Customer Analytics is the process of analyzing historical customer behavior (Descriptive Analytics), such as how, what, and where they shop to derive actionable customer segments. This process helps companies predict future outcomes (Predictive Analytics) and personalize their customers’ experience by targeting them and showing them highly relevant offers at the right time and place (Prescriptive Analytics).
The most fundamental part of Customer Analytics is Descriptive Analytics. With Descriptive Analytics, SaaS and E-Commerce companies can interpret their historical customer data to learn what worked and what hasn’t worked.
Descriptive Analytics also helps visualize changes in trends to highlight growing parts of the business and how product lines perform across seasons. SaaS companies use Product Analytics for their Descriptive Analytics to get in-depth behavioral analysis of their product and across all touchpoints.
To implement Descriptive Analytics, companies start by collecting behavioral data in an actionable format.
This behavioral data consists of actions that users take such as viewing a page, signing up, submitting a form or engaging with an email (active action) and activities that the company takes to engage a user such as calling the customer, sending them an email or initiating a chat conversation with them (passive actions).
Every action tracked will have three required components:
- Who: A unique user-id (e.g., email address, phone number, or cookie)
- What: The type or name of the action taken and all the metadata associated with it. (e.g., pageview with page URL and title)
- When: A timestamp of when the action occurred
As the company collects this data, they can start generating analytics reports by aggregating it. Data aggregation can be done in different ways to answer different types of questions.
Data can be filtered to specific actions and aggregated by date/time units to visualize the number of people who completed a particular action per day, week, or month over a period of time. This type of analysis will reveal how user engagement is progressing over time. For example, a Trends Analysis can show the number of weekly user signups over the previous 12 months.
Trends Analysis is usually the first step in the analysis to learn what happened. But to develop a deeper understanding of why events are trending in a particular direction, companies need to leverage customer journeys.
Customer Journey Analysis
Companies can also aggregate user data to analyze the effectiveness of actions in the context of a user’s journey. For this type of analysis, having consistent unique identifiers for users is critical to tie together the series of steps taken by a user throughout their journey.
By aggregating the customer journey data, companies will map and visualize the funnel and identify precisely where the users are getting stuck and aborting the user journey. Companies will be able to reveal the attribution of any initiative they launch towards the customer experience.
This type of analysis can help answer questions such as:
- Is viewing the pricing page deterring people from signing up for the service?
- How many emails should marketing send to re-engage a disengaged user?
- At what point in the journey is the company losing most of its users?
- Which campaigns are the most effective for acquiring successful customers?
- What is the retention rate of the new cohorts, and how does it compare to older ones?
When analyzing customer journeys, companies use Product Analytics tools to answer these types of questions. Tools like SQL databases can be very challenging to use to answer these questions.
Descriptive Analytics relies exclusively on historical data. To take things further, companies can leverage this historical data to predict the future using Predictive Analytics.
Predictive Analytics can predict the performance of the business in the future. (This is used heavily in business analytics.) Those predictions are never 100% accurate, but they can be reliable enough to forecast near and far future outcomes. While companies have used statistics and algorithms for Predictive Analytics since the 1980s, the techniques have evolved exponentially in the last decade using deep learning, a form of machine learning that uses neural networks.
Statisticians generally rely on historical data to find correlations between variables. They then develop models or equations that can predict outcomes, sometimes based on sets of assumptions. One of the most famous examples of statistical modeling is Moore’s Law. Gordon Moore, the founder of Intel Corporation, predicted that companies would pack twice as many transistors on the same sliver of silicon every two years. That observation has held until today.
The most used statistical modeling algorithm is Regression, where companies develop correlations between one or more variables to predict an outcome.
When it comes to Customer Analytics, analysts extract the customer and environment attributes that can influence an outcome. Using regression analysis, they build models to test and project outcomes based on sets of assumptions. For example, a company can produce a model around the following variables: age, gender, product category, and season with the measured outcome being revenue generated. The company will use this model every season to predict what to merchandize and who to market to in order to maximize the ROI on the inventory and advertising spend.
In the previous decade, we witnessed the rise of a new era of machine learning. Machines can now learn a lot faster from historical data and develop models beyond human abilities thanks to a new Deep Learning technique, also referred to as Neural Networks.
These Neural Networks are also trained with historical data. Unlike traditional statistical models, they can develop correlations between hundreds of variable inputs and predict the probability of an outcome with a high level of accuracy.
While this method frees us humans from scratching our heads and spending months developing and testing models, it’s worth highlighting the following problems:
- The more variables we’re training the neural network on, the more data we will need to develop an accurate model. This makes it a lot harder for companies that don’t have enough data to take advantage of this technique.
- The prediction models generated are like a black box. While they do a good job making predictions, it’s impossible to understand the reasoning behind those decisions. Since the data fed into these systems can be limited due to human bias, the predictions made by neural networks will inherit that bias, which is difficult to detect.
- There are no transparent best practices behind how neural networks should be designed. A neural network consists of layers, nodes with different shapes and colors (to simplify things). It’s more of an art than a science to architect a neural network that will work for the dataset available. This means there’s quite a bit of trial and error involved when architecting the most optimal neural network that can be trained with minimal data and produce the highest accuracy possible.
Recurrent Neural Networks
A more promising approach to deep learning for Customer Analytics is Recurrent Neural Networks (RNN). This neural network architecture uses the customer journey data instead of summarized user attributes to generate behavioral patterns. While this approach has been successfully solving many problems around sequential data, it hasn’t been fully leveraged for customer analysis yet, but it won’t be long before we get there.
Unlike Predictive Analytics, Prescriptive Analytics makes short-term predictions and on an individual user basis. Companies use Prescriptive Analytics to prescribe the next best course of action for every individual user to personalize their customer experience and show them relevant products at the right time and place.
With Prescriptive Analytics, companies can also optimize their teams’ productivity by predicting a customer’s likelihood of converting or churning. Systems can rank users based on their likelihood to convert, and salespeople can reach out to help seal the deal. Customers showing signs of churn can receive attention from the customer success team to diagnose the problem and remedy it.
When a user visits an online store, companies must start building their profile immediately. Even when they’re not explicitly providing any information about themselves, companies can learn a lot about their preferences and intents from their behavior. Interestingly enough, behavioral data can reveal information about users that they don’t even consciously know about themselves. Here’s a list of activities that companies should be tracking:
- What are they searching for?
- Which images are they clicking on?
- What category of products are they looking at?
- What colors do they look for?
Using this collected data, the company will then be able to develop attributes that can be used for personalization.
While systems can be automated to personalize experiences for customers, companies that follow a high-touch model - usually B2B companies - can improve their productivity by identifying who to sell to, in order to convert more customers, and who to give more attention to, in order to retain more customers effectively.
For example, a company could look for users who visit the Pricing page at least three times and trigger an alert to a salesperson who can start a live chat conversion to guide the user through the sales process.
On the other hand, a company may monitor app logins and compare the logins over the last 30 days to the previous 30 days. If a significant drop is detected, a customer success person will be notified. They will then reach out early to help reevaluate the customer’s needs before the customer pulls the plug.
Get Started with Customer Analytics
To get started with Customer Analytics, companies must put the Descriptive Analytics processes in place correctly before they start exploring Predictive and Prescriptive Analytics. It all begins with a solid data foundation.