Delivering deep-link analysis

How you can harness the power of graph analytics to achieve a 360 customer view without rebuilding the entire IT system. By Martin Darling, VP EMEA, TigerGraph.

  • 2 years ago Posted in

Achieving a 360-degree view of their customer base is the dream for many companies wishing to boost sales, drive operational efficiency and ultimately attract more customers. However, they often find it difficult to ask complex questions about the business because their data is trapped in siloed, legacy relational databases that lack the flexibility and power to perform deep link analysis.

Graph databases solve this fundamental data link problem and introduce a powerful new computational capability – graph analytics.

Relational databases struggle with deep-link analysis because of the very nature of the way data is organised in lists. In list format, drawing a link between one record in one table and another record in another table involves a table join, but relational databases become increasingly inefficient as the number of table joins rises above three or four, especially if the tables are large and the number of queries is high.

Graph databases can connect data in multiple relational databases and organise it into a set of vertices representing business objects such as a customer, payment or order connected by edges that represent the relationships such as this order was placed by this customer. This ability to map the data into pre-connected business entities without disturbing the original datasets effectively turns it into an analytical layer, sitting above the original data.

What’s more, it can maintain query performance even as it grows to billions of edges and vertices because, unlike a relational database, it doesn’t have to load entire tables into memory to perform a query. Only the data that is required for the query need be loaded as it ‘hops’ from one node to another, so queries of 10 hops or more are no problem for a graph database.

In-query processing adds another dimension to graph analytics, enabling queries to be constructed in such a way that it is not necessary to hop in and out of the database to complete a function.

This is a fundamental shift in the way we think about data storage, and it enables the use of a host of computing techniques or algorithms which are incredibly powerful in graph but virtually impossible within a traditional relational database.

Mapping your data

There is nothing inherently wrong with relational databases for many business applications. Developed in the 1970s and 80s, relational databases are well suited for the storage and retrieval of data and simple records management.

However, to achieve a 360 customer view across multiple databases and unearth new business intelligence, you need to be able to ask questions about the relationships among different pots of data which involves bringing together disparate data held across marketing, sales, accounting, customer service and other business functions.

Organisations are harnessing graph analytics to create a digital map of their customer data while leaving existing databases undisturbed. The result is a 360 customer view across accounting, sales, customer service, marketing and other domains – but it doesn’t end there. Businesses are using graph analytics to:

1. Boost the conversion rate for digital channels

2. Identify new customer segments to target for future growth

3. Reduce the cost of acquiring new customers

4. Shorten the time of the buying journey

5. Access new customer segments

Connect datasets and pipelines

A distributed graph database enables you to connect internal and external datasets and pipelines to extract invaluable business intelligence in real time, as demonstrated by Xandr, the advertising and analytics division of AT&T’s WarnerMedia. It works across 15 WarnerMedia channels including Cinemax, CNN, HBO, and TNT, each holding data on millions of customers. The challenge for Xandr was creating a unified picture of all of these customer databases to deliver a seamless experience across the portfolio for both viewers and advertisers.

They created Community, an advertising platform which would be able to disambiguate user data from multiple platforms. The goal was to create unified entities across the multiplicity of data sources and deliver a joined-up advertising experience – even as viewers jumped between their personal devices (mobile, tablet, laptop) and across channels. To achieve this, Xandr built the first and largest identity graph of its kind in the advertising industry using a graph database that scales horizontally to accommodate more than five billion vertices (business entities such as users, devices and identifiers) and seven billion edges (relationships among entities) and ingests a billion updates per day. Graph analytic capabilities, such as entity resolution and centrality algorithms, stitch together identities across the separate databases to create a view of households and devices that is both unified and granular.

Now Xandr not only knows how many times an ad has been seen on a particular device, it can tell how many times that ad has been seen across all the viewer’s devices – and target advertising accordingly. And the result is clear: Xandr wins advertisers from the competition by offering higher quality data and more precise targeting.

Target customers with personalised real-time email

In online retailing, product recommendation is a never-ending battle for accuracy and timeliness. Studies show that businesses can attract customers by doing a better job of recommendation than their competitors. Yet, despite this, 74% of marketing leaders report they struggle to deliver customer recommendations to all their customers all the time. It’s not surprising then that Gartner says companies are actively looking for new solutions for this perennial problem.

Kickdynamic found the solution in graph databases. It’s an email marketing platform that helps more than 200 leading brands boost customer engagement and sales, and it chose to build its recommendation engine on graph rather than a relational database because of the superior ability to pull CRM data from multiple sources, connect sequences of 10 or more datapoints to extract meaningful business intelligence and deliver all of this at scale and at speed.

The ability to process queries in real time allows it to deliver live pricing based on customer preferences and product availability, capture key ‘business moments’ and deliver targeted recommendations – before the customer disappears. This resulted in increased engagement, brand loyalty and sales conversions – all thanks to graph analytics streamlining the process of building recommendation emails. “Kickdynamic knows that compelling, individualized experiences are the most effective way to create customer loyalty and drive revenue” confirms Gabriele Corti, Chief Product Officer at Kickdynamic.

In-database machine learning for fraud detection

Retail businesses everywhere struggle with fraud, but financial services company NewDay has found a way to automate a lot of its anti-fraud measures with the help of graph analytics. It has achieved a big jump in detections and prosecutions, turning the tables on criminals by identifying fraud at all stages in the credit card lifecycle including application fraud (trying to obtain credit cards with stolen personal information), transactional fraud and first-party fraud (fraud by existing customers).

With revenues of nearly £1 billion per year, five million consumers and an operation spanning the largest online retailers and best-known credit cards, as well as access to third-party fraud prevention and identity checking databases, NewDay had no shortage of data on which to draw. However, bringing all of this information together in a meaningful way was a time-consuming manual process that required fraud investigators to jump in and out of internal and external databases.

Using graph analytics, NewDay created a system that brought together myriad databases, both internal and external, empowering fraud investigators to speed up the analysis of complex frauds. In addition, sophisticated algorithms enable it to analyse more than 10 million transactions per month and identify suspicious cases while minimising false positives. Consequently, anti-fraud at NewDay is more intelligence-driven, allowing it to block cards and issue new ones more quickly and refer a greater number of cases to the police.

NewDay says that their graph-based system is still in its infancy but has already yielded a 10-15% uplift in the number of fraud cases being detected which the head of fraud prevention, Danny Clark, says is only the beginning.

With graph analytics, organisations can achieve a top-down view of customer data, generating new insights to grow and develop the business.

By Raja Rao, Head of Growth Marketing, Redis.
By Joe Beaumont, Head of Hospitality at Exponential-e.
By Simon Spring, Account Director EMEA, WhereScape.
By James Fisher, Chief Product Officer, Qlik.
By Dale Murray, CEO at SalesAgility.
By Mathias Golombek, CTO of Exasol.