Supply Chain Optimization Model
End-to-end supply chain redesign reducing average lead time by 34% across a six-node distribution network — delivered as a board-ready briefing and twelve-month implementation roadmap.
Situation · Complication · Resolution
Situation
A mid-market industrial components manufacturer was operating a six-node distribution network developed incrementally over an eight-year period of organic growth. What had begun as a single regional warehouse had expanded, through acquisition and market entry, into a geographically dispersed configuration: two domestic manufacturing facilities, three regional distribution centres, and a national retail fulfilment operation. The network had never been designed — it had accumulated.
By the time the engagement began, the firm's CFO had identified three persistent operational pain points: lead times to key retail accounts were running at 18.2 days against an industry benchmark of 11–13 days; finished goods inventory was running at 47 days of cover against a target of 28–32 days; and service levels, measured as order-to-invoice fill rate, had declined to 91.4% from a high of 96.1% two years prior. The aggregate cost of this underperformance — in lost sales, carrying costs, and expediting spend — was estimated internally at $2.3M annually.
Complication
Six months of transactional data existed across the firm's ERP system, but it had never been cleaned, reconciled against physical stock records, or modelled as a continuous flow. Three separate product families moved through the network on different replenishment triggers: one using a reorder-point model, one on manual periodic review, and one on a vendor-managed inventory arrangement that had not been renegotiated in four years.
The result was a system optimised locally — at each node — but producing chronic inefficiency at the network level. Demand signals from retail were not reaching manufacturing in useful form, safety stock parameters had not been recalibrated since the third node was added, and the VMI arrangement was, in practice, creating forward accumulation of low-turn SKUs at the regional distribution centres.
Resolution
We built a network flow optimisation model using six months of cleaned transactional data, structured around a linear programming formulation that treated the entire six-node system as a single optimisation unit. Demand signals from point-of-sale were reintegrated into the replenishment cadence, safety stock was recalculated using a service-level-driven approach calibrated per SKU velocity tier, and the VMI arrangement was restructured around a consignment model with weekly review.
The recommendations were presented to the CFO in a single working session, alongside a twelve-month phased implementation roadmap. The client chose to implement Phases 1–3 internally, with our team providing model validation support through the live transition.
Distribution Network Architecture
Six-node network as modelled. Arrows indicate primary flow direction; thickness indicates volume weighting post-optimisation.
Before / After Comparison
| Metric | Baseline | Post-Optimisation | Change |
|---|---|---|---|
| Avg Lead Time | 18.2 days | 12.0 days | −34.1% |
| Inventory Days Outstanding | 47 days | 29 days | −38.3% |
| Stockout Rate | 8.2% | 2.1% | −74.4% |
| Cost per Unit Shipped | $4.60 | $3.85 | −16.3% |
| Order Fill Rate | 91.4% | 97.8% | +6.4 pp |
| On-Time-In-Full (OTIF) | 83.2% | 94.6% | +11.4 pp |
Results measured at 90 days post-implementation of Phases 1–3. Phases 4–6 implementation ongoing at time of publication.
Optimisation Approach
The model was built around a mixed-integer linear programming (MILP) formulation treating the network as a unified system. Objective function: minimise total landed cost subject to service-level constraints. Four analytical workstreams ran concurrently before convergence into the final optimisation run.
Data Reconciliation & Cleansing
Six months of ERP transaction data was extracted, deduplicated, and reconciled against physical cycle counts. 14.3% of records required correction before the dataset was fit for modelling.
Demand Signal Reconstruction
Point-of-sale data from retail partners was reintegrated into the supply model using a 12-week rolling demand-sensing window, replacing the static monthly forecasting cadence.
Network Flow Modelling
A MILP formulation was solved using branch-and-bound optimisation across 847 SKUs, six nodes, and four distribution lanes. Total model runtime: 4.2 hours on standard hardware.
Safety Stock Recalibration
Safety stock levels were recalculated per SKU using a service-level-driven approach (target: 97.5th percentile fill rate) segmented by velocity tier (A/B/C classification across 847 SKUs).
Engagement Phases
Stakeholder interviews, ERP data extraction, network mapping and KPI baseline setting.
Data cleansing, demand pattern analysis, root-cause identification across all six nodes.
MILP formulation, parameterisation, and initial solve across full SKU set.
Monte Carlo sensitivity analysis across demand variance and lead-time shock scenarios.
CFO presentation: findings, model outputs, phased implementation roadmap.
Model validation, go-live support, and 90-day post-implementation review.
Stakeholder interviews, ERP data extraction, network mapping and KPI baseline setting.
Data cleansing, demand pattern analysis, root-cause identification across all six nodes.
MILP formulation, parameterisation, and initial solve across full SKU set.
Monte Carlo sensitivity analysis across demand variance and lead-time shock scenarios.
CFO presentation: findings, model outputs, phased implementation roadmap.
Model validation, go-live support, and 90-day post-implementation review.