MandM Direct：使用 Dataiku + GCP 大规模管理模型
公司规模
1,000+
地区
- Europe
国家
- United Kingdom
产品
- Dataiku
- Google Cloud Platform (GCP)
- Google BigQuery
技术栈
- Python
- Machine Learning
实施规模
- Enterprise-wide Deployment
影响指标
- Digital Expertise
- Productivity Improvements
技术
- 分析与建模 - 机器学习
- 平台即服务 (PaaS) - 数据管理平台
适用行业
- 零售
适用功能
- 商业运营
- 销售与市场营销
用例
- 库存管理
- 预测性维护
- 供应链可见性(SCV)
服务
- 云规划/设计/实施服务
- 数据科学服务
关于客户
MandM Direct 是英国最大的在线零售商之一，拥有超过 350 万活跃客户和七个专门的欧洲本地市场网站。该公司每年向全球 25 多个国家/地区提供 300 多个品牌。2020 年，他们经历了快速增长，客户和数据量增加。这种增长加剧了他们的一些挑战，并迫使他们寻找更具可扩展性的解决方案。核心数据团队由四人组成（两名数据科学家、一名高级分析师和一名数据分析师），但他们通过利用中心辐射模型作为其数据中心卓越中心来扩大其覆盖范围，这意味着他们与嵌入在业务线中的分析师合作以扩大他们的工作。
挑战
MandM Direct 是英国最大的在线零售商之一，在快速发展的过程中面临着巨大的挑战。该公司在欧洲拥有超过 350 万活跃客户和 7 个专门的本地市场网站，每年向全球 25 多个国家/地区提供 300 多个品牌。他们的加速增长意味着更多的客户，因此也意味着更多的数据，这加剧了他们的一些挑战，并迫使他们寻找更具可扩展性的解决方案。两个主要挑战是将所有可用数据从孤岛中转移到统一的、可用于分析的环境中，并以可跟踪、透明和协作的方式扩展 AI 部署。最初，该公司的首批机器学习模型是用 Python（.py 文件）编写的，并在数据科学家的本地机器上运行。然而，随着生产中模型数量的增加，团队很快意识到维护模型的负担。
解决方案
为了应对挑战，MandM Direct 选择了 Dataiku 和 Google Cloud Platform (GCP) 的强大组合。借助 Google BigQuery 完全托管的无服务器数据仓库，MandM 可以打破数据孤岛，实现团队间数据访问的民主化。同时，得益于 Dataiku 用于数据管道、数据准备、模型训练和 MLOps 的可视化协作界面，MandM 还可以轻松地在生产中扩展其模型，并且透明且可追溯，不会出现故障或中断。MandM 现在拥有数百个实时模型，可以执行从评估客户倾向到生成定价模型等所有操作，所有这些都可以查看模型性能指标，明确区分设计和生产环境，并且平台中还内置了更多 MLOps 功能。现在，团队可以轻松地将数据准备和机器学习的计算推送和卸载到 GCP。
运营影响
数量效益
Case Study missing?
Start adding your own!
Register with your work email and create a new case study profile for your business.
相关案例.
Case Study
Improving Production Line Efficiency with Ethernet Micro RTU Controller
Moxa was asked to provide a connectivity solution for one of the world's leading cosmetics companies. This multinational corporation, with retail presence in 130 countries, 23 global braches, and over 66,000 employees, sought to improve the efficiency of their production process by migrating from manual monitoring to an automatic productivity monitoring system. The production line was being monitored by ABB Real-TPI, a factory information system that offers data collection and analysis to improve plant efficiency. Due to software limitations, the customer needed an OPC server and a corresponding I/O solution to collect data from additional sensor devices for the Real-TPI system. The goal is to enable the factory information system to more thoroughly collect data from every corner of the production line. This will improve its ability to measure Overall Equipment Effectiveness (OEE) and translate into increased production efficiencies. System Requirements • Instant status updates while still consuming minimal bandwidth to relieve strain on limited factory networks • Interoperable with ABB Real-TPI • Small form factor appropriate for deployment where space is scarce • Remote software management and configuration to simplify operations
Case Study
How Sirqul’s IoT Platform is Crafting Carrefour’s New In-Store Experiences
Carrefour Taiwan’s goal is to be completely digital by end of 2018. Out-dated manual methods for analysis and assumptions limited Carrefour’s ability to change the customer experience and were void of real-time decision-making capabilities. Rather than relying solely on sales data, assumptions, and disparate systems, Carrefour Taiwan’s CEO led an initiative to find a connected IoT solution that could give the team the ability to make real-time changes and more informed decisions. Prior to implementing, Carrefour struggled to address their conversion rates and did not have the proper insights into the customer decision-making process nor how to make an immediate impact without losing customer confidence.
Case Study
Digital Retail Security Solutions
Sennco wanted to help its retail customers increase sales and profits by developing an innovative alarm system as opposed to conventional connected alarms that are permanently tethered to display products. These traditional security systems were cumbersome and intrusive to the customer shopping experience. Additionally, they provided no useful data or analytics.
Case Study
Ensures Cold Milk in Your Supermarket
As of 2014, AK-Centralen has over 1,500 Danish supermarkets equipped, and utilizes 16 operators, and is open 24 hours a day, 365 days a year. AK-Centralen needed the ability to monitor the cooling alarms from around the country, 24 hours a day, 365 days a year. Each and every time the door to a milk cooler or a freezer does not close properly, an alarm goes off on a computer screen in a control building in southwestern Odense. This type of alarm will go off approximately 140,000 times per year, equating to roughly 400 alarms in a 24-hour period. Should an alarm go off, then there is only a limited amount of time to act before dairy products or frozen pizza must be disposed of, and this type of waste can quickly start to cost a supermarket a great deal of money.
Case Study
Supermarket Energy Savings
The client had previously deployed a one-meter-per-store monitoring program. Given the manner in which energy consumption changes with external temperature, hour of the day, day of week and month of year, a single meter solution lacked the ability to detect the difference between a true problem and a changing store environment. Most importantly, a single meter solution could never identify root cause of energy consumption changes. This approach never reduced the number of truck-rolls or man-hours required to find and resolve issues.
