Last-Mile Delivery Optimisation in Grocery Ecommerce

The Last-Mile Grocery Challenge

Grocery ecommerce is fundamentally different from other ecommerce. When a customer orders a book or a gadget, delivery in 3-5 days is acceptable. When a customer orders groceries, they expect delivery in hours, not days. The perishability of products, the weight of orders, and the geographic complexity of delivery create unique challenges.

At Coles, managing a $1.1B ecommerce business across 800+ storefronts meant solving last-mile delivery at scale. This wasn't about delivering as cheaply as possible—it was about delivering fast, reliably, and profitably to hundreds of thousands of customers across Australia.

The core challenge: last-mile delivery costs are typically the highest component of ecommerce logistics cost, often 50-70% of total delivery expense. For grocery, that's exacerbated by product perishability (you can't leave fresh produce sitting for 6 hours) and order weight (grocery orders average 8-12 kg, compared to 2-3 kg for general retail). Traditional courier models don't work for grocery. You need speed, reliability, and temperature control.

I inherited a delivery network that was functional but expensive. Same-day delivery was rare. Delivery time windows were broad (6-hour windows). Costs were eating margins. Customer satisfaction with delivery quality was moderate. We needed to fundamentally redesign how we thought about last-mile delivery.

Partnerships vs In-House: Making the Strategic Choice

The first decision: should we build our own delivery fleet or partner with existing delivery services?

The case for in-house:

• Complete control over customer experience and delivery timing
• Long-term scalability and no dependency on external partners
• Better margins (eventually) as you optimize operations
• Brand control—your team, your vehicles, your service standards

The case against in-house:

• Massive capital investment (vehicles, warehouses, workforce infrastructure)
• Significant operational complexity (staffing, vehicle maintenance, fuel, routing)
• Limited flexibility during demand spikes or geographic expansion
• Retail experience, not logistics expertise—we'd be starting from scratch
• Long time to profitability; you're burning cash for years

We chose a hybrid model: we didn't build a full delivery fleet, but we didn't entirely outsource either. Instead, we developed strategic partnerships with specialized delivery providers—each bringing different strengths.

Selecting Delivery Partners: Criteria and Trade-offs

We evaluated and partnered with three primary delivery services, each serving different use cases:

UberEATS: Fast, on-demand, gig economy-based delivery. Uber's strength is speed and coverage—they operate in dense urban areas and can achieve 30-45 minute delivery windows. The trade-off: drivers aren't trained on perishable handling, temperature control is inconsistent, and quality can be variable. We used Uber for orders in high-density areas where speed was the priority and order complexity was lower.

Airtasker: Task-based delivery marketplace. Airtasker allows businesses to post specific tasks (grocery delivery) and local contractors bid to fulfill them. Strength: flexibility and local presence. Weakness: less control over driver training and quality. We used Airtasker for suburban areas with moderate density where we needed reliability but could accept longer delivery windows (2-3 hours).

Deliveroo: Logistics platform with focus on restaurant delivery but expanding to grocery. Deliveroo had temperature-controlled options and better driver training than Uber. They were more expensive but more reliable for fresh products. We used them strategically for cold-chain products (dairy, meat, frozen) and in areas where reliability was more important than speed.

The key insight: don't try to have one delivery partner do everything. Different partners excel at different geographies, order types, and delivery windows. By strategically allocating orders based on delivery requirements and partner capabilities, we reduced cost per delivery by 18% while maintaining service quality.

Evaluation Criteria: When selecting delivery partners, consider:

• Geographic Coverage: Do they serve your current markets? Your future markets?
• Speed Capability: Can they achieve your target delivery windows?
• Temperature Control: For perishables, this is non-negotiable. Can they maintain cold chain?
• Driver Training: Are drivers trained on handling fragile items, perishables, and customer service?
• Real-time Visibility: Can you track deliveries in real-time? Can customers track?
• Cost Structure: How do they price? Per-delivery fixed cost? Per-mile? Surge pricing?
• Reliability: What's their on-time delivery rate? Their cancellation rate? Their customer satisfaction score?
• Scalability: Can they grow with you? Can they handle 10x volume growth?
• Integration: How easily do they integrate with your systems (order management, customer communication)?

We created a scorecard for each provider and reviewed quarterly. Partners that consistently underperformed on reliability or customer satisfaction were either improved or replaced. This discipline ensured we maintained high service standards.

Reducing Delivery Time: From 2 Hours to 45 Minutes

When we started, standard delivery was a 2-hour window (e.g., "between 2-4pm"). This was operationally easy but customer experience was poor—customers had to stay home for a 2-hour window. We committed to a 45-minute delivery window as our standard. This required fundamental operational changes.

