The Hyperlocal Fulfillment Equation: Building 2-Hour Delivery Networks That Profit

Between 2021 and 2024, North American pure-play quick commerce collapsed almost entirely. Gorillas exited the US market. Buyk ceased operations. Fridge No More closed. Getir withdrew from the US in 2024. Jokr pulled back from North America. The pattern was consistent across the category: well-funded venture-backed operators, dense networks of dark stores, 15-minute delivery promises — and a unit-economics reality that no marketing budget could outrun

Meanwhile, a different group of operators kept quietly delivering at 2-hour windows and getting more profitable every year: the big-box and grocery retailers using their existing store networks as fulfillment nodes. Walmart through Walmart+ and Spark. Target through Target Same-Day and Shipt. Amazon through Whole Foods. Kroger through its Boost program and Ocado partnership. Albertsons across Vons, Safeway, and Jewel-Osco. None of these operators promised 15-minute delivery. All of them are scaling.

The difference between profitable hyperlocal fulfillment and venture-burned hyperlocal fulfillment isn’t speed. It’s orchestration. The North American operators succeeding at 2-hour delivery aren’t running faster trucks. They’re running better answers to four orchestration questions about distributed inventory, node selection, repositioning, and dispatch density. The ones who failed treated hyperlocal as a delivery-speed problem when it was always a distributed-inventory-orchestration problem.

According to Capgemini Research Institute, last-mile delivery accounts for 41% of overall supply chain costs in retail parcel — making every hyperlocal architectural decision a direct margin event.

Why the Pure-Play Model Failed in North America

Three structural problems sank the pure-play quick commerce model in NA, and understanding them clarifies why the store-fulfillment alternative works.

1. Inventory replication across dense node networks destroyed margins. Pure-play operators built standalone dark store networks across major metros — Manhattan, Brooklyn, Chicago, Boston, San Francisco — replicating inventory across dozens of small fulfillment locations. Each node carried inventory it could never turn fast enough to justify. Carrying costs, dead inventory, and stockouts on fast-moving SKUs all compounded simultaneously.

2. Order density was structurally insufficient. A profitable last-mile route requires multiple orders within batchable distance and time. Standalone dark store networks with novel customer bases couldn’t generate the density. Drivers ran routes with two or three stops where five or six were needed for unit economics to work.

3. Customer acquisition cost exceeded contribution margin per order. With 15-minute delivery as the marketing promise, operators paid heavily for customer acquisition while losing money on every order delivered. There was no pathway to profitability at scale because the underlying unit economics never crossed.

Why Store-Fulfillment Models Work

The retailers profiting at 2-hour delivery in North America share an architectural choice: they treat existing stores as fulfillment nodes rather than building new dense-fulfillment networks. This decision unlocks four economic advantages simultaneously.

Inventory is already deployed. A Walmart, Target, Kroger, Albertsons, or Whole Foods location in Dallas-Fort Worth, Chicago, NYC metro, or Los Angeles already carries the SKUs customers want. Hyperlocal delivery uses inventory that was capitalized and stocked for an existing in-store customer base. There is no incremental carrying cost.

Order density follows existing demand. Stores are sited where demand already exists. Delivery routes draw from a customer base that overlaps with the store’s existing trade area. Density is structural, not something to be manufactured through marketing.

Also Read: Retail Ecommerce Fulfillment Strategy Guide 2026

Real estate is sunk in cost. The fulfillment-node real estate was paid for years ago and serves a dual purpose. The marginal cost of using a Whole Foods location as an Amazon Fresh fulfillment node is operational, not capital.

Pick labor scales with order volume in proportion. Store associates can be allocated to picking based on hourly demand, with cross-utilization for restocking and customer service when delivery volume is low. Standalone dark stores carry fixed labor costs regardless of order volume.

This is the architecture that works. The question is what orchestration capabilities are required to actually run it.

The Four Orchestration Questions Every Profitable Hyperlocal Network Answers

For Heads of E-Commerce Operations evaluating hyperlocal infrastructure, the technical decisions that determine profitability live in four distinct orchestration questions.

1. Which Fulfillment Node Should Fulfill Which Order?

When a customer in Brooklyn places an order, multiple nodes might be capable of fulfilling it: the nearest Whole Foods store, an Amazon Fresh fulfillment center, a third-party dark store partner, a 3PL location. The orchestration decision selects the optimal node based on inventory availability, distance to customer, current pick capacity, order composition, SLA requirement, and cost-to-serve.

