How Dark Store Routing Shapes Network Economics for North American Retailers

A VP of Supply Chain at a North American grocery retailer reviews the dark store network plan with the executive team. Real estate has identified eleven candidate sites across two metropolitan areas. Network design consultants have modelled location coverage and inventory distribution. The technology team has started evaluating warehouse management and pick automation. Finance has built unit economics for each site. The capital request is ready for the board.

The routing layer that will actually run the network — store-to-order assignment, multi-origin dispatch, inventory-aware decisioning, pick-capacity management, driver allocation, and surge response — is treated as a downstream implementation detail.

That is the common pattern in North American dark store planning. It leaves network economics exposed.

Dark store conversations focus heavily on location selection and inventory distribution. They should. Both are hard, capital-intensive decisions supported by specialist systems and consulting practices. But once the network goes live, economics are shaped order by order: which store fulfils the order, which driver is dispatched, whether the order can be batched, whether the store can pick within the promise window, whether traffic conditions threaten the SLA, and whether a nearby node should absorb demand before a store saturates.

Operators that treat dark store routing as a first-class design input are better positioned to protect delivery promises, reduce avoidable miles, improve route density, control cost-to-serve, and keep exception handling out of manual escalation queues.

This is a strategic guide for North American VP Supply Chain leaders, Heads of E-Commerce Operations, and Heads of Logistics evaluating dark store network strategy and the wider retail distribution architecture behind fast fulfilment.

The North American Dark Store Reality

Before discussing routing, market context matters.

The fifteen-minute delivery category that dominated dark store conversations in 2021–2022 has largely consolidated. Pure-play q-commerce operators including Gorillas, Jokr, Buyk, and Getir exited or scaled back US operations through 2022–2024; Gopuff retreated significantly from earlier expansion. The category survived, but the model changed.

The dominant North American fulfilment model in 2026 is one-to-two-hour grocery and convenience delivery: Amazon Fresh, Whole Foods, Instacart partnerships, big-box programmes from Walmart, Kroger, and Target, and surviving q-commerce operators that moved towards longer SLA windows. This is where hyperlocal routing for urban same-day delivery becomes central to network economics.

In-store fulfilment — store associates picking from regular store shelves — accounts for the majority of fast grocery delivery in the US. Pure dark stores still exist, but usually as one channel in a broader fulfilment network.

According to Bain & Company research on retail fulfilment economics, the strategic question for most North American retailers is not simply whether to operate dark stores. It is how to optimise the mix between:

• in-store fulfilment,
• pure dark stores,
• regional micro-fulfilment,
• centralised distribution,
• owned fleet,
• third-party carriers,
• marketplace delivery partners,
• and gig delivery capacity.

Dark store networks live inside this mix. Routing determines how well they perform inside it.

Recent market data reinforces why dark store routing deserves executive attention:

• Fact.MR estimates the global dark store market at USD 36.57 billion in 2026, with projected revenue of USD 719.14 billion by 2036.
• Dark stores and similar rapid-fulfilment formats processed about 9–12% of total online grocery orders in North America in 2025, compared with less than 5% in 2022.
• Future Market Insights estimates groceries accounted for 53.2% of global dark store revenues in 2025, making grocery the largest product category for dark stores.

These figures do not mean every retailer should build a pure dark store network. They mean that for retailers already operating dense, fast-delivery fulfilment models, routing decisions increasingly affect the economics of a material fulfilment channel.

Also Read: Delivery Under 2 Hours: How Quick Commerce Leaders Can Scale Fulfillment

Static routing vs dynamic dark store routing

Decision area	Static or rules-based routing	Dynamic dark store routing
Store-to-order assignment	Assigns by zip code, fixed service area, or nearest store	Assigns using inventory, pick capacity, driver capacity, delivery promise, batching potential, and cost-to-serve
Inventory visibility	Uses periodic inventory updates	Uses near real-time inventory signals from WMS, OMS, or inventory systems
Dispatch	Each store dispatches independently	Optimises routes across multiple origins and fleet types
Capacity	Assumes stores can absorb demand within their catchment	Treats pick capacity, staging capacity, and fleet capacity as constraints
SLA adherence	Reacts after delays appear	Reoptimises before capacity or traffic risk turns into SLA failure
Surge response	Manual escalation and ad-hoc reassignment	Automated diversion, batching, fleet allocation, and route reoptimisation
Cost control	Limited visibility into cost per order by assignment decision	Optimises against cost-to-serve, miles per order, utilisation, and service promise

