# Deck 03 · DC Optimisation Platform — Seed brief

**Status:** Draft (not yet built)
**Saved:** 2026-06-28 by Jeff (verbal)
**Owner:** Plex / OMX Strategy + DC Operations

---

## One-line thesis

The **full OMX DC optimisation platform** — live picker tracking, e-ink routing, digital-twin layout optimisation, voice-confirmed picks, and end-of-conveyor robotic Tetris packing.

A unified, IoT-enabled DC stack that replaces the patchwork of labour forecasting / replen / pick rates / put-away with one closed-loop system.

## What the platform does (Jeff's verbal scope · raw)

1. **WhoOnSite — live staff location**
   - Real-time positions of pickers / receivers / forklift drivers across the DC floor

2. **Pick rates → labour forecasting → leaderboards**
   - Live LPPH per picker → fed into labour forecasting → daily/weekly leaderboards + efficiency metrics

3. **Replen influence loop**
   - Demand shifts → recommend moving stock to better-picking locations
   - Move stock around the DC to reduce walk times

4. **E-ink labels on bin locations**
   - Run in/out of locations
   - Manage put-away through replen instructions
   - Flash a colour to show which picker should go to that location next
   - Speeds up finding things → faster pick times

5. **Button-press voice checksum read-back save**
   - End-of-pick confirmation — picker presses the e-ink button to save the voice readback ("3 of SKU 12345")
   - Closes the audit loop

6. **Digital twin**
   - Model the ideally-optimised DC layout
   - Simulate moving stock locations to reduce walk times
   - Feeds back into the replen-influence loop (recommends layout changes)

7. **Robotic Maxi-packing at conveyor end**
   - Arms know the carton (Maxi) sizing coming down the conveyor
   - "Tetris" the items into optimal pack arrangement
   - Reduces back strain (no more manual heavy/awkward packs)

8. **CHCH DC — pragmatic existing-conveyor + smart-tech overlay**
   - **Existing conveyor stays** — manually pushed around (no automation rebuild)
   - **Hand-held pick device** in the picker's hand (BYOD-class scanner / phone)
   - **E-ink bin labels** layered on top of existing racking (flash colour for next pick)
   - **Box-size Python script** splits the pick into N optimal carton sizes
   - **Cheap box-making robot** (sub-$50k category) builds and slides the right-sized box onto the pack table
   - Barcode printed on each box already encodes the **pick locations** in the order the picker will walk
   - Eliminates "wrong box size" waste, eliminates picker walking back to the label printer
   - **Key principle: don't reinvent the conveyor, layer smart tech on top.**
   - Each DC (AKL conveyor + CHCH bulk) gets the platform shape tuned to its layout

9. **Box-shunt cheap robot** (no-pack-slip lane)
   - Small shunt robot pushes boxes that don't need a pack slip down the line
   - Manual cut + seal still happens — that's fine, low value to automate further
   - **Whole stack philosophy:** lots of cheap, simple, easy automation layered together — not one expensive integrated rig

## Why now

- E-ink + IoT costs are at viable per-shelf-label economics
- Snowflake + the new AI tooling makes the digital twin tractable
- The pieces (WhoOnSite tracker, pick rates, replen, labour forecasting) all exist as separate streams — pulling them into one closed loop multiplies the value
- Robotic packing is mature enough for the consistent Maxi (OMX standard carton) form factor

## Data + info to work through tomorrow

### Problem framing (what's broken)

- **DC walk time** dominates pick-time variance. AKL conveyor pickers walk huge distances when product placement doesn't match daily demand.
- **WhoOnSite / Pick rates / Replen / Labour forecasting** are **separate streams of work today**. No closed loop = each one is locally optimised, globally wrong.
- **Manual put-away** errors → wrong-bin stock → pickers can't find product → time + error compounding
- **No digital twin** = layout changes happen by gut, not by simulation
- **Back-strain claims** from manual heavy packs (cost: workers' comp, lost days)