Optimization #1: Dark Store Network

Traditional grocery ecommerce fulfills from retail stores—a team picks orders from shelves used by in-store shoppers. This creates operational complexity and inventory conflicts. A faster model is dark stores (also called fulfillment centers or micro-fulfillment centers)—dedicated facilities optimized purely for online order picking.

Dark stores are smaller than retail stores (typically 2,000-3,000 sq meters vs 10,000+ for a retail store), located strategically closer to customer density, and organized for picking efficiency rather than customer shopping. A order that takes 20 minutes to pick in a retail store might take 5-7 minutes in an optimized dark store.

Coles had 800+ retail storefronts. We didn't build 800 dark stores—that would be cost-prohibitive. Instead, we strategically built dark stores in 15-20 highest-density population areas, paired with traditional store fulfillment in suburban areas. This hybrid model achieved the speed benefits of dark stores (45-minute delivery in major cities) while maintaining economics in lower-density areas.

Optimization #2: Predictive Ordering and Demand Forecasting

Speed requires knowing what to pick before the customer orders. We implemented demand forecasting that predicted which products would be ordered in each dark store on each day. This allowed us to pre-position inventory and stage products for picking, reducing picking time further.

Optimization #3: Routing and Geolocation

With multiple delivery partners, we needed intelligent routing. When an order came in, we'd assess:

• Delivery location (urban core, suburban, rural)
• Order weight and product requirements (cold, fragile, standard)
• Real-time availability of delivery partners
• Customer delivery window preferences

We'd then automatically route to the delivery partner best suited for that order. This maximization problem required data science and logistics optimization, but the payoff was significant: faster delivery, lower cost, and better partner utilization.

Optimization #4: Order Batching and Consolidation

Delivery partners optimize when they're delivering multiple orders on the same route. We built systems to batch orders intelligently—grouping orders by geography and delivery partner, then handing them off as batches rather than individual orders. This reduced delivery cost per order by 22% while maintaining 45-minute delivery windows.

The Result: 35% delivery time reduction. We moved from an industry-standard 2-hour delivery window to a 45-minute window as default in major cities, with same-day delivery available on-demand. Customer satisfaction with delivery improved from 76% to 89%.

Dark Store Operations and Merchandising

Dark stores are different from retail stores, and merchandising needs to reflect that difference. In a retail store, a customer browses; merchandising is about visual appeal and impulse purchase. In a dark store, a picker receives a digital order and retrieves items; merchandising is about operational efficiency and accuracy.

Physical Layout: Dark stores are organized by product category in a linear flow designed for picking efficiency. Dry goods, produce, dairy, meat, and frozen are arranged to minimize picker movement. In contrast, retail stores optimize for customer journey and cross-selling.

Inventory Depth: Dark stores stock fewer SKUs than retail stores but higher quantity per SKU. A retail store might stock 50 types of apples; a dark store stocks 5-8 high-velocity varieties. The inventory depth reflects demand forecasting rather than customer choice breadth.

Range Selection: Not all products are picked from dark stores. Our dark store range selection was driven by demand data—we stocked the products with highest online velocity, excluding items with low online demand. This meant a dark store might stock 10,000 SKUs vs 30,000+ in a retail store.

Quality Standards: Picking for online delivery requires different quality standards. Pickers must be trained to select optimal ripeness for produce, appropriate expiration dates for perishables, and undamaged products. We implemented training programs and quality audits to ensure standards were met.

Staff Incentives: We tied staff bonuses to order accuracy, speed, and customer satisfaction, not just volume. A picker who fulfilled 50 orders with 98% accuracy and high customer satisfaction was rewarded more than a picker who fulfilled 60 orders with 92% accuracy.

The dark store network was operationally complex—it required different staffing models, inventory systems, and quality controls than retail. But the operational efficiency gains justified the complexity. A dark store could fulfill 3-4x more orders per day per square meter than a retail store.

Cross-Functional Coordination: Online + Store Teams

Here's a challenge that most retail companies underestimate: online and store operations have competing interests and competing teams.

Store teams care about in-store customer experience: product availability on shelves, shelf life, customer browsing experience. Online teams care about online customer experience: fast delivery, accurate orders, low cost. When you fulfill online orders from retail stores, these interests conflict.

