Distribution AX Cases

Challenge

Amazon's fulfillment network now spans more than 300 facilities, so robot routing becomes a network optimization problem rather than a local automation task. Fast-delivery competition makes inbound identification, storage, and order processing speed central to service reliability. As mobile robots scale across the network, congestion and repetitive manual work become management issues for both throughput and safety.

AX Intervention

Amazon builds the robotics layer around integrated physical systems and a fleet-level AI model. Sequoia combines mobile robots, gantry, the Sparrow robotic arm, and ergonomic employee workstations to containerize inventory in totes and redesign inbound storage, picking, and order flow. Amazon says DeepFleet, a generative AI foundation model built with AWS SageMaker and trained on its warehouse and inventory data, will coordinate robot movement across the fulfillment network. The company announces DeepFleet alongside deployment of its 1 millionth robot, with the milestone unit delivered to a Japan fulfillment center.

Outcome

Amazon says DeepFleet will improve robotic fleet travel time by 10%. Its Sequoia robotic system identifies and stores incoming inventory up to 75% faster and reduces fulfillment-center order processing time by up to 25%. As of October 2023, Amazon had over 750,000 robots working with employees on highly repetitive tasks.

Company data shows that in 2022, recordable incident rates and lost-time incident rates were 15% and 18% lower, respectively, at Amazon Robotics sites than at non-robotics sites. Amazon says its robotics rollout also created 700 categories of new skilled job types that did not exist within the company before. Secondary reporting adds that, according to the company, robots already assist in roughly 75% of Amazon customer orders; TechCrunch also cites Wall Street Journal reporting that 75% of global deliveries are robot-assisted in some way.

Challenge

Coupang makes fast Rocket Delivery depend on a directly operated fulfillment network. In traditional warehouse operations, workers travel across large spaces to find inventory, creating fatigue, process handoffs, and error risk. Daegu FC concentrates the bottleneck: picking, packing, sorting, replenishment, and safety must move as one operating system.

AX Intervention

Coupang applies AI order assignment to predict efficient work sequences from inventory, product-location, and delivery-route signals. It shifts picking to a goods-to-person model: AGVs bring shelves to workers, light guidance identifies the target item, and cameras plus sensors manage route recognition and obstacle avoidance. Auto baggers package completed picks, sorting robots route parcels by delivery region, and unmanned forklifts move pallet loads in access-controlled zones. Coupang uses Daegu FC as a future-logistics test bed and says later automation adds AutoStore vertical storage with AI robots.

Outcome

At Daegu FC, Coupang's AGV fleet reaches 1000 units, and ETNews reports automation-system investment of ₩320B. According to the Daegu FC center head, automation reduces work intensity by 65% and eliminates human error. Logistics robots compress the same workload from 3 hours to 1 hour, while Seoul Economic Daily reports AGVs cut overall work steps by 65% and deliver items within 2 min on average. Seoul Economic Daily reports Coupang has invested about ₩6.2T in its Rocket Delivery logistics network, including ₩320B for Daegu FC; ETNews reports Daegu FC creates new employment of 2500+ people, centered on technical roles.

Challenge

Lotte Mart operates in a Korean hypermarket market where fresh-food trust directly affects repeat purchase. Fruit is difficult to grade by appearance alone because sweetness, ripeness, internal cracking, and moisture can be hidden. Manual or sample-based inspection therefore leaves a quality gap: bad fruit can reach stores, create complaints, and weaken the premium proposition.

AX Intervention

Lotte Mart builds its "Fresh Intelligence" program around non-destructive near-infrared Brix sorting and deep-learning analysis. The system measures sweetness without cutting the fruit and screens internal cracking, ripeness, and moisture signals before store intake. Lotte Mart started AI sorting with melons in 2022 and expanded the range to 9 types fruit varieties. It then packages verified higher-sweetness fruit into premium tiers and extends the same data-led operating logic toward a planned Busan CFC using Ocado automation, AI demand forecasting, inventory management, and robotics.

Outcome

Lotte Mart says AI sorting keeps fruit defects within 0.01% of sales volume and improves the pre-adoption level by about 30%. MoneyToday reports that 2025 AI-sorted fruit sales grew more than 7x versus the 2022 adoption year, while News2Day reports that high-Brix product sales reached approx. 2x year over year in January-August 2025. Lotte Mart reports that high-Brix premium fruit sales rose approx. 20% year over year from June 2025 through May 2026. Korea Economic Daily reports that premium products' share of total fruit sales reached approx. 20% from approx. 6%, a more than 3x expansion. ETNews reports that Lotte Mart's Busan CFC plan would handle over 30K deliveries deliveries per day and more than 2x throughput versus existing online fulfillment centers.