Five Routing Decisions That Shape Dark Store Economics

1. Store-to-Order Assignment

In a multi-store metro, every order creates an assignment decision: which dark store should fulfil it?

The default answer — assign by zip code, fixed territory, or closest store — is usually too blunt. The closest store may be out of stock, pick-capacity constrained, short on drivers, or poorly positioned for batching with other active orders.

A stronger store-to-order assignment model considers:

• SKU availability by store,
• inventory confidence and substitution risk,
• remaining pick capacity in the promise window,
• driver availability and vehicle type,
• batching opportunities with nearby orders,
• current traffic and access constraints,
• customer SLA tier,
• cut-off times,
• delivery density,
• and cost-to-serve by fulfilment option.

This is where how AI route optimization works becomes economically relevant. Route optimisation is not only a transport planning tool; it is a node-selection, promise-protection, and cost-control mechanism.

A zip-code rule might send an order to the closest dark store. A dynamic assignment engine may send it to a slightly farther store if that store has the full basket, spare pick capacity, a driver already routed towards the customer cluster, and a lower expected cost per successful delivery.

The difference between basic and intelligent assignment in dense metros is real and measurable in delivery cost per order, on-time delivery, and exception handling. According to McKinsey & Company research on grocery delivery economics, last-mile cost variation across operational decisions can produce significant per-order economic differences, particularly at scale.

2. Multi-Origin Dispatch Routing

A network of fifteen dark stores creates fifteen origin points, not one.

That changes the routing problem. Each dark store has its own:

• available inventory,
• pick queue,
• staging capacity,
• driver pool,
• delivery density,
• service area,
• order mix,
• and local congestion pattern.

But customers experience the network as one brand. Finance sees one cost-to-serve model. Operations is accountable for one SLA.

That is why multi-origin dispatch routing is not the same as running fifteen separate store-level routing plans. It requires assignment and route optimisation across the metro as a system.

Most legacy routing systems were architected for single-origin operations: one distribution centre, one route plan, one driver pool, one dispatch cut-off. Dark store routing requires a multi-origin Vehicle Routing Problem (VRP), where the platform must decide both:

1. which origin should fulfil the order, and
2. how that order should be sequenced into a route.

The multi-origin VRP is computationally harder than single-origin VRP. Routing engines built natively for this model produce materially different outcomes from systems retrofitted through configuration.

In practical terms, the difference shows up in:

• route density,
• driver utilisation,
• miles per order,
• dispatch productivity,
• SLA adherence,
• store workload balance,
• and the number of manual overrides needed during peaks.

3. Inventory-Aware Routing

Routing decisions need real-time inventory visibility by store.

When an order arrives for a specific basket, the system needs to know which stores have the required SKUs, which have enough inventory confidence to avoid substitutions, which have pick capacity, and which can fulfil within the promise window at the lowest operational cost.

This has to happen at the moment of routing — not against inventory data refreshed four hours ago.

Many dark store operations still run inventory updates on cycles measured in hours rather than minutes. Routing against stale inventory creates visible customer and operational problems:

• unavailable items after order acceptance,
• late substitutions,
• partial picks,
• picker rework,
• dispatch delays while baskets are corrected,
• driver wait time at store,
• failed or delayed handoffs,
• customer refunds and service contacts.

The integration depth between the routing engine, WMS, OMS, e-commerce front end, and inventory systems matters more than dark store operating models often acknowledge.

A capable dark store routing stack should be able to consume inventory availability, inventory confidence, order promise windows, pick status, and exception signals quickly enough to keep routes aligned with operational reality.

4. Capacity-Aware Dispatch

Every dark store has finite hourly capacity.

