All case studies
AI Optimisation
Fulfilment & Distribution · Quick Commerce · Industrial 5 min read

The design space is too large for humans to search. So we built an AI that searches it for them.

AI-Generated Warehouse Layouts
Validated, Selected, and Built

A quick-commerce operator expanding its fulfilment network needed to design multiple new distribution centres under tight timelines — each with different building footprints, product category mixes, and throughput targets. Manual layout design was too slow and too narrow in the options it could explore. We built an AI framework using constrained beam search that automatically generates, validates, and scores hundreds of warehouse layout candidates — giving their design team a ranked shortlist of feasible, optimised options to choose from, rather than a blank canvas to start from.

Note on applicability: The challenges and solution described here apply equally to automotive parts warehouses, spare parts distribution centres, and manufacturing line-side buffer zones — any facility where layout directly affects operational efficiency and the design space is too large for manual exploration.

Client
Quick-commerce fulfilment operator (name withheld)
Core Technology
Constrained Beam Search · Pareto Scoring
Scope
Multi-facility layout optimisation programme
Delivered by
Eclectic Systems

Why Manual Layout Design Breaks Down

As demand for faster fulfilment grows, the limitations of manual and basic automated layout approaches become business-critical constraints — not just operational inconveniences. These are the conditions we found across the client's existing design process.

Inefficient space utilisation

Warehouses must simultaneously maximise storage density and preserve operational accessibility — two objectives that directly conflict. Manual layouts routinely sacrificed one for the other, leaving capacity on the table or creating congestion in high-traffic zones.

Layout decisions driven by habit, not data

Designers relied on experience and precedent. There was no systematic way to evaluate whether a given layout was optimal or simply the first feasible option that came to mind — and no mechanism to compare alternatives objectively.

Manual design doesn't scale

For large fulfilment and distribution centres with thousands of storage locations, the combinatorial design space is intractable manually. Exploring even a fraction of viable alternatives was impractical within any realistic planning timeline.

No adaptability as operations change

Product mix shifts, seasonal demand peaks, and facility expansions regularly invalidated existing layouts. Each redesign started from scratch — expensive, slow, and dependent on the same designer who created the original.

Competing optimisation objectives

Maximising storage capacity, maximising accessible pick faces, minimising average picking distance, and maintaining safe aisle clearances all pull against each other. No single objective captures the full picture — yet manual processes forced a single design pass.

Limitations of classical approaches

Integer linear programming and genetic algorithm approaches suffered from high modelling complexity, poor flexibility when requirements changed mid-design, and prohibitive computational cost at realistic warehouse scales.

The AI Framework We Built

The system combines automated optimisation with human expert judgment — covering the full loop from constraint-driven generation, through multi-criteria scoring, to on-site implementation validation.

Constrained Beam Search Layout Generator

At the core of the framework is a constrained beam search algorithm — a directed search that systematically explores the warehouse layout design space while hard-enforcing spatial constraints at every step. The beam width controls how many candidate layout branches are kept alive simultaneously; wider beams explore more of the space, narrower beams run faster. Spatial constraints — minimum aisle clearances, accessibility requirements, fire egress paths, and storage location rules — are encoded as hard boundaries: any partial layout violating a constraint is pruned from the search immediately, so the algorithm never wastes compute on infeasible branches.

Multiple feasible, constraint-satisfying candidate layouts generated in hours, not weeks.

Multi-Criteria Scoring Function

Each candidate layout is automatically scored across three dimensions: total storage capacity, number of accessible pick faces, and average accessibility cost (a weighted proxy for mean picking distance under expected order profiles). The scoring function surfaces the Pareto-optimal frontier — layouts where no further improvement on one objective is possible without degrading another. This gives designers a principled basis for comparison rather than gut feel.

Objective, ranked evaluation across every generated candidate — no more subjective trade-off calls.

Automated Feasibility Verification

Before scoring, every candidate layout passes through a feasibility checker that validates three conditions: accessibility (every storage location is reachable via a compliant path), clearances (all regulatory and operational minimum distances are met), and aisle connectivity (no dead ends or isolated zones exist). Infeasible layouts are rejected before they reach the scoring layer — and certainly before they reach the designer.

Designers only ever see layouts that will work in practice.

Human-in-the-Loop Designer Interface

Warehouse designers review the ranked candidate layouts, inspect scoring breakdowns per objective, and apply additional operational constraints — product velocity zones, cross-docking requirements, hazmat segregation, or equipment turning radii. They can either select the best-ranked layout or refine candidates further based on domain expertise. The system narrows the field; the expert makes the call.

Expert judgment applied to the best options, not invested in generating them.

On-Site Validation Before Implementation

The selected layout undergoes detailed review by an on-site team before any physical work begins — checking practical feasibility against real equipment dimensions, checking safety compliance against local regulations, and verifying implementation readiness. The framework also notes integration touchpoints with existing WMS, ERP (including SAP), and plant layout tools (AutoCAD, DELMIA) to ensure the selected layout translates cleanly into the client's existing systems.

No surprises at build — every layout is field-validated and systems-checked first.

How a Layout Engagement Works

From objective setting through to a validated, implementation-ready layout — expert judgment is in the loop at every decision point.

01

Layout Objective Definition

Define design goals: storage capacity targets, required access points, navigation efficiency requirements, projected retrieval throughput, and WMS/ERP integration constraints.

02

Automated Candidate Generation

Constrained beam search generates multiple layout alternatives satisfying all physical, operational, and accessibility constraints.

03

Multi-Criteria Scoring & Shortlisting

Each candidate is scored across capacity, access points, and accessibility cost. Top-ranked candidates are shortlisted for designer review.

04

Human-in-the-Loop Refinement

Designers review ranked shortlist, apply operational preferences, and select or further refine the preferred candidate.

05

On-Site Validation & Systems Check

Selected layout reviewed for practical feasibility, safety compliance, WMS/ERP integration readiness, and implementation handover.

Before vs. After

AreaBeforeWith AI Framework
Layout generation time3–6 weeks per design cycleCandidate shortlist in hours
Alternatives evaluated3–5 options from experienceHundreds of feasible candidates scored
Feasibility checkingManual review, often incompleteAutomated constraint verification before scoring
Objective evaluationSubjective trade-off by designerPareto-scored across capacity, access, and navigation
Redesign on changeFull manual restartRe-parameterise and regenerate
Designer roleConstructs layout from blank canvasSelects and refines from ranked, valid options

Measured Outcomes

3 weeks → hours
Layout design cycle time — from initial brief to ranked candidate shortlist
100%
Of shortlisted layouts pass automated feasibility checks before reaching the designer
100s vs. 3–5
Candidate layouts evaluated per engagement versus historical manual practice
Zero redesigns
Post-build redesigns required — layouts that pass on-site validation are built as-selected

The core shift is from “design a layout and check if it works” to “generate every feasible layout and pick the best one.” Warehouse designers spend their expertise on selection and refinement — not on construction — which means better decisions made faster, with full confidence that every option in front of them is physically, operationally, and systemically viable. Storage utilisation improvement data from client deployments is under ongoing measurement and will be published as part of a follow-on update.

Designing or expanding a warehouse or distribution facility?

We can scope an AI layout optimisation engagement for your facility type, operational requirements, and WMS/ERP stack — starting with a free discovery session.