### Memory references I can pull tomorrow

- AKL conveyor pick rates: **flow 180/hr, static 110/hr** per Jeff confirmed 2026-06-09
- Optimal AKL conveyor staffing: **24 pickers (static + flow)** per Jeff 2026-06-10
- Minimum 1 picker per zone (10 zones on main conveyor)
- HITS = PICKS (PICKEVENT) / LINES = order lines (can produce multiple HITS each)
- Dematic PICKEVENT = conveyor only; non-conveyor in Pronto
- DG (dangerous goods) is separate workflow / zone 47
- Customer cutoff = 15:00; carrier cutoffs 18:30 regional / 20:30 metro
- SMALL_DIFOT already at 99% — labour case is cost/efficiency, **not SLA uplift**

### Benefits (the value story)

| Lever | Rough $ value | Note |
|---|---|---|
| **Walk-time reduction** | High | Even 10% walk reduction across AKL conveyor = significant LPPH lift; quantify from PICKEVENT |
| **Error reduction** | Med | Voice checksum + smart routing → fewer mis-picks |
| **Workers' comp / back strain** | Med | Robotic Maxi packing reduces manual heavy lifting |
| **Labour planning accuracy** | High | Live pick data → tighter labour forecast → fewer no-shows + fewer over-staffed shifts |
| **Layout optimisation (digital twin)** | High | Stock placement that matches actual demand patterns = persistent ongoing saving |
| **Box-size waste (CHCH)** | Med | Right-size box → less air shipped → courier $ |
| **Cheap automation philosophy** | Strategic | $50k box-making + $20k shunt vs. $1M+ integrated conveyor = order-of-magnitude cheaper |

### Open questions to resolve tomorrow

- Closed-loop diagram (below) — confirm directionality + arrows
- DC pilot starting zone (which conveyor segment? which CHCH bay?)
- Cheap automation vendor list (box-making robot, shunt robot, e-ink supplier)
- Whether to integrate with existing WMS or sit alongside (Pronto + Dematic)
- Funding model — capex per device, opex per month

## Closed-loop architecture (concept)

```
    Demand signal (Snowflake)
            ↓
    Replen recommendations  ─→  E-ink put-away routing
            ↑                          ↓
    Pick-rate analytics  ←──  Live picker positions (WhoOnSite)
            ↓                          ↓
    Labour forecasting  ←──  Voice checksum confirmation
            ↑                          ↓
    Digital twin  ←──────────  Robotic Maxi packing (end of conveyor)
            ↓
    Recommended layout changes  →  back to replen
```

## Open questions for next session

- Which DC starts where?
  - **AKL conveyor**: WhoOnSite + e-ink + voice checksum + robotic Maxi packing
  - **CHCH DC**: box-splitter + box-making robot + smart barcode labelling
  - Both DCs share: demand → replen → labour forecasting → digital twin loop
- Which sub-component is the POC starter? My guess: e-ink + voice confirmation in one zone, prove the unit-economics
- What's the existing IT stack we extend (Pronto / WMS)? Don't reinvent
- Robotic arms — vendor research needed (ABB, Fanuc, KUKA, Universal Robots)
- Budget? Likely much larger than Ask Max — this is hardware-heavy

## Layout candidates from the gold standard

- **Cover**: WeWork-style — DC floor at first light
- **Problem vector grid (6 or 8)**: Walk time / Manual put-away / Pick errors / Back strain / Labour planning / Layout fixed-by-history / Replen reactive / No closed loop
- **Solution mirror grid (6 or 8)**: One per problem
- **Architecture pipeline**: A LOT of components — WhoOnSite → E-ink → Voice → Replen → Snowflake → Digital twin → Robotic arms
- **Composite + callouts**: One "DC floor" hero image with pins on each technology component
- **Mind-map**: Centred on "The DC" with spokes to each capability
- **How-it-works**: One picker's shift — image-left, 4-step flow right
- **Roadmap (6-stage)**: POC zone → expand → full DC → second DC → robotic arms → digital twin live
- **Ask**: significant ($M+) — show the payback against current labour + error costs

## Cross-links

- See **Ask Max** (Deck 01) for the voice + AI agent layer that overlaps with picker voice confirmation
- See **PPSS** (Deck 02) for the demand/promo planning that feeds the replen-influence loop
- See OMX Lens for the analytics + leaderboard surfaces