I led a cross-functional team of 20+ people (including Accenture offshore partners) to manage this tension. The team included:

• Store Operations: Managing fulfillment from retail locations, coordinating with store managers
• Dark Store Operations: Managing dedicated fulfillment centers
• Logistics and Delivery: Partner management, routing, delivery performance
• IT and Systems: Order management systems, routing algorithms, dark store WMS (warehouse management system)
• Merchandising: Range selection, inventory depth, replenishment strategies
• Supply Chain and Planning: Demand forecasting, inventory management, stock allocation
• Customer Service: Managing delivery issues, returns, refunds

Weekly Operations Meetings: We held weekly syncs where all functions reviewed metrics and issues. Store teams would surface inventory conflicts (online orders exhausting in-store stock). Logistics would surface delivery partner issues. Merchandising would propose range changes. IT would raise system constraints. We'd collectively solve problems rather than allowing functions to work in silos.

Unified Metrics: Success metrics were shared across functions: customer satisfaction, on-time delivery rate, order accuracy, profitability per order. No team optimized locally at the expense of overall performance.

Inventory Allocation: One of the most contentious issues: how to allocate inventory between online and store? We implemented rules-based allocation: physical stores got inventory to maintain in-store shelf stocking; online orders were fulfilled from remainder. When demand spikes pushed online orders to constrain store inventory, we raised prices or limited online availability rather than shortchanging store experience. This required trade-off discussions but ensured long-term health of both channels.

Technology as Coordinator: Ultimately, coordinating 20+ people across functions at scale requires automation. We implemented:

• Automated demand forecasting that predicted daily demand by store, by category
• Inventory allocation systems that automatically reserved inventory for online or store based on rules and forecasts
• Exception management dashboards that flagged conflicts or issues early
• Automated replenishment that coordinated dark stores, retail stores, and suppliers

Technology didn't replace the cross-functional team—it enabled them to focus on strategic decisions rather than manual coordination.

Scaling Across 800+ Storefronts

Coles had 800+ retail storefronts across Australia. Scaling last-mile delivery and online fulfillment across 800 locations is a massive undertaking. Here's how we approached it:

Phase 1: Major Cities (Months 1-6)

We didn't launch everywhere simultaneously. We started with 5 major metro areas (Sydney, Melbourne, Brisbane, Adelaide, Perth). This allowed us to validate the operating model, work out issues with store teams and delivery partners, and establish standard operating procedures. Each phase took 6-8 weeks including staff training, system setup, and soft launch.

Phase 2: Secondary Markets (Months 6-12)

With major cities stable, we expanded to secondary cities and high-density suburban areas. By now, we had replicable processes, trained teams, and proven systems. Rollout accelerated.

Phase 3: Regional and Rural (Months 12-24)

Expanding to regional and rural areas required different economics. Delivery times are longer. Delivery partners are less available. Demand is lower. We adapted: delivery windows were longer (2-3 hours instead of 45 minutes), fulfillment was primarily from retail stores (not dark stores), and we partnered with local logistics providers in addition to national ones.

Standardization vs Flexibility: Across 800 locations, standardization was critical for scalability. Standard systems, standard processes, standard training materials. But 800 locations also demanded flexibility—each store has unique demand patterns, staff capabilities, and local market dynamics. We maintained a core standard playbook but allowed local customization where needed.

Continuous Improvement: With 800 storefronts, someone's always experimenting with something new. We captured learnings from high-performing stores and scaled them. One store found a more efficient picking process—we documented it and rolled it out to 50 similar stores. Another store found a better delivery window that improved customer satisfaction—we tested it in a few similar locations and scaled if successful.

Local Store Manager Engagement: Store managers were initially concerned about online orders disrupting in-store operations. We brought them into the decision-making process. We showed them data on profitability of online orders. We involved them in designing store layouts to minimize online-store conflict. Many store managers became champions of online fulfillment because they saw how it contributed to store profitability and kept jobs secure.

Scaling to 800 storefronts took 24 months. By the end, we had built a network that could fulfill 50,000+ orders per week across 800 locations with 45-minute delivery windows in major cities, a 35% reduction from starting delivery times, and a profitability per order that was competitive with retail margins.

Conclusion: Last-Mile Success is a System

Last-mile delivery optimization isn't one thing—it's a system of decisions, partnerships, operations, and technology working together. Choosing the right partners, building dark stores in strategic locations, implementing intelligent routing, coordinating across functions, and scaling systematically are all necessary.

The companies that will win in grocery ecommerce aren't those that obsess over one element. They're those that optimize the entire system: product sourcing, inventory management, fulfillment operations, delivery partnerships, customer communication, and returns handling all working in concert.

At Coles, we achieved a 35% delivery time reduction and maintained profitability while building the largest online grocery business in Australia. That wasn't because we had the best delivery partner or the biggest dark store network. It was because we solved last-mile delivery as a complete system.