Warehouse

Pilot wave 3 (M9). Impact: high. See the hub. Warehouse is deliberately late-wave: its use cases touch physical operations and bonded stock, so they build on SoftOne integration proven in wave 2 and on process discipline established first.

Current state (assumptions to validate)

Prime Products operates bonded warehouses (goods held under customs suspension for ship supply) plus regular stock across Piraeus/Perama locations, including temperature-controlled storage for provisions with cold-chain obligations under ISO 22000. The defining rhythm: orders confirmed against a vessel’s port window trigger pick-pack-load runs where the deadline is a sailing time, not a service level. Picks span wildly heterogeneous goods — frozen meat, engine spares, charts, PPE cartons, hazmat items with segregation rules — often consolidated onto one truck to a berth.

Assumptions — to validate in discovery: team ~15–25 across sites; stock in SoftOne, possibly without granular bin locations; picking from printed lists; bonded stock movements requiring customs-matched documentation (with logistics); cycle counting informal; receiving paperwork manual; temperature logs manual or standalone loggers. Pain points: pick errors and shortages discovered at packing (too late against a sailing), stock-accuracy gaps between SoftOne and shelf, bonded/duty-paid stock confusion risk, peak-day load balancing by improvisation, tribal knowledge of where things are.

Target AI-first operating model

By M12 the day starts with an agent-generated pick plan: the day’s orders ranked by sailing deadline, grouped into pick runs, shortage risks flagged before pickers move (stock check vs order lines), and hazmat/cold-chain handling notes attached automatically. Receiving is assisted — supplier delivery notes photographed/scanned and matched to POs, discrepancies flagged on the spot. Inventory questions (“do we have X, where, bonded or free?”) are answered by the assistant instead of a walk to the shelf. Warehouse leads consult exception dashboards; supervisors, not the system, decide sequencing overrides. All stock postings remain human-confirmed.

AI use cases

Use case	Pain addressed	Data needed	Complexity	Impact	Pilot
Deadline-ranked daily pick planning with shortage pre-check	Late shortage discovery, improvised sequencing	SoftOne orders + sailing deadlines, stock by location	M	H	Y
Receiving assistant: delivery-note scan → PO match → discrepancy flag	Manual matching, delayed booking	Scanned delivery notes, SoftOne POs	M	H	Y
Natural-language stock lookup (incl. bonded vs free status)	Walk-and-check culture, phone interruptions	SoftOne stock, warehouse/bin data	L	M	N
Cycle-count planning & variance narrative	Poor stock accuracy, unfocused counting	SoftOne stock movements, count history	M	M	N
Hazmat/cold-chain handling-note generation per pick run	Compliance depends on individual memory	Item hazard/temperature attributes, handling SOPs	L	M	N
Packing/delivery document pre-check against order & customs status	Errors found at the berth	Order docs, customs status (with logistics)	M	H	N

Process transformation opportunities

• Pick-run planning: from printed lists in arrival order to deadline-ranked, shortage-pre-checked runs — the core redesign.
• Receiving: scan-first receiving with same-day SoftOne booking, killing the paper backlog.
• Stock accuracy: standing ABC cycle-count program driven by movement data, replacing sporadic counts.
• Bin location discipline: introduce/complete bin-level locations in SoftOne so lookups and pick paths are data, not memory (prerequisite — assumption on current state, to validate).

Required data sources

• SoftOne: stock by warehouse/location, orders with delivery deadlines, POs, stock movements, item attributes (hazard class, temperature regime — likely incomplete, to enrich).
• Bonded-stock records and customs status flags (with logistics/customs docs).
• Scanned delivery notes and packing lists (paper → scan pipeline).
• Temperature logs (loggers or manual sheets); handling SOPs (with compliance).

Potential AI agents

• Pick Planning Agent — proposes the day’s deadline-ranked pick runs with shortage flags; warehouse lead approves the plan.
• Receiving Match Agent — matches scanned delivery notes to POs and drafts the goods-receipt; storekeeper confirms posting.
• Stock Query Agent — answers stock/location/bonded-status questions read-only; no approval needed (no writes).

Automation opportunities

• n8n: morning pick-plan generation and Teams post to warehouse leads.
• Shortage alert flow: order line vs stock check on order confirmation.
• Temperature-logger data collection and out-of-range alerts (if loggers support export).
• Scheduled stock-accuracy report per site.

Required integrations

• SoftOne read (stock, orders, POs) and goods-receipt draft write — see integrations.
• Scanning pipeline (multifunction devices → SharePoint intake folder); Teams alerts.

KPIs

KPI	Baseline	M12 target
Pick errors per 1,000 lines	TBD (M2)	−50%
Shortages discovered at pick/pack (vs pre-flagged)	TBD	−70%
Stock accuracy (count vs SoftOne, A-items)	TBD	≥98%
Receiving-to-booking lead time	TBD	Same day ≥95%
Deliveries late to berth for warehouse reasons	TBD	~0

Risks

• Bad pick plan disrupts a sailing-critical delivery → supervisor approves every plan; manual override always available; pilot on one site first.
• Bonded/free stock misstatement creates customs exposure → bonded status shown with source record; no AI-inferred customs status, ever.
• Low digital familiarity on the floor → hands-on training, tablet/terminal UX, warehouse leads as champions per the change plan.
• Data quality (locations, item attributes) too poor for planning → M3–M6 data-enrichment workstream is a hard prerequisite.

Training needs

• Warehouse leads: pick-plan review, exception dashboard (M8–M9).
• Storekeepers/pickers: receiving-scan workflow, stock-query assistant (M9).
• All: hazmat/cold-chain note usage refresher with compliance (M10).

Deliverables

• Bin/attribute data-enrichment workstream output (with BI).
• Receiving-scan pipeline; Pick Planning + Receiving Match Agents (pilot M9, production M11).
• Cycle-count program; warehouse KPI dashboard.

12-month execution milestones

Month	Milestone
M1–M2	Discovery: floor observation, process mapping, data-quality audit, baselines
M3–M5	Data enrichment: bin locations, hazard/temperature attributes (prerequisite work)
M6	Stock Query Agent available (read-only, low risk, early win)
M8	Training for leads; receiving-scan pipeline ready
M9	Wave-3 pilot: pick planning + receiving match at one site
M10	Pilot evaluation; second site
M11	Agents in production both sites; cycle-count program running
M12	KPI review; handover to warehouse management