That capacity includes:

• picker availability,
• pick rate by category,
• staging space,
• packing capacity,
• chilled or frozen handling constraints,
• dispatch dock capacity,
• driver check-in and handoff throughput,
• and available fleet capacity for each delivery window.

This is a capacity planning for omnichannel retailers problem as much as it is a routing problem. Saturation at one store while a nearby store has slack is system-level waste. Capacity-blind routing creates this pattern repeatedly: high-demand stores overload, underused stores stay underused, SLA failures cluster around constrained nodes, and dispatch teams compensate manually.

Capacity-aware dispatch treats store workload and fleet availability as constraints in the optimisation problem. Orders are not routed only to the geographically closest store. They are routed to the node that can pick, stage, dispatch, and deliver within the customer promise at an acceptable cost-to-serve.

This is where auto-dispatch logistics software can reduce manual intervention. A store can look optimal geographically and still be the wrong fulfilment node if its pick queue is already at risk of breaching the SLA.

5. Surge Orchestration Across the Network

Promotional spikes, weather events, holidays, and viral product moments rarely hit a network evenly.

One store may receive a disproportionate share of orders because of a local promotion, weather-related demand, or a high-volume customer cluster. Nearby stores may still have capacity. Physical inventory rebalancing is slow. Routing rebalancing is immediate.

That makes the routing layer the practical surge-control layer.

Strong surge orchestration allows retailers to:

• divert orders before a store saturates,
• rebalance across nearby dark stores,
• protect priority SLA windows,
• allocate owned, 3PL, and gig capacity dynamically,
• batch compatible orders more aggressively,
• adjust dispatch waves,
• flag high-risk orders early,
• and reduce exception-driven decision-making.

Operators that treat surge response as a routing-layer problem absorb peaks with less SLA degradation and lower exception cost. Operators that do not typically experience every peak as a recurring crisis managed through calls, spreadsheets, manual driver reassignment, and efforts to manage delivery exceptions after service risk has already appeared.

Also Read: Supply Chain Control Tower: How to Build Real-Time Logistics Visibility That Delivers ROI

Composite Illustrative Scenario

Consider a mid-size grocery retailer running fifteen dark stores across a major metropolitan area — Chicago, Boston, or the San Francisco Bay Area — with daily order volume around three thousand and a ninety-minute delivery promise.

Without routing as a first-class layer

Store-to-order assignment runs by static zip code. Each store dispatches its own drivers without cross-store coordination. Inventory data refreshes every four to six hours. Dispatch teams plan locally, with limited visibility into metro-wide fleet capacity. Surge response is manual.

When Black Friday, a weather event, or a sudden product spike hits, the operations team works through escalation calls and ad-hoc reassignments. Popular stores saturate. Quieter stores remain underused. Drivers wait for late picks at constrained stores while available capacity sits elsewhere. Basket exceptions create rework. SLA failures concentrate in predictable pockets of the network.

The result is not simply “late delivery”. It is higher cost-to-serve:

• more manual dispatch intervention,
• more driver idle time,
• more unproductive miles,
• more failed batching opportunities,
• more substitutions and customer contacts,
• poorer route density,
• and lower confidence in the delivery promise.

SLA hit rates sit at the lower end of operationally acceptable. Delivery cost per order trends higher than the network’s potential.

With routing as a first-class layer

Store-to-order assignment runs dynamically per order. The routing engine evaluates inventory state, pick capacity, driver capacity, batching opportunities, current demand, traffic, and the customer promise window.

Multi-origin dispatch routes drivers across the metro with system-level optimisation, rather than treating each dark store as an isolated operation. Inventory integration runs continuously rather than on hours-long refresh cycles. Capacity-aware routing diverts orders before saturation becomes an SLA failure. Dispatch automation reduces the number of decisions that supervisors need to make manually.

During a surge, the routing layer:

• detects capacity risk at the store level,
• diverts eligible orders to nearby nodes,
• prioritises high-risk SLA windows,
• adjusts batching where customer promises allow,
• reallocates fleet capacity across owned, third-party, and gig pools,
• and reoptimises routes as traffic and fulfilment status change.

