In October 2025, Lingmou Intelligence and Swire Coca-Cola renewed a strategic cooperation agreement covering 2025 through 2030.

The relationship had begun in 2019 as a technology project and developed into a broader retail operating system.

According to the companies, six years of AI deployment generated more than RMB 250 million in cost and efficiency benefits and made more than two million retail outlets more visible and actionable.

The figures in this article are company-reported, but the operating model illustrates how AI can connect frontline execution with headquarters decisions.

The Complexity of Managing Beverage Distribution

Swire Coca-Cola operated a portfolio of thirty brands and more than one thousand beverage SKUs across convenience stores, supermarkets, campus stores, and other outlet types.

The network included more than two million retail terminals, ten thousand sales representatives, and seven hundred thousand coolers.

Traditional management struggled with that scale.

Manual counting produced missing items and flavor errors. Different lighting, shelf angles, and overlapping products reduced the reliability of earlier image-recognition systems.

Data also moved slowly. A distribution plan developed at headquarters could take weeks to reach stores, while decision-makers lacked current information about actual availability and sell-through.

The company needed more than one isolated tool. It needed to connect products, stores, people, processes, and equipment.

Four AI Capabilities

1. Computer Vision for SKU and Shelf Recognition

Sales representatives could photograph a shelf rather than count every item manually.

The system recognized products, variants, facings, and distribution compliance across different store environments.

The vendor reported that a new product could be added to the recognition system within seven days and that recognition accuracy reached 99.3 percent.

Headquarters could see where a product had insufficient facings or where a planned SKU was missing without waiting for a manual report.

The companies reported that counting became six times faster, audit expense fell by RMB 120 million annually, and product turnover improved by 18 percent.

2. A Mobile AI Workspace for Sales Representatives

The mobile workflow adapted to the type of visit.

A quick convenience-store visit used a photo to complete stock and display checks. A deeper supermarket visit presented the relevant account information automatically.

Photos moved to headquarters in real time, while the application highlighted missing facings, replenishment needs, and competitor activity.

The reported time per store fell from twenty minutes to eight. Representatives could visit more outlets and use more time discussing growth opportunities with store owners.

The companies also reported lower competitive-audit cost and higher sales, though those outcomes should be interpreted as project claims rather than independent measurement.

3. Competitive Intelligence across the Category

The system developed a beverage database covering 25,000 non-alcoholic and 8,000 alcoholic beverage SKUs.

Combined with location and time data, the recognition results created market heat maps.

Headquarters could observe competitor distribution, category momentum, and shelf position by outlet type and region.

When Costa bottled coffee launched in 2021, the system analyzed the placement of competing ready-to-drink coffee in leading convenience chains. Swire then prioritized 28,000 high-potential outlets.

The project reported a return on investment 3.6 times the industry average for that deployment.

4. Connected Coolers as Managed Assets

Seven hundred thousand coolers were distributed across community stores, fuel stations, and restaurants.

The connected system used location, temperature, and door-opening information to identify movement, failure, and replenishment demand.

A fuel-station cooler with much higher opening frequency could receive more frequent replenishment. A temperature problem could trigger maintenance. A relocated asset could trigger an alert.

The companies reported that cooler power-on rates increased from 78 percent to 96 percent, loss fell from 5 percent to 0.1 percent, and out-of-stock rates fell from 15 percent to 4 percent.

The cooler became both a physical asset and a source of local demand data.

The Defensible Asset Is the Operating Data

After six years, the project had accumulated three forms of capability.

The first was a beverage knowledge graph covering twelve outlet types and more than 25,000 SKUs.

The second was faster model improvement. A federated-learning approach reportedly reduced iteration time from three months to seven days.

The third was field reliability. Edge processing provided rapid responses even when store connectivity was limited.

The difficult part was not acquiring an image model. It was building the labeled data, store workflow, device network, and organizational adoption required to make the model useful at scale.

Digitalization as a Revenue and Productivity Asset

The project connected four elements:

  • people: sales representatives;
  • products: beverage SKUs;
  • places: retail outlets;
  • equipment: coolers.

More than a billion outlet images supported model improvement. Cooler data informed replenishment and maintenance. Outlet observations moved into headquarters decisions.

That closed loop turned terminals from hard-to-manage endpoints into observable operating nodes.

The next stage described by the companies included dynamic pricing experiments, simulated shelves for testing new products, carbon accounting, FMCG-specific AI assistants, and temperature and humidity monitoring.

Not every idea will produce the reported value, and other brands should not assume they can copy the economics directly.

The transferable lesson is narrower and stronger: AI creates value in FMCG when it is embedded in a specific frontline workflow, supported by proprietary operating data, connected to an accountable decision, and measured through execution outcomes.