Most retailers answer this with static rules — nearest store with inventory wins. Profitable operators answer it dynamically. The “right” node varies by order, by current operational load, and by margin contribution. A store running at 95% pick capacity may not be the optimal node even if it’s nearest, because the marginal pick adds delay risk and SLA exposure. A nearby dark store with ample capacity may be the better choice for that specific order, even at slightly higher distance.

2. What Inventory Sits at Which Node?

Each fulfillment node type performs differently against different inventory profiles. Stores work for full-assortment fulfillment because they’re already stocked for in-store customers. Dark stores work for top-velocity SKUs in dense urban zones where store coverage is thin. Micro-fulfillment centers (MFCs) work for the highest-velocity, automation-friendly SKUs.

The architectural mistake — the one that destroyed pure-play q-commerce — is replicating full inventory across all nodes. The architectural success is dynamic SKU allocation by node, based on velocity, basket associations, and zone-specific demand patterns. The operators profiting at 2-hour delivery don’t carry the same inventory in every node. They carry the right inventory in each node for that node’s role.

3. When and How Is Inventory Repositioned?

Even with correct node-level SKU strategy, demand drifts. A SKU that was high-velocity in one neighborhood last quarter may be slowing; a new SKU pattern may be emerging in another. Profitable hyperlocal operations continuously reposition inventory across nodes through pre-positioning based on demand forecasts, inter-node transfers when imbalances develop, and seasonal rotation reflecting local patterns. The repositioning logic depends on real-time inventory visibility across the entire node network — a prerequisite the pure-plays often lacked.

4. How Are Orders Batched and Dispatched to Maintain Density?

The last mile of hyperlocal delivery represents the single largest cost line. Density engineering — multiple orders per route — is the only way to bring per-order delivery cost into profitable range. Dynamic slot pricing nudges customers toward batchable windows. Wave picking at MFCs groups orders with overlapping SKUs to reduce picker travel. Driver dispatch is timed to route density rather than individual order receipt. According to McKinsey & Company, AI-driven last-mile routing optimization typically delivers cost reductions in the 10–25% range in production deployments — concentrated in operations that solve the density problem at the dispatch layer.

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

Why the Orchestration Layer Has to Be Integrated

The four questions are not independent. The node selected affects inventory consumed, which feeds repositioning decisions, which constrains future node selection. Order batching decisions affect SLA confidence at each node, which affects which orders should be assigned where. Treating these as four separate systems — running them in parallel through different vendors or different teams — is the architectural pattern that produces the “$18 loss per $9.99 order” outcome that pure-play operators experienced.

The integration matters because the data flows in both directions. The orchestration layer needs operational truth from each node (current pick capacity, inventory accuracy, time-to-fulfillment) and the dispatch layer (route density, driver supply, current cost-to-serve) to make decisions that hold across all four questions simultaneously. Without that bidirectional integration, hyperlocal optimization is local rather than systemic — and pure-play q-commerce demonstrated where local optimization without systemic orchestration leads.

According to Brick Meets Click, US online grocery sales have crossed $100 billion annually — large enough that operational architecture differences translate into material balance-sheet impact, fragmented enough that the operators with the strongest orchestration will pull ahead.

What Heads of E-Commerce Operations Should Evaluate

Five questions to apply when assessing hyperlocal fulfillment infrastructure for North American operations.

1. Are we treating hyperlocal as a delivery-speed problem or as a distributed-inventory-orchestration problem? The framing determines which capabilities are evaluated and which are skipped.
2. Does our orchestration layer dynamically select fulfillment nodes per order based on live inventory, capacity, and cost-to-serve, or does it use static nearest-with-inventory rules?
3. Is our node-level SKU strategy deliberate, or are we replicating inventory across all nodes by default? The latter is the pattern that destroyed pure-play unit economics.
4. Do we have real-time visibility across our entire node network — stores, dark stores, MFCs, partner locations — sufficient to make repositioning decisions weekly or daily?
5. Does our dispatch layer optimize for route density, with slot pricing and order batching engineered to maintain it, or does it dispatch each order independently as received?

The Real Question for North American Heads of E-Commerce Operations

The pure-play quick commerce category is mostly gone in North America. The big-box and grocery retailers leveraging existing store networks are scaling. The architectural lesson is now clear: hyperlocal profitability is determined by orchestration, not by speed. The technology and operational frameworks to run it well exist. The strategic question for Heads of E-Commerce Operations isn’t how fast can we promise? It is: do our orchestration capabilities — node selection, inventory placement, repositioning, dispatch density — actually answer the four questions that determine whether each hyperlocal order makes money?

To learn more, visit locus.sh