Capacity utilisation balances across stores. SLA hit rates improve into the higher operational range. Delivery cost per order trends towards the network’s economic potential.

The qualitative difference is meaningful. The quantitative difference depends on metro density, network configuration, demand patterns, fleet model, inventory accuracy, and existing operational maturity — and varies enough across operators that publishing a single universal number would mislead.

Operational metrics to baseline before and after routing optimisation

KPI	Why it matters in dark store routing
Cost per order	Shows whether routing decisions are improving the unit economics of fulfilment and delivery
Cost-to-serve by node	Identifies which dark stores, zones, or promise windows are structurally expensive
On-time delivery rate	Measures whether routing and dispatch are protecting the customer promise
SLA adherence by window	Shows performance against one-hour, ninety-minute, two-hour, or scheduled delivery promises
Miles per order	Indicates route density and the cost impact of poor assignment decisions
Driver utilisation	Measures productive driver time versus waiting, deadheading, or underloaded routes
Pick capacity utilisation	Shows whether fulfilment workloads are balanced across stores
Order batching rate	Indicates whether the network is creating efficient multi-stop routes where service promises allow
Substitution and stockout rate	Reveals whether routing is aligned with inventory reality
Manual dispatch override rate	Measures operational friction and the limits of automation
Failed or delayed handoff rate	Shows whether store operations and dispatch are synchronised

Where the Routing Layer Lives Architecturally

Routing platforms like Locus that handle multi-origin dispatch, inventory-integrated routing, capacity-aware decisions, and cross-network surge orchestration as core capabilities provide the routing-layer architecture dark store networks need to capture full economics.

The architectural choice matters.

A dark store routing stack should not sit at the end of the process as a basic route planner. It should sit between order orchestration, fulfilment operations, and fleet execution.

A practical flow looks like this:

1. E-commerce front end captures the order
   The customer selects a delivery promise: one hour, ninety minutes, two hours, or a scheduled window.
2. OMS validates the order and promise
   The order management system confirms basket, customer location, payment, serviceability, and promise rules.
3. Inventory and WMS systems expose node-level availability
   The routing layer needs current SKU availability, substitution risk, pick status, and fulfilment constraints by store.
4. Routing assigns the fulfilment node
   The platform evaluates inventory, pick capacity, fleet capacity, batching opportunity, distance, traffic, and SLA risk.
5. Dispatch automation creates and updates routes
   Orders are sequenced into routes across owned fleets, 3PLs, marketplace partners, or gig capacity where relevant.
6. Driver execution and customer visibility run in real time
   Driver apps, ETAs, proof of delivery, exceptions, and customer notifications feed back into the control tower.
7. The network reoptimises continuously
   As inventory, pick status, driver location, cancellations, traffic, and demand change, routes and assignments are recalculated where operationally appropriate.

Dark store routing also needs to operate across broader omnichannel fulfillment models, not only dedicated dark stores.

Also Read: AI-Powered Dynamic Pricing: Solving the Last-Mile Delivery Crisis

According to INRIX congestion data, urban routing in dense North American metros faces variable conditions that further reward routing architectures with continuous reoptimisation rather than static assignment.

The routing decisions that matter most for dark store economics are made many times per hour, against changing inventory, capacity, traffic, and demand state. A static plan cannot keep up with that operating environment.

What to evaluate in dark store routing software

Supply chain and digital fulfilment teams evaluating a dark store routing platform should pressure-test the following capabilities:

