Event Tracking

What is event tracking?

Event tracking involves recording the interaction between a user and a digital interface. The ‘events’ captured describe user behavior; that’s why ‘event data’ is often called behavioral data.

This type of data has been used by some of today’s most successful companies to dominate their markets. Organizations like Amazon, Netflix, and Facebook have used behavioral data to gain a deeper understanding of their users, shape their products and/or services, and inform their marketing.

And whilst the effective use of behavioral data has been the preserve of tech giants for a number of years (mostly because of their access to resources), the emergence of new, innovative technologies is enabling organizations of all sizes to compete. With the right solutions, companies are able to ‘do more’ with behavioral data, and power advanced analytics and ML/AI use cases.

Empowered with rich, high-quality behavioral data, they’re able to deliver personalized product recommendations, dynamic pricing models, real-time fraud detection, churn reduction, credit scoring, basket analysis, help desk enablement, and supply chain optimization, to name just a few applications. With the right event-tracking tool, there are limitless possibilities on how to use behavioral data.

What is an event?

An event is defined as an interaction that can be observed at a particular point in time. Generally speaking, this interaction happens between a human and a digital interface (websites, apps, wearable devices), but it can also extend to include machine-to-machine interactions.

Here are some examples of events:

• Clicking on a blog post
• Pausing a video
• Submitting a form
• Searching for a product
• Adding an item to a shopping basket
• Sharing an article
• Visiting a physical store, and using a discount card
• Checking in baggage at an airport
• Starting a run on a fitness app

What is an entity?

An entity describes the environment that an event took place in. It adds rich, granular contextual information to an event, helping to drive actionable insights.

Here’s an example of the kind of contextual information (properties) contained within an entity (in this instance, a ‘blog entity’):

• User ID
• The device of the user
• The web page where the click occurred
• The category of the blog piece which includes the link
• The publish date of the blog which includes the link
• Engaged time on the blog which includes the link

It’s worth noting at this point that the whole concept of entities is unique to Snowplow (a Behavioral Data Platform – more info on this below).

Snowplow believes that the event/entity model of structuring behavioral data is the most effective way of making it more usable for advanced data applications. This is because it helps to maintain data meaning, ensure data quality, and maintain data governance across an organization; in essence, it helps you to drive more value from behavioral data.

To learn more about the structure of Snowplow data (and the importance of entities), watch this video here.

What does event data look like?

Event data is often expressed in JSON, using attribute-value pairs and arrays. The exact format of event data depends on what tool you’re using to track it, whether events are inbound or outbound, and whether they’re flat or nested. Typically, event data is represented in a one-row-per-interaction table in the data warehouse.

Here’s an example of how that would look, showing a single event:

An overview of event-tracking tools

There are many tools which have an SDK to generate event data to power reports or functionality from within their own platform. Most allow you to export event data to cloud data warehouses (for further analysis or to power a separate data app), but they’re not specifically built for it.

Snowplow BDP is currently the only tool on the market which creates data specifically for a data warehouse, lake, or stream.

Here’s an overview of what it is, alongside a non-exhaustive list of other tools/platforms which capture event data:

• Snowplow –Snowplow is an open source behavioral data platform (BDP), which generates, governs, and models event-level behavioral data. It’s an extremely flexible solution which enables data teams to reliably track events across multiple sources (in a single, consolidated format) for use in advanced analytics, AI, and ML.
• Google Analytics (GA) – By far the most used web analytics platform, GA collects, processes, stores, and analyzes behavioral data; it’s an end-to-end packaged analytics platform. GA4, which is replacing Universal Analytics, operates on an event-data model.
• Segment – A customer data platform (CDP), Segment’s primary strength is that it makes customer data integration simple; it supports a broad set of sources and destinations, and promises user-level analytics.
• Mixpanel – An end-to-end platform dedicated to product analytics. Although Mixpanel has an SDK to generate event data, it’s gradually improving support to ingest (Snowplow) data from the warehouse.

What to look for in an event-tracking tool

• Security – Data Privacy regulations are evolving. The application of GDPR and CCPA has a direct impact on how companies track, store, and use behavioral data. The collection and transfer of personally identifiable information (PII) is particularly relevant (as demonstrated recently with the French and Austrian DPAs decisions to enforce ‘Schrems II’). This means that whatever tool or platform you use for event-tracking, it needs to be able to adapt to changing legislation – here, control and visibility on what you track, and the ability to host your pipeline in your own private cloud account (preventing access to third-parties) are key.
• Flexibility and scalability – As a company, your behavioral data needs are likely to change over time. You may want to start by building a deeper understanding of your customers through advanced analytics, before deploying AI/ML models and driving automated actions through a data app. The behaviors you track and how you use them should be able to flex alongside changes in your business, and your event-tracking tooling should allow for this.
• Data quality/Observation of data – A key challenge for data teams today is how to build trust in their work across the wider organization. One of the main obstacles in doing this is ensuring that only high-quality, reliable data reaches data consumers within the business; if there are discrepancies between datasets and there’s no ‘single source of truth’, trust in your data erodes. And whilst a >whole category of tools exists to deal with this, your event-tracking tool should also take account of data quality and observation – up-front data validation and failed events monitoring are important features.

Event-tracking<>Data Creation – Example architecture

‘Data Creation’ refers to the process of purposefully ‘creating’ data for advanced analytics, AI, and ML data applications. Event data that has been ‘created’ stands in contrast to ‘exhaust data’; this is data that is a bi-product of another tool, and was collected for a purpose other than what it’s intended for. >

The problem with ‘exhaust data’ is that it needs to be wrangled for use in downstream applications. This takes a significant amount of time and effort, preventing data teams from working on more meaningful, impactful projects. Even once it has been formatted and cleansed, furthermore, its quality can still be called into question – an aspect which prevents it from being democratized across an organization.

Data creation is a response to this, and involves the generation, enhancement, and modeling of behavioral data.

Here’s an example architecture showing how behavioral data can be ‘created’, and used to power machine learning data applications and ‘unbundled’ CDPs.

Understand, Predict, Automate: The event data rocketship

More organizations are becoming aware of the huge potential behavioral data offers. They want to create high-quality, granular behavioral data in order to gain a deep understanding of their customers, and use this information to drive the growth of their businesses.

Having worked with many companies over the past decade, we’ve found that the process of ‘doing more’ with behavioral data follows three distinct steps: Understand, Predict, Automate. Here’s what we mean.

Understand with Advanced Analytics

The first stage in driving value with behavioral data is to gain a deeper understanding of your customers. By tracking behaviors across all digital touch points, you can build a comprehensive picture of how they interact with your brand and what motivates their actions. This information can then be used to inform decisions across the business.

Predict with ML and DL

Once you have a deep understanding of your customers’ historic behavior, you can begin to predict their future behavior; machine learning and deep learning models can be trained with behavioral data in order to identify signals of future actions. Again, this can help to inform decisions around the business.

Automate with a Composable CDP

Automations can then be set up to take actions and influence customer behavior by bucketing users based on their propensities to take a specific action. ML-driven product and content recommendations, dynamic pricing and paywalls, and personalized messaging in real time are just a few examples of automations you can deploy at this stage.