Multi-Echelon Inventory Optimisation (MEIO)

Multi-echelon inventory optimisation (MEIO) treats the entire supply network — factory → central DC → regional DCs → stores or 3PLs — as a single integrated system, jointly computing optimal stock levels, reorder points, and service times across every node simultaneously. It contrasts with single-echelon (siloed) optimisation, in which each warehouse or point of sale independently sets its own inventory parameters, failing to account for interdependencies between nodes. MEIO is the natural extension of Safety Stock Optimisation once a retailer operates multiple distribution tiers.

---

Theoretical foundations

The Clark-Scarf (1960) paper introduced "echelon inventory" — measuring a node's stock position as its own stock plus all downstream stock and in-transit quantities it covers — and proved a decomposition result: optimal base-stock levels can be found by minimising independent one-dimensional convex cost functions at each echelon sequentially, making multi-echelon problems tractable. This remains the foundational result in Clark-Scarf Model theory. (SCIRP review, accessed 2026)

Extensions since 1960 include assembly/convergent structures, fixed batch sizes, replenishment intervals, service-level constraints, advance demand information, and batch transfers. (SCIRP)

A 2025 arxiv paper (2503.18201) proposes Iterative Multi-Agent Reinforcement Learning (IMALR) as an alternative to classical Clark-Scarf heuristics for non-stationary demand environments. (arxiv 2503.18201, 2025-03)

A 2026 arxiv paper (2603.16815) argues that forecast accuracy metrics (MAPE, RMSE) and inventory cost are not perfectly correlated: the best-performing forecast model under standard metrics may not produce the lowest-cost MEIO policy. This challenges the standard MEIO pillar that improving demand forecast accuracy directly improves inventory outcomes. (arxiv 2603.16815, 2026-03)

---

Network topology

MEIO network structures are classified into four types (o9 Solutions, 2026-03-04):

Type	Description
Serial	Each node has one upstream and one downstream connection
Assembly	Each node has one downstream, multiple upstream connections
Distribution	Each node has one upstream, multiple downstream connections (e.g. central DC → regional DCs)
General	Nodes can have multiple connections both upstream and downstream

Ecommerce and omnichannel retail typically map to the distribution type. A real automotive manufacturer supply chain reportedly contained approximately 60 echelons. (ASCII/Lexia MEIO webinar, ~2018)

---

Core mechanics

MEIO calculates target stock levels, reorder points, and order-up-to quantities for every item at every location using shared inputs: demand variability, lead times, service-level targets, and transportation constraints. (Interlake Mecalux, 2026-02-20)

Service Time Optimisation is the key optimisation variable: the period (in days) between order receipt and order dispatch at each node. The MEIO solver simultaneously optimises service times across all nodes to minimise total holding cost. (ASCII/Lexia, ~2018)

A 2024 academic lecture on a five-stage network demonstrated two fundamental propagation rules (Five-Stage MEIO Example, YouTube, 2024-11):

• Demand variance propagates backwards (upstream): a node's variance equals the sum of the downstream variances it serves
• Cumulative holding cost propagates forwards (downstream): a node's cost equals its own added value plus all upstream costs

The counter-intuitive placement result

Full enumeration of all service-time configurations in the 2024 five-stage example produced (Five-Stage MEIO Example, YouTube, 2024-11):

Policy	Total cost
Push all stock to furthest downstream stages (4–5)	42.17 (worst)
No-stock baseline (zero service time everywhere)	~34.42
Optimal: stock at upstream stages 1–2 AND downstream stages 4–5	32.17 (best)

The optimal solution held stock at both upstream and downstream nodes, not at the intuitive middle decoupling point. Pushing all stock closest to customers produced the worst outcome. This directly contradicts the naive "keep inventory close to customers" heuristic.