## Source references

- WhoOnSite (live staff location) — internal OMX system, in flight
- DC pick rate data — Snowflake `PICKEVENT` table (conveyor-only); Pronto for non-conveyor
- Replen — existing Thursday cadence, see DC Replen project
- Labour forecasting — 10:30am same-day model (canonical, not T+1)
- Digital twin — ongoing exploration with OMX DC Operations

---

## Research deepening (background-agent, 2026-06-28)

### E-ink shelf-label vendors (NZ-shippable)
- **Solum** — $8–12/unit at 5k+ vol · APAC-strong · HTTP API for location-triggered routing
- **SES-imagotag (Pricer)** — $5–9/unit · batch update + colour-flash commands · proven for picker routing
- **Pricer** — $6–11/unit · European leader · conveyor zone mapping hooks
- **Hanshow** — $3–7/unit · aggressive pricing · API-first · best fit for Pronto/Dematic stitch
- **Integration pattern:** WMS webhook → ESL gateway → location flash + colour cue. ~2sec lag.

### Box-making robot vendors (<$50k)
- **CMC Cartonwrap** — $25–35k · 80 boxes/hr · best fit for CHCH bay footprint · 12–18mo payback
- **Packsize** — $35–45k · modular · 18–24mo payback at 500+ picks/day (CHCH-scale)
- **Sealed Air I-Pack** — $40–50k · ~50 boxes/hr · bulky-SKU strong, slower cadence
- **Payback frame:** ~$2–3k/mo courier savings (right-size) + ~$5–8k/mo labour = 8–14mo typical

### End-of-conveyor robotic arm
- **Universal Robots UR10** — $38–55k cobot · 10kg · 3–4 picks/min · no cage needed (AKL-safe)
- **ABB IRB1200** — $50–75k traditional · 7kg · 6–8 picks/min · needs safety cage
- **Fanuc M-10iA** — $45–70k · 10kg · mature integration in APAC food/pharma
- **Integration:** Vision + Snowflake carton dimensions → Tetris algorithm → arm executes → feedback to Dematic/Pronto

### Digital twin DC modelling
- **AnyLogic** — $3–8k/seat/yr · drag-drop · Snowflake ODBC · fastest OMX on-ramp · **recommended starter**
- **AWS Twinmaker** — free tier but $15–25k integration · best if AWS lock-in
- **FlexSim** — $5–15k/seat/yr · C++ depth · best 3D stakeholder demos

### WhoOnSite ↔ WMS stitch (already in flight at OMX)
- **AKL (Dematic):** WhoOnSite → n8n webhook → Dematic API (batch location flush every 10sec)
- **CHCH (Pronto):** WhoOnSite → Snowflake ETL (Fivetran) → Pronto batch sync (hourly) + zone-override
- Location-tracking layer sits independent — polls WMS, emits zone-exit to replen queue, no WMS code change

### Existing OMX files to quantify against
- `D:\20-PROJECTS\officemax\dc-pick-optimization\DC-PICK-WALK-TIME-OPTIMIZATION-BRIEF.md` — AKL baseline 180/hr flow vs 110/hr static, target 15–25% walk-time reduction
- `D:\20-PROJECTS\officemax\dc-pick-optimization\00-DATA-INVENTORY.md` — Snowflake PICKEVENT 90-day pick history ready
- `D:\20-PROJECTS\officemax\dc-labour\02-app\README.md` — **WhoOnSite already deployed prod + UAT**, Gallagher access control integrated, captures NFC zones
- `D:\20-PROJECTS\officemax\lens-subsystems\dc-labour-modeling\README.md` — real-time workforce analytics, ready for LPPH leaderboard
- `D:\20-PROJECTS\officemax\dc-replenishment\docs\SOLUTION-ARCHITECTURE.md` — current replen architecture
- `D:\20-PROJECTS\officemax\dc-labour\02-app\docs\cab-release\01-SOLUTION-DESIGN.md` — labour CAB design

**Critical finding:** WhoOnSite is **already in production**. Deck framing shifts from "build new" to "stitch existing tools into a closed loop" — much faster ROI story.

