USE NETWORK GRAPHS TO GET MORE OUT OF DATA

Network graphs are doing wonders across industries.

They are used by banks and insurance firms, narcotics bureaus, revenue, military intelligence, and other government regulators to investigate money laundering, fraud, terrorism financing, non-obvious suspects, and the movement of funds and people.

The graphs depict the strength of the relationship between entities (persons, products, interactions) based on granular analysis. It then aggregates network performance to provide insights by uncovering hidden relationships and devising recommender systems. The key to network performance—reliance and predictability of decisions based on the customer-consumer-product network—are updates based on events such as customer purchases or digital footprints.

A network graph to study customer behavior will have three data layers:

• Foundation data, including data on top customers, maximum sales, sales based on time of day, periods of maximum or minimum footfall, market-basket analysis, and stores needing attention.

• Network relationship data, including products bought together in three orders, top or bottom N products by customers, by stores, discovering consumers basis purchase pattern.

• Relationship type, including customer clusters loyal toward a brand or product, which requires supple inventory; customers likely to shift from a product, which would require the product to be replaced; periodic product needs; and investigating the cause for change in relation.

The complexity of analysis increases as we move from one layer to another. Storing diverse, high-fidelity data together persistently will provide businesses with valuable insights; from a technology viewpoint, it saves the effort to tie different data together every time a new business problem or query emerges. This is the fundamental idea of having a visual network that helps personalize as well as aggregate information.

The three layers are filled with finer nuances to strengthen the customer-product relationship network. It is not just about technology; it is also about the expertise needed to contextualize data for an insightful network graph. Merely weaving a network may not help if the data is broken, aged, or imputed. Governed data will ensure trust and faith and better decision-making.

Network graphs have multiple features

Retailers should create a web of customer interaction and mine relationships by holding a complex network of data.

They should enhance the application that contains customer-product data with details of customer-customer relationship. If, for example, a retailer discovers that two friends come together on Sundays to shop periodically, the business could offer group discounts to improve loyalty.

As the system evolves, retailers should have details on a range of relationship: customer-customer, customer-consumer, customer-product, consumer-product, and product-product. Inferring relationship repulsiveness (customers repulsive to a particular product or brand) or snap points gives meaning to machine learning. Insights into supplier relations or home-to-store distances are add-on benefits.

However, just mapping relationships with a relational database is archaic, cumbersome, manual, and error-prone with computational overheads. The problem accentuates with growing underlying relationships and entities. A graph helps overcome these challenges by presenting immediate value in the form of nodes, relations, labels, and properties. A network graph has many features:

• Offers semantic insights to carve out personalized customer journeys to enhance customer experience

• Feeds data on relations into demand forecasting and streamlines upstream supply chain

• Manages allocation of products and time of delivery for online or offline purchases based on customer behavior depicted by available data

• Etches product seasonality and brand loyalty, providing details of a customer’s purchase frequency

• Provides need analysis of a customer household to serve the exact product from the nearest store

Graphs make personalized offers and products possible

Newer, precise revelations through the graph help retailers to increase wallet share.

Need and behavioral analysis will help them meet customers at the right time of the journey on a self-sustained network using the right channel. They can provide personalized products and better offers, thereby not missing opportunities. The graph can help a grocery retailer provide monthly nutritional recommendations based on a customer’s consumption pattern and provide fashion, jewelry, and eye-wear retailers insights on size, color, shade, texture, or fitness requirements.

Cross-sell, up-sell, mark-down strategies and tier promotion loyalty strategies could be laid out with the insights. Marketing strategies would get more contextualized to customer segments rather than being broad-based, thus limiting marketing expenses.

Customer intuitively see all that they are looking for—from the brand and size to quantity and color—on time, at the right place. Machine learning models over such layered data can make more accurate predictions. As a result, aggregate cost is optimized by knowing which product will get sold where, when, and in what quantity. The impact on profitability is instant. Over time, a detailed customer behavior study will help prevent customer churn and give rise to a new customer acquisition strategy.

A broken link—lack of customer cross-sectional data—between base data and advanced analytics yields inefficient results due to incomplete analysis, skewing the cost of investment (CoI), cost of operations (CoO), and return on investment (RoI).

For example, retailers do top-down demand forecasting (advanced analytics), going by historical sales performances of product categories across ranges. If the customer data is not included, there is the possibility of errors in projections. But, if base customer data is considered along with product data using graphs, then even bottom-up, data-driven forecasting becomes possible, making decisions more accurate. Similarly, all other advanced analytics models dependent on only products data could benefit from customer data and vice-versa.

Network data improves RoI

Network graphs give a fillip to retailers’ bottom line.

By adopting a layer of network data, retailers will not only reduce CoO but also propel RoI. Data science models for segmentation, lifetime value (LTV), channel preference, customer profitability scores, the propensity to buy (PTB), cross-sell and up-sell (CSUS) opportunities, product substitutability, loyalty and offer analytics, cost-to-serve (CtS) details, product rank, range, category, and affinity grouping will feed off the graph data points.

This will help anticipate customer behavior better, improve workforce productivity and sales performance, and optimize inventory and customer experience. Based on the vintage cohort of customers, newer offers and mark-downs can be planned, keeping profitability levers in check.

Retailers can provide product recommendations using cross-network analysis. Product correlation, inventory, past purchases, supplier information, logistics, and social data like ads clicked, and products searched could be easily unified. Due to its ability to traverse quickly through the network, a network graph also helps identify fake social media accounts (sockpuppets) like bot-run accounts.

A live-wire network

Today, data scientists take a disjointed and short-sighted approach to solve business problems.

They spin available data and apply crude data engineering to solve a problem. While retailers capture customer data, the insights gleaned from them—on consumer preferences and future needs—will feed into the product recommendation analytics. Any vintage cohort analysis cannot be done in isolation; it depends on the interaction of customers with products and channels. Integrating digital footfalls and bounces, expressions, and reviews and following a network of influencers will fine-tune the buying behavior.

Data managed centrally and visually will enable granular analysis for hyper-personalization, as weak links and strong bonds are easily decipherable. A comprehensive and inclusive aggregate analysis is possible as high-density nodes (areas of concern) are easily noted, enabling one-stroke decision-making.

It is a live-wire network with changes (household, preference, digital footprints) being accommodated with every transaction. This reduces the time invested in model training, building, and model fine-tuning, and the system also becomes auditable, ensuring retailers are not caught off guard by changes in consumer behaviors.