Risk pooling and demand aggregation

Demand that is 97% "low variability" at the customer level may shift to 32% erratic and 68% smooth once aggregated to facility level, enabling smaller safety stocks at higher echelons. (ASCII/Lexia, ~2018)

---

Performance benchmarks (as-of 2026-06-22)

[!unverified] The widely-cited 15–30% inventory reduction figure traces to unattributed or vendor whitepaper origins (logicamatrix.com, mcpanalytics.ai) — no named primary study was found. Treat as directional only.

Sourced case studies:

• Australian manufacturer/distributor: Holding safety stock at non-DC locations reduced total safety stock cost from AUD 13.5M to AUD 10.9M — ~19% reduction at the same 97% service level. (ASCII/Lexia, ~2018)

• Retail company case study (ASCII/Lexia, ~2018): MEIO showed the same 80% service level could be maintained with 25% less safety stock, or 90% SL achieved with only ~5% more working capital, with AUD 75M in prevented lost sales.

• Caterpillar: 50% scrap reduction after MEIO implementation, attributed to better customer forecast visibility. (o9 Solutions, cited via ShipBob, 2026-04-20)

[!unverified] o9 is the origin; verified second-hand only. Manufacturing context — not retail.

Lead time variability is the dominant safety stock driver

In the ASCII/Lexia case analysis (~2018): upgrading service level from 97% to 99% increased safety stock cost by ~5%. Doubling transport time variability (standard deviation) increased safety stock cost by ~37%. This has direct implications for fashion retailers with variable international lead times: investing in Lead Time Variability reduction (e.g. closer sourcing, buffer stock at origin) yields more safety stock benefit than tightening service level targets.

---

Fashion and ecommerce applications

MEIO directly applies to the omnichannel fashion network topology: factory → central DC → regional DCs → stores / 3PLs. It replaces the siloed model where some stores accumulate slow-moving items while others stockout, by computing reorder points and safety stock per node based on demand variability and transit times. (Interlake Mecalux, 2026-02-20)

Fashion-specific modelling approach: use deterministic planning for basics and proven bestsellers (predictable replenishment cadence); apply stochastic MEIO for new, seasonal, or high-variability items. (The Retail Exec, 2025-05-06, updated 2026-02-24)

MEIO is especially relevant where lead time variability is high — typical in fashion with international sourcing — because lead time variance dominates safety stock cost far more than service level targets.

Same-Day Delivery and Click and Collect each add a downstream echelon (dark stores, ship-from-store nodes) with different demand variability and lead time profiles from a regional DC — extending the MEIO network scope.

New Product Forecasting is the unresolved cold-start gap: MEIO requires demand history to calculate variability parameters; new fashion SKUs launched without history cannot be classified or optimally positioned.

Size Curve Optimisation extends MEIO into the fashion variant dimension: safety stock must be allocated not just across network echelons but across size variants within each echelon.

---

Implementation challenges

Three core structural barriers (o9 Solutions, 2026-03-04):

1. Data management — real-time inventory data must flow across all tiers; managing separate inventory policies per echelon creates coordination and security challenges
2. Supplier coordination — all suppliers must use compatible systems, difficult for large multi-supplier networks
3. Legacy system integration — siloed ERP and WMS systems are incompatible with MEIO data requirements

Vendor-estimated timeline: 9–12 weeks for software go-live (Streamline/GMDH, 2026-06-11). This likely understates the full change management and data governance workload documented by practitioners.

MEIO requires continuous operation, not a one-off project. Parameters need monthly or quarterly review; static models leak value. (IBF Journal / demand-planning.com, 2024-02)

Complexity management at scale

In a 60-location, 13-echelon company, the implementation team compressed ~200,000 item/location combinations into ~300 segments. Planners review one safety stock factor per segment rather than every individual SKU/location pair — without this compression, the model was predicted to be abandoned. (IBF Journal / Maarten Driessen PhD, demand-planning.com, 2024-02)

---

Change management — the dominant failure mode

[!unverified] The "60–80% of supply chain planning implementations fail to achieve projected benefits" figure is cited by Procurement Insights (2025-11) without a named primary study.