---

## Research deepening v2 (background-agent, 2026-06-28 — vendor + competitor + OMX-scale)

### OMX DC scale — confirmed from live fixtures (DCReplen Frontend)
Anchors any deck numbers against these, not estimates:

- **AKL DC = 47 zones total** (not 46). Conveyor zones 14-23 (10 zones, static+flow). Bulk zones 25-28. DG zone 47. Confirmed in `D:\20-PROJECTS\officemax\DCReplen Frontend\js\page-dc-labour-modeling.js` line 75-77 + zone classifier line 1442-1443.
- **Optimum conveyor staffing = 24 pickers** combined static+flow, 1-picker-per-zone floor (line 538-539, "jeff-cto-lock" memory reference).
- **Pick rates locked**: static conveyor 110 hits/hr (108 in fixtures), flow conveyor 180 hits/hr (line 683-684, 766).
- **AKL DC physical** = 28 rack rows / ~1,800 bins (Suohling AS-BUILT racking drawing AR00, 22-Jun-2026), source DWG at `D:\00-INBOX\2026-06-23\sterling-racking-cad\OfficeMax_Highbrook_-_Auckland_Racking_Layout.dwg` (22.5MB AutoCAD).
- **Current baseline** = 22 permanent pickers single-shift; proposed = 24 two-shift overlap; **labour saving fixtures = AUD $750,000/yr** (`dcp-fixtures.js` line 136, "Consultation Pack §5").
- **Overtime cost basis** = 8 conveyor pickers running past cutoff at loaded_hourly_rate * overtime_multiplier (line 833-835).
- **Digital twin prototype already exists** at `D:\20-PROJECTS\officemax\digital-twin\prototype\` — Three.js viewer, 28 rack rows generated from Suohling drawing, click-to-inspect, synthetic inventory fallback, ready to swap to Snowflake feed. Run `python -m http.server 8765`. Not a paper concept — there's working code.

### E-ink shelf labels — NZ-relevant detail
- **VusionGroup (SES-imagotag)** — global market leader, 350M units sold 2023 across 45,000 stores in 60 countries. Pick-to-Light flashing-LED workflow proven. No public NZ warehouse case studies but APAC retail footprint is deep (Co-op rollout 2025). Source: [VusionGroup](https://www.vusion.com/products/sesimagotag/electronic-shelf-labels/), [Wikipedia](https://en.wikipedia.org/wiki/VusionGroup).
- **Hanshow** — pick-to-light reported to lift order picking productivity up to 50%. Gateway coverage = 1 gateway per ~1,000-2,000 sq ft retail floor (50,000 sq ft = 25-50 gateways). Bracket ~AKL Highbrook scale. Source: [Hanshow ESL](https://www.hanshow.com/en/solutions/category/electronic-shelf-labels).
- **Cost benchmark (2026)**: ESL units US$5-20 each typical, advanced US$50+. Software/integration US$3k-10k. Implementation support US$2k-8k. Source: [Retail Digitals 2026 ESL Guide](https://retaildigitals.com/electronic-shelf-labels-guide/).
- **At AKL scale (~1,800 bins)** = NZ$25k-65k hardware + ~NZ$15-25k gateways (AKL Highbrook ~14,000 sqm bracket) + NZ$10-20k integration = **~NZ$50-110k for pick-face ESL coverage**.

### Box-on-demand machines — NZ/AU presence confirmed
- **CMC CartonWrap** — **800-1,000 boxes/hour** (one box every 3.5 sec). NZ/AU distributor = **Abbe Corrugated** (Coolaroo, Melbourne) — local demo centre operational. **NZ benchmark deployment: Catch.com.au Truganina VIC** — full integrated CMC line live. Sources: [Abbe](https://www.abbe.com.au/cmc-cartonwrap-makes-peak-season-distribution-a-breeze-for-catch-com-au/), [MHD Supply Chain](https://mhdsupplychain.com.au/2018/03/14/cmc-cartonwrap-makes-peak-season-distribution-a-breeze-for-catch-com-au/).
- **Packsize iQ Fusion 2** — sold in AU/NZ **exclusively through Visy**. Smallest footprint in category; single-operator workflow. Source: [Packsize ANZ via Visy](https://packsize.com/au/iq-series).
- **OMX volume reality check**: CHCH DC is much smaller volume than the 800/hr Cartonwrap warrants. Right-sized fit may be **Packsize iQ3** (entry tier) or a CMC entry-tier machine. The seed brief's "sub-NZ$50k box-making robot" assumption needs vendor quote validation — public pricing not disclosed by either.