• Multi-origin optimisation: Can the platform optimise across multiple dark stores, in-store fulfilment locations, MFCs, and depots?
• Store-to-order assignment: Does it assign orders dynamically using inventory, capacity, SLA, distance, and cost-to-serve?
• Inventory integration: Can it consume near real-time inventory and order status from WMS, OMS, and inventory systems?
• Pick-capacity constraints: Can it account for picker availability, pick queue, staging capacity, and store-level saturation?
• Fleet orchestration: Can it allocate across owned fleets, 3PLs, contract carriers, marketplace platforms, and gig drivers?
• Route reoptimisation: Can it update routes as demand, traffic, pick status, and driver availability change?
• Batching intelligence: Can it batch orders without compromising service promises or product handling requirements?
• SLA prioritisation: Can it protect different delivery promises and customer tiers?
• Exception handling: Can it automate decisions for delays, failed picks, driver no-shows, cancellations, and stockouts?
• Control tower visibility: Can operations teams see network health, route risk, capacity risk, and SLA exposure in real time?
• Analytics: Can leaders measure cost per order, on-time delivery, driver utilisation, miles per order, and manual override rates by store, zone, and promise window?

The architectural choice — routing platforms purpose-built for multi-origin operations versus single-origin systems adapted to dark store deployment — determines whether the routing layer creates network value or leaks it.

---

Benefits of Treating Dark Store Routing as Core Infrastructure

When dark store routing is designed as core infrastructure, not a post-launch configuration exercise, retailers can improve both operational performance and financial control.

1. Lower cost-to-serve

Routing determines how many miles a delivery requires, whether the order can be batched, whether a driver waits at store, and whether a nearby fulfilment node could serve the customer more efficiently.

Supply Chain Management Review data cited in Trax Technologies analysis states that strategically located dark stores can reduce average last-mile delivery distances by 23% versus traditional store-based fulfilment networks. Location matters, but routing determines whether the network actually captures that distance advantage.

2. Better pick and delivery synchronisation

Fast delivery fails when picking and dispatch operate on separate clocks. A driver dispatched too early waits at the store. A driver dispatched too late puts the SLA at risk. Inventory-aware and capacity-aware routing aligns pick progress, staging readiness, driver location, and delivery sequencing.

3. Higher order accuracy and fewer exceptions

Food Marketing Institute data cited in Trax Technologies analysis states that dedicated dark-store-style fulfilment facilities can reduce per-order labour costs by 28% and improve order accuracy from 94% to 98% compared with in-store picking. Routing cannot replace fulfilment process discipline, but it can prevent avoidable exceptions caused by assigning orders to the wrong node or dispatching against stale inventory.

4. Faster order processing

National Retail Federation data cited in Trax Technologies analysis states that dark store operating models can achieve average online order processing times of about 12 minutes versus 20 minutes for conventional store-based fulfilment. Those processing gains translate into customer value only when routing can convert faster picks into reliable dispatch.

5. More resilient surge performance

Promotions, storms, holidays, local events, and viral product demand create nonlinear spikes. Routing helps the network flex before capacity breaks by diverting orders, adjusting batches, reallocating fleet, and reoptimising routes against live conditions.

6. Lower emissions and more sustainable delivery

Efficient dark store routing reduces avoidable miles, deadheading, idle time, and fragmented deliveries. For retailers moving toward electric fleets, zone-based charging constraints, or sustainability reporting, route optimisation becomes part of the green logistics operating model.

---

Key Features of a High-Performance Dark Store Routing Layer

A routing layer built for dark stores needs more than route sequencing. It needs to coordinate fulfilment, inventory, capacity, and delivery execution as one decision system.

Core capabilities

Capability	Why it matters
Dynamic store-to-order assignment	Prevents fixed territories or zip-code rules from overloading specific nodes
Multi-origin route optimisation	Allows the network to optimise across several dark stores, not one origin at a time
Inventory-aware fulfilment logic	Reduces stockouts, substitutions, partial picks, and customer-visible failures
Pick-capacity awareness	Prevents routing orders into stores that cannot pick within the delivery promise
Fleet-capacity orchestration	Allocates orders across owned fleets, 3PLs, contract carriers, and gig capacity
Real-time reoptimisation	Adjusts routes as traffic, pick status, demand, and driver availability change
Batching intelligence	Improves route density without compromising customer promises or product handling
Exception automation	Reduces manual decisions during delays, failed picks, cancellations, and driver no-shows
Control tower visibility	Gives operations teams real-time visibility into SLA risk, route health, and capacity exposure
Performance analytics	Links routing decisions to cost per order, miles per order, on-time delivery, and utilisation

How to Make Dark Store Routing a First-Class Design Input

Dark store routing should influence network design before launch and guide continuous optimisation after launch.

