
RFQ email drafting to shortlisted suppliers	Repetitive drafting, slow fan-out	Supplier master, item specs, order context	L	H	Y
PO drafting in SoftOne from selected quote	Re-keying, errors	SoftOne PO module, selected quote data	M	H	N
Supplier certificate & document tracking (ISO 22000, product certs, expiry alerts)	Manual chasing, audit exposure	Supplier documents (SharePoint/scans), supplier master	L	M	N
Spec-vs-offer compliance check for tender/CBRN purchases	High-stakes manual reading	Tender specs, supplier datasheets	M	H	N
Import/customs document pre-drafting with logistics	Paperwork latency on imports	Shipment docs, supplier invoices, HS-code references	M	M	N
Price-anomaly flagging on incoming quotes vs history	Overpaying under time pressure	SoftOne purchase price history	L	M	N

Process transformation opportunities

• Urgent sourcing loop: from ad hoc phone/email scramble to a standard assisted pipeline (shortage → RFQ fan-out → extracted comparison → PO), with cycle time measured.
• Supplier master data: consolidate supplier records, certificates, and performance into one maintained dataset (with BI) instead of tribal knowledge.
• Quote archive: all supplier quotes captured and structured, making price history queryable rather than lost in mailboxes.
• Certificate compliance: expiry-driven alerts replace annual panic before audits.

Required data sources

• SoftOne: suppliers, purchase orders, purchase price history, item master, stock.
• Procurement mailboxes (quotes, order confirmations); SharePoint supplier-document folders; paper certificates → scan.
• Excel comparison sheets (historical, for pattern mining); HS-code/customs reference data.

Potential AI agents

• Quote Comparison Agent — extracts supplier quotes into a normalized comparison with history context; buyer selects.
• PO Drafting Agent — drafts the SoftOne PO from the selected quote; buyer approves before posting.
• Supplier Document Agent — tracks certificates and drafts renewal-request emails; procurement approves sends.

Automation opportunities

• n8n: quotes mailbox → extraction → comparison-ready notification.
• Certificate-expiry scheduled alerts (60/30/7 days).
• Weekly open-PO and overdue-delivery report; reorder-point alert flow from SoftOne stock levels.

Required integrations

• SoftOne read (suppliers, price history, stock) and PO write — see integrations.
• M365: Outlook (quote intake), SharePoint (supplier docs), Teams alerts.

KPIs

KPI	Baseline	M12 target
Urgent sourcing cycle time (shortage → PO)	TBD (M2)	−40%
Quotes compared per sourcing event (avg)	TBD	≥3
POs drafted by agent (share of total)	0	≥60%
Supplier certificates current at any audit	TBD	100%
Purchase price variance vs history flagged before PO	0	100% of anomalies

Risks

• Extraction errors distort a comparison and drive a wrong award → source-quote citation next to every extracted figure; buyer verifies before PO.
• Supplier-sensitive pricing data leaking across users → role-scoped access in the assistant.
• PO write integration errors corrupt SoftOne data → drafts-only writes, human posting, staged rollout per architecture.

Training needs

• Buyers: assistant basics, quote-comparison workflow, verifying extractions (M6–M7).
• Head of procurement: KPI dashboard, price-anomaly review routine (M8).

Deliverables

• Quote-comparison pilot (M7) and production sourcing loop (M9–M10).
• Supplier master + certificate register (with BI); quote archive structure.
• PO Drafting Agent; procurement KPI dashboard.

12-month execution milestones

Month	Milestone
M1–M2	Discovery: sourcing process mapping, quote-format sample collection, baselines
M3	Supplier master data assessment; certificate register design
M4	Assistant onboarding
M6	Training on pilot workflow; quote-extraction accuracy testing offline
M7	Wave-2 pilot: quote extraction & comparison on live sourcing events
M8	Pilot evaluation; RFQ drafting added; PO-draft integration tested
M9–M10	Quote Comparison + PO Drafting Agents in production; certificate alerts live
M12	KPI review; handover