Customer engagement has long benefited from data and analytics. Knowing more about each of your customers, their attributes, preferences, behaviors and patterns, is essential to fostering meaningful engagement with them. As technologies advance, and more of people's lives are lived online, more and more data about customers is captured and made available. At face value, this is good; more data means better analytics, which means better understanding of customers and therefore more meaningful engagement. However, volumes of data measured in terabytes, petabytes, and beyond are so big they have spawned the terms "Big Data" and "Big Analytics." At this scale, there are practical considerations that must be understood to successfully reap the benefits for customer engagement. This article will explore some of these considerations and provide some suggestions on how to address them.

Customer Data Management (CDM), also known as Customer Data Integration (CDI), is foundational for a Customer Intelligence (CI) or Customer Engagement (CE) system. CDM is rooted in the principles of Master Data Management (MDM), which includes the following:

• Acquisition and ingestion of multiple, disparate sources, both online and offline, of customer and prospect data
• Change Data Capture (CDC)
• Data cleansing, parsing, and standardization
• Entity Modeling
• Entity relationship and hierarchy management
• Entity matching, identity resolution, and persistent key management for key individual, household, company/institution/location entities

• Rules-based attribute mastering, "Survivorship" or "Build the Best Record"
• Data lineage, version history, audit, aging, and expiration

It's useful to first make the distinction between attributive and behavioral data. Attributive data, often referred to as profile data, is discrete fields that describe an entity such as an individual's name, address, age, eye color, and income. Behavioral data is a series of events that describe an entity's behavior over time, such as phone calls, web page visits, and financial transactions. Admittedly, there is a slippery slope between the two; a customer's current account balance can be either an attribute or an aggregation of behavioral transactions.

MDM typically focuses on attributive data. Being based on MDM, the same is true for CDM. Personally Identifying Information (PII) such as name, email, address, phone, and username are the primary drivers behind identity resolution. Other attributes such as income, number of children, or gender are attributes that are commonly "mastered" for each of the resolved entities (individual, household, company).

Enter Big Data. As more devices are developed - and adopted - that capture and store data, huge quantities of data are generated. Big Data, by definition, is almost always event-oriented and temporal, and the subset of Big Data that is relevant to a CE system is almost always behavioral in nature (clicks, calls, downloads, purchases, emails, texts, tweets, Facebook posts). Behavioral data is critical to understanding customers (and prospects). And, understanding customers is critical for establishing meaningful and welcome engagement with them. Therefore, Big Data is, or should be, viewed as an invaluable asset to any CE system.

Further, this sort of rich, temporal behavioral data is ripe for analytics. In fact, the term Big Analytics has emerged as a result. Big Analytics can be defined as the ability to execute analytics on Big Data. However, there are some real challenges involved in executing analytics on Big Data, challenges that drive the need for specialized technologies such as Hadoop or Netezza (or both). These technologies must support Massively Parallel Processing (MPP) and, just as importantly if not more so, they must bring the analytics to the data instead of bringing the data to the analytics. Having recently completed a course for Hadoop developers (an excellent course that I highly recommend), I have a heightened appreciation for the challenges related to managing and analyzing data "at scale" and the need for specialized technologies that support Big Data and Big Analytics.

A few significant points regarding Big Analytics should be considered:

1. Big Analytics allow the build of models on an entire data set, rather than just a sampling or an aggregation. My colleague, Jack McCush, explains: "When building models on a small subset and then validating them against a larger set to make sure the assumptions hold, you can miss the ability to predict rare events. And often those rare events are the ones that drive profit."
2. Big Analytics allow the build of non-traditional models, for example, social graphs and influencer analytics. Several useful and inherently big sources of data such as Call Detail Records (CDRs) generated from mobile/smart phones and web clickstream data both lend themselves well to these models.
3. Big Analytics can take even traditional analytics to the next level. Big Analytics allows the execution of traditional correlation and clustering models in a fraction of the time, even with billions of records and hundreds of variables. As Revolution Analytics points out in Advanced 'Big Data' Analytics with R and Hadoop, "Research suggests that a simple algorithm with a large volume of data is more accurate than a sophisticated algorithm with little data. The algorithm is not the competitive advantage; the ability to apply it to huge amounts of data-without compromising performance-generates the competitive advantage."

Big Data is great for a CE system. It paints a rich behavioral picture of customers and prospects and takes CE-enabling analytics to the next level. But what happens when this massive behavioral data is thrown at a CDM/MDM system that is optimized for attributive data? A "basketball through the garden hose" effect might occur. But this doesn't have to happen; there are ways to gracefully extend CDM to manage Big Data.

The key is data classification. Attributive, or profile, data is classified separately from behavioral data. While both contain Source Native Key (e.g., cookie-based visitor id, cell phone number, device id, account number), attributive data can be structured only. Behavioral data, on the other hand, can be structured and unstructured and contains no PII. Big Data almost always falls under the behavioral category.

Importantly, behavioral data requires different processing than attributive data. Since the processing is different, the two streams can be separated just after ingestion, like a fork in the road, with the attributive data going one way and the behavioral data going the other. This is the key to integrating Big Data into a CDM-MDM system without grinding it to a halt. To be fair, the two streams aren't completely independent. The behavioral stream will typically require two things from the attributive stream: Dimension Tables and Master ID-to-Natural Key Cross-References - both of which can be considered as reference data.

Dimension Tables
For example, the "subscriber" dimension table may be required in the Big Data world so that it can be joined to the "web clicks" table. This is done in order to aggregate web clicks by subscriber gender, which only exists in the subscriber table.

Master ID-to-Natural Key Cross-References
Master IDs are created and managed in the CDM-MDM world, but they are often needed for linkage and aggregation in the Big Data world. Shadowing cross-references that map master IDs, such as master individual id, to "source natural keys" into the Big Data world solves this problem.

The two classifications of data are separated into two streams and processed (mostly) independently. How do they come back together? One way this architecture works is that both streams, attributive and behavioral, contain a "source natural key." This is a unique identifier that relates the two streams. For example, web clickstream data typically has an IP address or a web application-managed, cookie-based visitor ID. Transactional data typically has an account number. Mobile data will have a phone number or device ID. These identifiers don't have to mean anything, per se, but are critical for stitching the two streams back together.

It's not just the dimensionalized, aggregated data that is reunited with the profile data, but also the high-value, behavioral analytics attributes (predictive scores, micro-segmentations, etc.) created courtesy of Big Analytics. The attributive data is now greatly enriched by the output of the Big Data processing stream. And, to get things really crazy, these enriched behavioral analytics profile attributes can be used as part of the next cycle of matching; similar, complex behavior patterns can help tip the scales, causing two entities to match that might not have matched otherwise. In the end, CDM-MDM and Big Data can live together harmoniously; Big Data doesn't replace CDM-MDM, but rather extends it.