Step 1: Map demand, service promises, and fulfilment nodes

Start with customer density, order frequency, basket composition, promised delivery windows, traffic patterns, and existing retail footprint. Use this to identify overlapping service areas, high-cost zones, and areas where one fulfilment node is likely to become overloaded.

Step 2: Define assignment logic before the network goes live

Decide which nodes are eligible to fulfil each order type. Assignment should reflect SKU availability, inventory confidence, pick capacity, fleet availability, delivery promise, and cost-to-serve — not only distance.

Step 3: Build inventory and capacity integrations early

Routing cannot be inventory-aware if inventory signals are stale. It cannot be capacity-aware if pick queues, staging constraints, and fleet capacity are invisible. OMS, WMS, inventory systems, and routing software need shared operating data.

Step 4: Design dispatch rules for mixed fleets

Most North American retailers do not rely on one fleet type. They use some combination of owned drivers, 3PLs, contract carriers, marketplace delivery partners, and gig capacity. Routing rules should reflect cost, service level, capacity, vehicle type, reliability, and customer promise.

Step 5: Establish surge protocols before the first peak

Surge orchestration should not depend on supervisors discovering problems manually. Define diversion thresholds, batching rules, priority SLA logic, driver reallocation rules, and escalation paths before promotions, weather events, or holidays hit.

Step 6: Measure routing ROI continuously

Track cost per order, cost-to-serve by node, on-time delivery, SLA adherence by window, driver utilisation, miles per order, pick-capacity utilisation, batching rate, stockout rate, and manual dispatch override rate. Routing economics change as demand density, inventory behaviour, and fleet mix change.

---

Why Choose Locus for Dark Store Routing

Dark store routing is an orchestration problem across orders, inventory, capacity, fleets, service promises, and real-world constraints. Locus is designed for this operating environment.

Locus supports:

• multi-origin routing across dark stores, fulfilment centres, depots, and retail nodes,
• dynamic order allocation based on operational constraints,
• automated dispatch planning and reoptimisation,
• owned fleet, 3PL, contract carrier, marketplace, and gig fleet orchestration,
• SLA-aware route planning,
• control tower visibility,
• exception handling,
• analytics for cost, service, utilisation, and operational performance,
• and integration with enterprise systems that sit around fulfilment and delivery execution.

For dark store networks, the critical question is not whether routes can be drawn. It is whether the routing layer can make better decisions continuously as orders, inventory, traffic, capacity, and fleet conditions change.

The Real Question for Supply Chain Leaders

Dark store network design is mostly about location and inventory. But the routing layer that operates the network produces a meaningful share of total economics — and many North American operators under-invest in it relative to its impact.

The strategic question is:

Given the network we have or are building, is dark store routing a first-class design input, or an afterthought we will handle later?

A first-class routing strategy changes the questions leaders ask before launch:

• Which fulfilment nodes should be eligible for each order type and SLA?
• How will store-to-order assignment work when nearby stores have different inventory and capacity states?
• What pick-capacity thresholds should trigger diversion?
• How will owned, 3PL, marketplace, and gig fleets be allocated across stores?
• Which delivery promises are profitable by zone, basket type, and time of day?
• What routing logic protects SLA adherence during weather, promotions, and holidays?
• How will dispatch automation reduce manual escalation?
• Which KPIs will define routing ROI: cost per order, on-time delivery, driver utilisation, miles per order, or all of them?

For some networks, a configured legacy routing tool may be sufficient: low density, few nodes, broad delivery windows, and stable demand. For high-volume grocery and convenience networks with multiple dark stores, tight SLAs, changing inventory, and mixed fleet capacity, the routing layer needs to be designed as core infrastructure.

At Locus, this is the operating view: last-mile performance is not only a route sequencing problem. It is an orchestration problem across orders, inventory, capacity, fleets, service promises, and real-world constraints.