Practitioners consistently identify change management — not technology — as the primary failure mode. (IBF Journal / demand-planning.com, 2024-02; Arkieva/Desmet, 2016, updated 2025-12)

Incentive misalignment is the structural killer: A CFO sees total network inventory fall ~30%. A regional DC manager — given a target to reduce local inventory 10–15% — may see their local stock increase 66% as safety stock is repositioned to their node. This contradiction kills implementations. (Arkieva / Bram Desmet PhD, 2016, updated 2025-12)

Planner acceptance rate is the single most predictive KPI. If planners override system proposals, the implementation fails. Most failures occur because software doesn't model real-world constraints (full container shipments, seasonality, finite capacity, slow movers), so planner trust never forms. (IBF Journal / demand-planning.com, 2024-02)

Recommended principles (IBF Journal, 2024-02):

• Start with right-sizing not reduction as the goal
• Use simulation not closed-form algorithms — enables gradual complexity increases and explainability
• Implement in small steps (e.g. reduce safety stock 100 units/month, not slash to target overnight)

---

Vendor landscape (as-of 2026-06-22)

Enterprise:

• o9 Solutions — 2025 Gartner Peer Insights Customers' Choice (Supply Chain Planning); o9 self-reported (Gartner report not independently fetched)
• Blue Yonder / Luminate Platform — Gartner 4.5★ / 255 reviews; positioned for retail forecasting; user sentiment: outdated UI, training-intensive, expensive support; ransomware attack November 2024 disrupted multiple clients (now a vendor selection risk factor); typical MEIO implementation 12–18+ months, ROI realisation 18–36 months
• RELEX Solutions — Gartner 4.8★ / 84 reviews; AI-native, markets "Rebot" agentic AI (2026); Lokad competitor review flags that quantitative logic is "not broadly inspectable" — a planner trust risk
• Manhattan Associates, e2open, OMP Unison Planning, Anaplan — enterprise suite alternatives

Mid-market:

• Streamline / GMDH — discrete-event simulation (not static formulas); free tier; 200+ partners; vendor performance claims unverified
• ToolsGroup — demand sensing and inventory optimisation
• Slimstock — European mid-market
• Arkieva — MEIO specialist with practitioner-cited thought leadership
• Deposco — fulfilment-adjacent planning

No independent pricing or TCO benchmark was found. Enterprise vendors (o9, Blue Yonder, RELEX, Manhattan) do not publish pricing. (as-of 2026-06-22)

---

Key terms

Term	Meaning
Echelon inventory	A node's on-hand stock plus all downstream stock and in-transit quantities it is responsible for covering
Service time	The quoted response time between order receipt and order dispatch at a given node (days)
Coverage	The net risk period that a node's safety stock must cover
Demand Pooling	Aggregation of variable individual-location demand into smoother combined demand at a higher echelon
Decomposition result	Clark-Scarf proof that multi-echelon optimisation can be solved by sequentially minimising one-dimensional convex functions at each echelon
Planner acceptance rate	Proportion of system-generated safety stock proposals implemented without override — most predictive KPI for MEIO success

---

Frontier links

• Clark-Scarf Model — foundational 1960 decomposition result
• Echelon Stock Policy — the base-stock reorder policy type underlying MEIO
• Service Time Optimisation — the solver variable MEIO minimises across the network
• Demand Pooling — the risk aggregation mechanism enabling upstream inventory reduction
• Lead Time Variability — dominant safety stock cost driver in MEIO (37% impact vs 5% for service level change)
• Demand Sensing — short-horizon demand signal feeding into MEIO parameters
• New Product Forecasting — cold-start gap for fashion seasonal SKUs
• Size Curve Optimisation — fashion-specific MEIO challenge across size variants
• DDMRP (Demand Driven Material Requirements Planning) — pull-based replenishment alternative
• Inventory Optimisation Software — vendor landscape deserves standalone page