Time for a Reality Check
Technology savvy customers with easy access to information and a complex supply chain ecosystem have affected a paradigm shift in the role of Item Management. This shift is further compounded by factors such as:
- Constant product information updates by the manufacturer
- Mergers and Acquisitions (M&A) across both the retailers’ and the vendors’ ecosystems
- Critical need for different levels of details for specific new channels
In the online channel, product affinities calculated from item attributes need to be derived based on click stream history, customer master and the overall performance.
With these complexities in place, let’s review Item Management in today’s context:
- Growing dimensions of item attributes: Social and mobile have introduced a new range of item attributes, including derived attributes—images, reviews, videos, product ratings, and so on—increasing the total attribute count to over 1,000. Retailers are struggling not only to capture new and evolving data attributes but also to optimally leverage these attributes for other merchandising areas such as promotions and pricing.
- Inflexible and high maintenance legacy systems: Customers today expect rich and relevant product information. Failure to meet this expectation can result in missed sales. The lack of design foresight in legacy database systems, often acts as a bottleneck in capturing new must-have dimensions of a product. The incremental structural database changes and quick fixes to adapt to new information imperatives are proving to be very expensive, compromising on robustness in the long term.
In addition, item attribute details, often entered manually, are prone to transposition, replication, and typographical errors—a source of revenue leakage every year. Retailers need quick and effective alternatives for tracing and rectifying such wrong entries.
- Costly insights: The era of analytics driven business excellence requires retailers to pose reciprocal queries which are costly to process. For example, in a Relational Database Management System (RDBMS), the query, “What is the rating of a product?” is relatively economical to process as compared to the reciprocal query, "”Which products are rated 5?”.
Although an index can help with simple reciprocal query processing, queries that require a slightly higher degree of recursion, such as "List the products that are rated 5 with more than 3 review comments and specific product specification" become prohibitively expensive.
Meeting these challenges requires remodeling the conventional Item Master into an adaptive one that is capable of managing new streams of data and customizing product configurations in line with shopper behavior and channel preferences.
Custom development challenges in RDBMS, the overall Return on Investment (ROI), ease of integration and compatibility, and increased performance expectations even with explosive data growth rates are some factors that are driving retailers to explore alternate data handling approaches.
Is RDBMS Losing Out to NoSQL?
The conventional data handling techniques of RDBMS cannot address the needs for fast response, bulk data handling, and ease of data management. In such scenarios, data retrieval and updates for a particular item with specific attributes could take a toll on time, cost, and quality. This is because of database structure complexities and the subsequent joins required to achieve the desired results. Retailers are therefore exploring newer techniques to support item attribute enrichment in real time.
Next generation data handling techniques such as NoSQL, which have evolved over the years, show promise in helping retailers to leverage the big pool of available data.
Is Graph Schema Here to Stay?
The traditional approach to a data structure definition relied on the RDBMS concepts of context and entity relations. NoSQL databases provide a new dimension to data structures—each database approaches data structure definition in a unique way (referred to as ‘schema-less modeling’), opening up opportunities with a range of benefits. Graph databases, characterized by their maturity and robustness in representing data in the form of a network of related nodes, are gaining traction in the NoSQL domain. The data hierarchies in graph schemas use graph structures with nodes, edges, and properties to represent and store data.
By definition, a graph database is any storage system that provides ‘index-free adjacency’. In the context of retailing, one might not know the item data upfront at each point in time. In addition, intricacies of data may not be fully established when it is created; for example, users adding dynamic content in the form of reviews, tags, and so on. Graph databases address this need perfectly. As they are naturally additive, they enable the addition of new kinds of relationships, nodes, and attributes to an existing structure without disturbing existing queries and application functionality. This renders them ideal for representing connected data. Retailers employing graph databases will have the flexibility to add or remove attributes for any item, at any time in the item’s lifecycle. Graph databases present a strong case for Master Data Management (MDM) and can assist in addressing some of the key challenges that retailers face in Item Management, especially with increasing attributes.
Graph Databases: The Road Ahead
The early adoption of graph databases by organizations such as Cisco and Pitney Bowes to manage their product and organization data lends credibility to graph database’s promise of superior access to complex data. Organizations are evaluating the usage of graph databases to interpret customer opinions in social media and gather meaningful insights into customer preferences. There are several players from the open source community competing with each other to build better versions of graph databases, with Titan and Neo4j showing potential to lead the pack.
The opening up of new channels and extension of the marketplace are forcing retailers to seek smarter ways to manage items and retain the right attributes consistently throughout the item lifecycle. With an increase in the number of entities in the supply chain, a new Item Master capable of capturing new attributes of different types with auto-evolving support becomes crucial. NoSQL approaches show promise in reworking the approach to data structures, offering a possibility of revamping Item Management, specifically through graph schema.
Although graph schemas have had limited penetration, the developer community is working hard to come up with Application Programming Interfaces to enhance the scope. This should lead to frictionless development of new systems with simplified maintainability while providing complete control. Retail-specific use cases of graph databases include:
- Social media traversals for opinion mining
- Recommendations engines with network structure modeling around ‘item’ and ‘user’ nodes
- Geospatial mapping on graph schemas to boost store intelligence
- Shopping basket pattern analysis to bundle and co-market products together
In the journey ahead, graph databases will be embraced in areas beyond Item Management. They will extend to other MDM elements including customers, suppliers, departments, geographies, sites, cost centers, and business units, given their ability to support better adjacency-based traversals and configurable lookups.
Published in the TCS Retail Forum Journal – Issue 2 (PDF, 1.76 MB)
Get More Out of Box: Macro Space Planning Based on Your Customer Behavior
What’s Your Bet on Omni-Channel Pricing?
Devising Your Online Assortment Game Plan