We're kicking off a new series about evidence-based marketing and its framework. This will be a six part series, the first of which will cover the integration and enrichment of event and customer data. Subsequent posts in this series will cover the entire evidence-based marketing framework; including modeling attribution and customer lifetime value, modeling high risk or high reward behaviors and activating audiences based on those models, forecasting enterprise performance, and modeling allocation of resources.
At the center of the evidence-based marketing framework is a scientific testing and measurement process, which acts as the flywheel for continuous improvement within the evidence-based marketing framework.
So to kick this off, what I'd like to do is set aside the framework for now and instead look at a timeline. Time is one of the essential elements to evidence-based marketing, and it's largely forgotten within many traditional marketing data constructs. Let's look at three examples of timelines of events and how they are structured in an evidence-based marketing data construct.
A Classic Retail Case
First, a classic retail case; a single customer interacting with your advertising platforms and your e-commerce website. On this customer's path to purchase, we first see an interaction with a Facebook display ad, they receive an email, they click on that email, they receive a direct mail piece, they go to Google and search for your product, and then they end up converting on the website.
There are two types of data you want to collect from these events as they occur. The first is an exact or close approximation of the time of purchase and the second is a set of metadata that will allow you to do powerful analytics at a later time. Since this is an e-commerce purchase example, we have a lot of exact timestamps, the only exception is the direct mail piece the customer received. Most often, we can only have an approximate date that the direct mail piece arrived at the home so we will have an approximate timestamp representing when that consumer was actually presented with the impression of the direct mail piece.
Metadata is information you capture from the body of the event. For most digital programs, like Facebook's advertising platform, you can capture very specific information about the campaign like the Campaign ID, Ad Set ID, and Ad ID. With this information captured and stored for that click that that user engaged in, you can get back to all of the information in Facebook about the financials of the campaign, the relevance, how the audience targeting was set up. This shows how much rich metadata is associated with this particular event.
You can find similar information in all digital programs. An interesting one to think about is the purchase event. What are you capturing for the purchase event today? A lot of information related to the specifics of the transaction, so product line item detail, item identifiers, and all of the information associated with couponing. One of the things that's rarely captured at the transaction or item level is the appropriate coupon ID, which is a very powerful piece of information. Think about all of those detailed metadata items that could inform questions later on in an evidence-based framework.
Before we move on from the classic retail case, I want to point out that this is one of the primary reasons that an event-based framework is not implemented at marketers today. The data is very complicated and it's spread around to a lot of different platforms. In this simple case for this one customer, data is retrieved from the Facebook ad platform, the ESP, the web analytics platform, a third-party direct mail fulfillment house, the Google ads platform, and the in-house e-commerce or CRM system.
Unhappy Customer Case
Let's move beyond the classic retail case of a path to purchase and look at a different interesting case for a consumer. We're going to look at a consumer who had a purchase and then subsequent to that purchase, there was an inbound call to the customer service center, a product return, negative feedback from customer on the website, and finally, a series of sent emails that were never opened or interacted with.
Similar metadata is captured for each of these events but there are three unique characteristics to think about on this type of event data. The first is the purchase information that you're capturing. We previously established that you're capturing the item ID and other information about the product during the purchase, many retailers don't capture the same level of granularity with their returns.
From an analytic perspective, that means they're going to have difficulty understanding what the causality was related to the customer dissatisfaction to analyze later. It's very important to make that connection between the return event and the purchase event, even in an omnichannel retailer where that might occur in different channels.
The second type of information that you should be capturing is related to non-structured events, like the phone call and web feedback from this example. When you think about each of those, what kind of information would you like to capture for those events? We would want to capture the full body of the text associated with those events, the category of the issue, and maybe even the sentiment of the customer feedback. These types of information can be extremely valuable for analysis later on.
The final concept I want to discuss is the lack of an event, which is sometimes a powerful predictor later on. In this particular case, the customer didn't open or click on any of the emails sent after their purchase event. These are powerful indicators when we look at predictive analytics and segmentation built on top of this event data framework.
Social Media Influence Case
The final example is quicker than the other two and it's driven by the fact that many retail and marketing organizations today are focused on recurring revenue models, especially those that are driven by subscriptions. For this example, let's look at a consumer who interacted with us via Instagram at first and then, as a result of that interaction, subscribed for the free trial of our service.
For a short period of time after they signed up, they had multiple periods of engagement. We sent them a clever email which was successful in driving additional engagement and before the end of the free trial, they subscribed to our service.
Most of the event data captured in this example has been described before. For Instagram and the email, the type of event metadata we want to capture has already been described. However, this example provides an opportunity to demonstrate that the engagement event can be a nested timeline in and of itself. This nested timeline can represent all of a user's sub-engagement with the content on your web or e-commerce site or even an application that you've developed.
Now that we have a foundational understanding of how to integrate, enrich, and structure the event data for your organization, let's take a look at how to use that event data to recreate a snapshot of every customer at any historic point in time. To get started, let's recreate the path to purchase that we used in the first example.
We have someone who engaged with a Facebook ad, they were sent an email, they clicked on that email, they received a direct mail piece, they searched on Google, and finally, they converted via your e-commerce website.
For your organization, you may have defined tens or even hundreds of characteristics to describe each of your individual consumers. Many times these characteristics are described as recency, frequency, or monetary characteristics. Recency is the amount of time that has elapsed since the customer's last purchase. Frequency is the number of times, either historically, cumulative, or over a rolling window of time, that the customer has purchased. Monetary will be either sum value of their past purchases or an average purchase value for that consumer.
These characteristics are sometimes extended to product and channel preferences, where product is the customer's preferred product class or family like luxury handbags and purchase channel is the customer's preferred way to purchase, such as in-store.
There are many other characteristics that we will discuss in the rest of this series, like propensity models, promotional sensitivity, velocities, such as value or email engagement velocity, and a wide variety of demographic enrichments. For this post however, I really want to make clear how these characteristics can be retrieved at any point in time for any given customer using the event data structure we've discussed.
Let's say we have a customer characteristic that defines a customer's product preference based on the last five times they've come to the website via a Google product listing ad or a non-branded search and what the term or product was that led them there. For this consumer, it was jewelry. Let's move forward in time and as this customer moves along their event path, she comes to her next Google search which is for footwear, making three of the last five searches for this customer footwear related and bumping jewelry out of the preference as she moves along their path to make another purchase.
Throughout this examples, we've been seeing things adjust. Recency has been increasing based on the amount of time that has elapsed from her last purchase. In this case, we're now going to see frequency increase by one because of her recent purchase. There may be an adjustment, either up or down, to average purchase value depending on the total cost of this particular purchase. Perhaps handbags are bumped from the preference and footwear, based on her last demonstrated Google search, becomes the product that is preferred by this customer. Lastly, her channel preference changes from in-store to the web as this recent purchase was made on the e-commerce site.
I hope this introduced you to the concepts of integration and enrichment of event data and how that event data foundation can be used to recreate a snapshot of each customer at any historic point in time. These two concepts are fundamental to the upcoming posts in this series. Stay tuned for our next post on marketing attribution.