### Robotic arm vendors — end-of-conveyor packing
- **Boston Dynamics Stretch** — designed specifically for truck/container unloading, not box-into-pallet packing. Multipick now moves multiple boxes per swing. **Deploys in <1 week**. Cost not publicly disclosed; quoted as package (hardware + end-effector + vision + integration + multi-year support). Source: [Boston Dynamics Stretch Brochure 2025](https://bostondynamics.com/wp-content/uploads/2024/01/Stretch-Brochure-2025-compressed.pdf). **Likely overkill for end-of-conveyor Maxi-packing** — Stretch's sweet spot is truck unload, not Tetris pack.
- **Universal Robots / ABB / Fanuc** cobots remain the right class for the Maxi-packing use case (lower cost, smaller footprint, no cage for cobots). Prior section v1 numbers stand.

### Digital twin platform — recommendation refined
- **AnyLogic** still the recommended starter for OMX scale — **process-driven** modelling (logistics, pick routing, labour) is the right fit, not Omniverse's high-fidelity physics. Source: [ScienceDirect 2026 DT review](https://www.sciencedirect.com/science/article/pii/S2405959526000093).
- **AnyLogic + NVIDIA Omniverse integration is now native** (Omniverse Connector element) — live AnyLogic agent positions feed 3D Omniverse scene. Means OMX can start AnyLogic-only, add Omniverse visuals for board-level demos later without rebuild. Source: [AnyLogic Omniverse integration](https://www.anylogic.com/features/anylogic-omniverse-integration/).
- **FlexSim** has Omniverse integration too but better suited for manufacturing throughput than picker-route modelling.

### Competitor benchmarks — direct + adjacent

**Officeworks AU (direct competitor — same parent group as OMX NZ)**:
- **Derrimut VIC CFC (2021)** — 116 Geek+ solar-powered AMRs + 32 sortation robots, 25,000 SKUs. **Walking reduced from 10-12 km/shift to near-zero**. Integrator: Körber.
- **Perth WA CFC (2023)** — 91 Geek+ AMRs, scalable to 150 via RaaS. 1:1 team-member:robot ratio.
- First solar-powered AMR fleet in Australia.
- Sources: [Office Products News](https://www.officeproductsnews.com.au/newsitem/how-officeworks-created-dc-future), [Geek+ case study](https://www.geekplus.com/case-studies/officeworks), [Körber MHD 2023](https://mhdsupplychain.com.au/2023/10/30/korber-automates-new-officeworks-cfc/).
- **Implication for OMX deck**: Officeworks AU has gone full AMR-replacement-of-walking. OMX deck thesis ("layer cheap smart tech on top of existing conveyor") is a deliberately cheaper, faster path — frame it as the *pragmatic NZ-scale alternative*, not as falling behind.

**Staples (US)**:
- **Locus Robotics AMRs** — pick errors down 73%, cycle time down 70%, picks/hr 2x, training cut from days to minutes.
- **RightHand Robotics piece-picking arms** deployed across multiple FCs (AI-vision suction grip).
- **Manhattan Active WM** consolidating 14 facilities + 2,000+ AGVs/robots.
- Sources: [SCMR Staples Canada](https://www.scmr.com/article/staples-canada-rethinks-its-fulfillment-model), [Supply Chain Dive RightHand](https://www.supplychaindive.com/news/staples-righthand-robotics-automate-fulfillment-centers-warehouses/707942/), [Manhattan](https://www.manh.com/about-us/newsroom/press-releases/staples-transforms-complex-fulfillment-operations-with-manhattan-active-warehouse-management).

