At Snowflake Summit 26 in San Francisco, Pulmuone’s head of supply chain management, Ahn Soon-sik, presented the food company’s effort to use AI to make its SCM operations smarter, highlighting how much of the work depended not on the model itself but on the quality of the data feeding it.
The project has drawn attention because SCM sits deep in the back end of a business, far from customer-facing functions such as marketing or sales. Snowflake said that applying AI at this...
Continue Reading This Article
Enjoy this article as well as all of our content, including reports, news, tips and more.
By registering or signing into your SRM Today account, you agree to SRM Today's Terms of Use and consent to the processing of your personal information as described in our Privacy Policy.
level is still relatively rare, including in the food sector, which helped put Pulmuone’s experience on the summit agenda.
Speaking after his session to Korean reporters, Mr Ahn described a process shaped by the pressures of fresh food logistics. Short shelf life means that even modest forecasting errors can quickly turn into excess stock, forcing firms to choose between discounting, moving inventory elsewhere or scrapping it. In his view, the financial consequences of supply chain decisions are only becoming more significant.
The challenge becomes more complicated overseas, where headquarters may hold most of the expertise in demand analysis, inventory risk, supply planning, profitability assessment and scenario modelling. Mr Ahn said Pulmuone saw AI agents as a way to extend that knowledge to local teams across time zones, allowing staff to ask questions at any hour rather than waiting for human support.
But the project’s biggest hurdle was not the AI layer itself. It was cleaning and standardising the underlying information. Mr Ahn said it took more than six months to convert SCM data into a form that AI could reliably use. Field names varied from system to system, with the same item code appearing under different labels and different numbering conventions, making it impossible to use the data directly without extensive harmonisation.
That work required more than technical skill. Mr Ahn said it depended on people who understood both the business and how to work across departments, because AI-ready data cannot be created effectively without domain knowledge. He added that the company also had to teach the system the context of its own business, after early versions produced answers that front-line staff found inaccurate.
To address that, Pulmuone built a separate knowledge layer combining general SCM know-how, the company’s internal terminology and reporting material. Mr Ahn said the firm used an Anthropic-based knowledge management system within Snowflake’s platform and split industry knowledge from Pulmuone-specific expertise before loading both into the agent.
He argued that the real competitive edge lies not in perfect prediction but in responding faster when plans and reality diverge. The harder task, he said, was not linking models together but teaching the agent how the business thinks.
Pulmuone also pushed the system beyond basic question answering. The agent was designed to examine why a problem appeared, whether it was temporary or structural, whether it related to pricing in a particular channel and whether it reflected a product’s lifecycle stage. In one example, if a stock-keeping unit is discontinued, the system can map the likely effects on remaining inventory, production schedules, cost allocation and profit and loss.
According to Pulmuone’s internal testing, the AI analysis matches data in Snowflake at a 100 per cent level. The company also said it was able to move data models far more quickly than before, with work that previously would have taken months completed in three weeks for its domestic, Japanese and US entities using Coco, Snowflake’s coding agent.
The SCM intelligence project is still being rolled out, and Mr Ahn said a major priority this year is simply getting more employees to use it. He said the long-term aim is a self-healing system that can automatically adjust plans when market conditions change.
Even so, he said people will remain central. Final responsibility, he added, still sits with humans, even if AI takes over more of the routine work.
Source: Noah Wire Services