Challenge

Online grocery is structurally hard: low margins, irregular baskets, and chilled, frozen, and ambient fulfillment must work as one operation. Waste directly erodes margin because fresh goods are fragile and short-dated. Reuters reports that Ocado says its end-to-end platform delivers food waste of 0.4% of sales, versus 2.5%-3% for most supermarket groups.

AX Intervention

Ocado Smart Platform packages AI control, grid robotics, robotic picking, forecasting, and last-mile routing into an operating platform for grocery fulfillment. In CFCs, an AI "air traffic" control layer coordinates bots, communicates with each bot 10 times per second, and supports speeds up to 4m/s. At Erith, Ocado's $310 million CFC runs more than 2,000 robots across the grid at four metres a second. Its last-mile engine runs 100m optimisation calculations per second, while the machine-learning picking arm could handle 15% of the range in 2021 and was expected by the company to reach 50% in the near term and 80% longer term.

Outcome

Ocado says OSP enables partners to pick a 50-item grocery order in five minutes. It also states its fulfilment centres are among the world's most labour-productive sites, with a 50-item order picked in minutes and simultaneous order picking. Reuters reports that Erith productivity is already 25% higher than Ocado's first British CFC, and Luke Jensen sees scope for 50-70% further gains. The result is not a single robotics deployment; it is a controlled fulfillment system that links demand, storage, picking, and delivery.

Challenge

CJ Olive Young's commerce model puts more demand into digital channels. Maeil Business Economy reports that online revenue rose from ₩60B in 2017 to more than ₩1T in 2023, while Oneul Dream's share of online-mall orders rose from 24.5% in 2021 to 41.3% in 2023. The same report says membership had moved past 10M people to approx. 15M people by September 2024, increasing the payoff from better personalization. Attribute-based recommenders still leave gaps when beauty and health products lack complete product attributes, and manual curation struggles to convert review and behavior data into discovery at app scale.

AX Intervention

CJ Olive Young buys AI startup Rocketview in 2022 and builds recommendation capability in-house. The company says ItemLM learns item similarity from search behavior rather than product attributes, using approx. 3M labels training labels and 20% as validation data. Because the model infers semantic proximity from behavior, it is positioned to recommend products even when product attributes are missing. CJ Olive Young also launches a review-based AI theme service; ETNews reports its self-developed sLLM is trained on 40M reviews customer reviews, extracts approx. 100K expressions expressions, and recommends up to 15 types products per theme. The operating choice is data and model internalization, not only an external recommender purchase.

Outcome

The primary performance claim is narrow and measurable: CJ Olive Young reports ItemLM improves recommendation-screen CTR by approx. 50% versus the previous model. The company also says the same project expands recommendation coverage for products and categories without attributes to 100%. In the broader personalization stack, ETNews reports that the review-trained sLLM creates new themes each day from 40M reviews reviews and approx. 100K expressions extracted expressions. The evidence does not establish a verified conversion-rate lift, so the defensible outcome remains better click response and company-reported full coverage for the previously uncovered catalog.

Challenge

Walmart runs a global store, distribution, and fulfillment network where inventory errors, spoilage risk, and manual handling compound operating cost. Urban stores with constrained shelf space face waste when excess inventory sits in the wrong node. The management problem is to sense imbalance earlier and move inventory, labor, and automation decisions before cost becomes locked in.

AX Intervention

Walmart deploys Self-Healing Inventory to detect overstocked supply and automatically reroute it to stores with demand. It pairs that AI inventory agent with predictive warehouse and transportation management for fresh-food routing and store demand alignment. Walmart also plans to implement Symbotic's robotics and software automation platform across all 42 regional distribution centers over the coming years. In 2025, the company extends the U.S.-tested supply-chain playbook into international markets while using early agentic AI to model complex store and supply-chain conditions.

Outcome

Walmart says Self-Healing Inventory alone has already saved more than $55 million by rerouting overstocked supply to stores with demand. Per Walmart U.S. CEO John Furner, next-generation automated fulfillment centers cut unit costs by 20% year over year versus manual sites.

Looking forward, Walmart expects automated centers to drive more than 30% cost-reduction improvement across its network by the end of 2025 and expects 65% of stores to receive merchandise from high-tech distribution centers by the end of 2025. At its 2023 investor meeting, Walmart projected that by FY2026 roughly 65% of stores would be serviced by automation, about 55% of fulfillment-center volume would move through automated facilities, and average unit costs could improve by approximately 20%.