**Costco (US)** — autonomous pallet jacks (LiDAR + 3D vision + IMU). AI demand forecasting + real-time inventory. Adding 27-29 warehouses in 2025 with AI/IoT baseline. Source: [Klover.ai Costco AI 2025](https://www.klover.ai/costco-uses-ai-agents-10-ways-to-use-ai-in-depth-analysis-2025/).

**NZ direct benchmarks (most relevant)**:
- **DHL Auckland healthcare DC (2025)** — 41 Geek+ AMRs, **NZ$10M automation in NZ$90M facility**, integrator **Automate-X NZ** (local capability confirmed). 14,100 sqm, 12,000 pallet spaces. Source: [NZ Herald](https://www.nzherald.co.nz/property/dhl-opens-new-90m-automated-medical-healthcare-goods-distribution-centre-at-the-landing/4KSAYSMR4VABVK5V5TQLN4EPVE/).
- **AS Colour AKL DC (2021)** — Dematic Multishuttle GTP, **344% throughput uplift** (45 → 200+ lines/hr/worker). Same Dematic vendor as OMX AKL. Source: [Dematic case](https://www.dematic.com/en-us/insights/case-studies/as-colour-new-zealand/).
- **Foodstuffs SI Hornby (2025)** — NZ$28M automated frozen DC, ASRS crane + shuttle.
- **Element Logic** entered NZ/AU via Safer Storage Systems acquisition Dec 2025 — primary AutoStore integrator now locally present.
- Full vendor + NZ-deployment table at `D:\10-CONTROLLER\jarvis\knowledge\consultant\industry-reports\2026-04-29_cost-benchmarks_robot-coffee_dc-automation_marketplaces.md` (Section B).

### Benchmark cost-per-pick anchors for deck
- **Manual ANZ B2B pick**: NZ$4.50-5.50/pick.
- **Automated GTP** (AS Colour benchmark): 6-10 US cents/pick (~NZ$0.10-0.17).
- **Travel = 50-60% of warehouse labour hours** (Automate-X NZ benchmark).
- **AMR capex**: NZ$40k-230k/unit. **RaaS**: NZ$1.6k-12.3k/robot/month.
- **OMX angle**: 110 hits/hr static vs 180 hits/hr flow at AKL → walk-time = ~39% of static-zone pick cycle. Even 25% walk reduction = static zones approach 140-150 hits/hr without adding pickers.

### WhoOnSite — what it actually is (clarification)
- "WhoOnSite" returns no results as a public-product name — **this is an OMX-internal system**, not a SaaS vendor.
- Memory + seed brief references confirm: NFC-zone capture, Gallagher access-control integration, deployed prod + UAT (no production code findable in current `D:\20-PROJECTS\officemax` snapshot — folder may have moved).
- **For the deck**: treat WhoOnSite as the staff-location data source already in flight; do NOT propose replacing it. The deck story is **stitching WhoOnSite to pick rates + replen + labour forecast + e-ink routing** into one closed loop.
- **Recommended verification before deck delivery**: ask the WhoOnSite owner (likely Jay / IT) for (a) NFC zone count, (b) update frequency, (c) JSON payload schema, (d) where data lands today (Snowflake? Gallagher API? n8n?).

### Updated deck framing recommendation
Three reframing moves vs the v1 brief:

1. **Lead with the closed loop, not the gadgets.** Officeworks AU spent serious capex on full AMR replacement. OMX's edge = the **closed-loop stitch of existing systems** (WhoOnSite + PICKEVENT + replen + labour forecast + Snowflake) plus *targeted* cheap automation overlays. That's a faster-payback, lower-risk story.
2. **Use Officeworks AU as the "what good looks like at scale" benchmark** — then position the OMX play as the NZ-right-sized version: 1 DC at a time, layered overlays, digital twin already prototyped, total spend an order of magnitude less.
3. **Anchor on real numbers from the live fixtures, not invented ones**: 47 zones, 24 optimum pickers, 110/180 hits/hr, ~1,800 bins, NZ$750k/yr labour saving already modelled, NZ$10M = DHL AKL automation benchmark, AS Colour 344% GTP uplift on same-vendor Dematic kit.