In a widely circulated and discussed article on Forbes, Nallan Sriram, Global Technology Strategist of Unilever makes a compelling argument for the need for master data for AI initiatives in the enterprise. The article describes that master data gets siloed in operational systems like ERP with the key decision-makers realizing the need for correct master data when faced with revenue loss or increased operational expense. As master data provides context to business transactions, it is fundamental to business operations. In earlier times, we could manage master data through human intervention. But now with cloud data lakes and our aspirations to build predictive algorithms for business operations and operations, the need for clean, contextual and unified master data is all the more enhanced.
Machine Learning depends fundamentally on the quality of input and training data, and it is impossible for the enterprise to derive value out of ML if the base master data is unclean. To ensure that proper links and relationships exist in the master data, the article advises focusing on key entities like customer master, supplier master, product master, supplies master and employee master by
- Improving processes and controls around onboarding of new customers and suppliers
- Data enrichment and validation through external sources
- Improved operations and technology to build relationships between customers, suppliers and products
With digital transformation, automation and adoption of machine learning in the enterprise, master data becomes all the more critical. Master Data could surely eat AI for breakfast! However, there are some points that we would like to add here.
Legacy master data applications can not fulfill the needs of a modern, transformation driven enterprise looking to up its operations by leveraging data. With all its promise of breaking data silos, older master data technology is riddled with long configuration cycles, complex deployments as well as hard-coded rules making both addition of new data sources as well as consumption of master data by base applications cumbersome. Take for example, address data. We need impeccable address data for legal, compliance and postal requirements. However, for a web store where services are also delivered digitally, address data may or may not be mandated as it is not needed for the business. To build linkages between ERP, Contracting, Product Based and other application systems, traditional MDM forces strict address structure and underlying applications are mandated to adapt to that, even if they need it or not. This leads to a lot of coordination across MDM owners and operational teams owning their part of the customer journey and master data. This change management across departments is tedious, time-consuming as well as unwarranted. Instead, if we have powerful algorithms that can match free form addresses with variations along with structured address data, we would not need so much effort from the underlying data sources and owners. If we look under the hood, a lot of normalization and standardization on the source data for the purposes of being able to match and link them is wasteful, especially when the source systems do not need the data in that format or with those attributes.
The idea that there is one single golden record, even when consuming applications can not actually handle the lengths or fields from that record is another example of the legacy MDM forcing unnecessary changes or processes on the source systems. Instead, contextual views of the masters, delivering high quality master data consumable by the enterprise systems, are both practical and necessary.
Process and inter-department collaboration are the key to run a successful business, but those should be driven by the business and not stem from technical limitations of the systems we adopt. To help us achieve our business goals and fully leverage AI/ML, our master Data Management systems need to keep up with the times and adapt, leveraging AI on the core mastering processes. AI is integral to the volume and variety of data we need to manage and master. Agile data mastering fuelled by AI/ML can prepare us for a smooth ride and return on investment on our predictive analytics journey. It can not stay the last cog in the wheel as only the consumer of mastered data. It has to be the oil to our data management and data mastering processes.