#Japan #Tokyo #DATAFLUCT #Airlake #Itochu Foods

DATAFLUCT and Itochu Foods Collaborate on AI Order Prediction

DATAFLUCT, based in Shibuya, Tokyo, led by CEO Hayato Kumemura, has recently finalized a promising proof-of-concept experiment aimed at revolutionizing the food logistics sector in collaboration with Itochu Foods, headquartered in Osaka. This initiative focused on building an AI model for order predictions, with real-world applicability demonstrating notable accuracy levels.

The experiment spanned five warehouses and involved approximately 4,500 items, utilizing a variety of algorithms and rigorous engineering techniques. Among the findings, one warehouse achieved a weighted average percentage error (WAPE) of 28.9%, confirming the model’s viability for practical use. Both companies are now looking to implement this technology for a broader deployment across their national networks.

Background and Challenges

Itochu Foods serves as a crucial hub within Japan's food infrastructure, linking around 4,000 manufacturers with approximately 1,000 retailers while managing 500,000 items. The main challenge faced by the company was the increasing workload of order management due to the complexities introduced by logistics efficiency initiatives. Furthermore, food distribution involves numerous constraints that go beyond simple demand forecasting, including seasonal variations, promotional impacts, varying lead times from different manufacturers, holiday closures, minimum order quantities, and shelf-life management.

To address these challenges, DATAFLUCT leveraged its data utilization platform, "Airlake," to integrate shipping data encompassing these complex factors. The aim is to develop a sophisticated forecasting system capable of predicting the unique and unpredictable demand patterns characteristic of food distribution using AI.

Overview of the AI Order Automation

1. Comprehensive Automation from Demand Forecasting to Order Optimization
This project aims at achieving seamless automation of the entire process from demand forecasting to order optimization. DATAFLUCT has taken the lead in developing the core demand prediction model while implementing order logic based on Itochu Foods’ operational expertise.
To enhance prediction accuracy, this initiative integrates past shipping data from Itochu Foods with external data such as weather conditions and major event information accessed through the "Airlake" platform. Orders will be predicted by warehouse, product, and day. Future enhancements will include automatic calculations of safety stock considering out-of-stock risks, as well as specific lead times and minimum order quantities to be included in the ordering logic.

2. Results of the Experiment
The experiment evaluated around 4,500 products across five Itochu Foods logistics sites, employing cutting-edge deep learning algorithms and versatile tree-based models like LightGBM in a unique ensemble methodology. This approach allowed the team to capture certain predictable trends in demand despite potential fluctuations caused by seasonality or promotional activities. Subsequently, a WAPE rate of 28.9% was recorded at some warehouses, reflecting the model’s predictive validity for real-world application.

Future Plans

Going forward, the focus will be on refining the models for specific product categories and enhancing the integration of external data to better manage sudden demand changes. The goal is to establish a robust order logic based on forecast results and design systems that can be scaled nationwide. Continuous operational improvements, referred to as MLOps, will be necessary to maintain high predictive accuracy over the long term, helping to lead the transformation of the food wholesale industry through a next-generation supply chain framework.

About DATAFLUCT's Airlake

"Airlake" is a comprehensive platform that aids in data analysis and the development of AI agents. It streamlines data automation through natural language interactions, generating insights from previously untapped data.

By offering a complete suite that includes the Airlake platform for data management, custom AI models, and AI agents, DATAFLUCT provides rapid and cost-effective AI implementations tailored to the specific needs of clients.

About DATAFLUCT

Founded with the vision "Transforming Data into Business," DATAFLUCT seeks to derive new value from previously hidden data to address societal challenges. The company excels in multimodal data utilization, handling various data types without constraints. It focuses on projects such as demand prediction for loss reduction, sustainable urban planning, and promoting carbon-neutral behaviors, striving to create a world where data is effectively harnessed for sustainable decision-making. Recognized as a JAXA Venture enterprise since 2019, DATAFLUCT continues to innovate in the realm of data applications and AI.

Headquarters: 1-4 Sakuragaokacho, Shibuya, Tokyo 7th Floor, Shibuya Sakura Stage
CEO: Hayato Kumemura
Established: January 29, 2019
Contact Number: +81-3-6822-5590
Capital: ¥1.49712 billion (including capital reserves)
Services: Data platform development and operation support, DX promotion support, sustainable data business initiatives
Website: DataFluct
Official Twitter: @datafluct
Facebook: DATAFLUCT Facebook

For more information and inquiries about their services, visit DATAFLUCT Services. For press inquiries, please contact [email protected].