Customer journey mapping has become very popular in recent years. Making it right can upgrade marketing strategy, boost personalized branding and offerings and result in increased revenue and marketing ROI.
To bring value to the business, it requires a healthy balance of qualitative knowledge of customer-facing functions about market and customers and quantitative insights. The last can be gained from an e-commerce platform, CRM and other market related external sources.
The purpose of this article is to share with fellow sales, marketing professionals and data scientists how to approach quantitative part of customer journey mapping, namely to answer questions:
- How many buyer personas do we have?
- What are their unique characteristics?
- How can we adapt the marketing strategy concerning buyer personas to increase marketing ROI?
- How accurately can we predict buyer persona from the first customer purchase transaction?
As most of the companies keep data related to customer journeys confidential, the use case will be demonstrated on Google Analytics Sample Dataset.
The sample contains obfuscated Google Analytics 360 data from the Google Merchandise Store, a real e-commerce store. It sells Google-branded merchandise.
The data is typical of what you would see for an e-commerce website. It includes traffic source, content, and transactional data.
How many buyer personas do we have?
It is not an easy question. What are the criteria to segment buyer personas? One approach is to compare customers using RFM analysis answering questions such:
- When recently did they buy (Recency)?
- How often do they buy (Frequency)?
- How much do they spend (Monetary value)?
The clustering algorithm can effectively group similar buyers to clusters which significantly differs from one another. It also suggests what should be the number of clusters (aka buyer personas).
Google Merchandise Store had 12028 transactions done by 9962 customers in 1 year. There are 3 naturally formed clusters:
- cluster 1 has the highest mean on frequency 3.2 and monetary value 1129 USD representing “frequent shoppers and high spenders”
- cluster 2 has the highest recency 273 days, low frequency 1, most probably representing “customer churn”
- cluster 3 has the lowest recency 87 days, low frequency 1, representing “new customers”
Clusters can be profiled across clustering variables by using their standardized scales which enable unit independent comparison.
What are the unique characteristics of buyer personas?
The profiling stage involves describing the characteristics of each buyer persona (cluster) to explain how it may differ in relevant dimensions as demographics, consumption patterns and behaviors. These are the variables not included in cluster analysis.
It is common that e-commerce platforms stores about 360 different features during each click of the visitor on the website. So which one of these 360 features are the most important differentiators of particular buyer persona from others?
This complex challenge can be solved using machine learning (ML) algorithms which have the ability to calculate “features’ weights” representing importance. Their value can range from -1 to 1 and the higher the absolute value of the weight the higher the importance of the feature.
It is also recommended to calculate “standardized difference of one-to-rest averages” which shows the magnitude and how much particular buyer persona differs from the rest of personas in feature of interest.
It simply a difference between the average value of a feature of customers classified to particular buyer persona and the average value of the same feature for the rest of customers classified as other buyer personas. The difference in averages is also standardized to enable comparison between features with different units.
Both feature weights and standardized average differences are plotted next to each other in a horizontal bar chart for the use-case of Google Merchandise Store. Only the top 14 out of 360 features are plotted.
“Frequent shoppers and high spenders” are mostly customers from the Americas, who visit Google Merchandise Store by typing its URL into a browser, using bookmark or landing from google search engine. They mostly buy apparel, office or drinkware.
“Customer churn” is represented by a group of customers who buy less apparel and office goods, rarely land to Google Merchandise Store from google search engine and are seldom from Americas. They browse more pages during one visit than the rest of the customers.
“New customers” browse fewer pages during one visit, are rarely from the Americas continent and seldom visit Google Merchandise Store by typing its URL into a browser or using bookmarks and rarely buy drinkware.
How can we adapt the marketing strategy?
This is how it may look like in case of the results above…
“Frequent shoppers and high spenders” should be target via advertising on google search engine with cross-selling opportunities among apparel, office and drinkware product category. The next step would be to build a recommendation engine to enable these cross-selling opportunities.
The objective here is to reduce “customer churn” by designing an effective incentive program that will target customers outside of the Americas via different channels that google search engine. The selection of the channels needs to be explored more in detail.
Browsing flow can be closer investigated to capture churn interests to better design effective incentives. This buyer persona can be further segmented and different incentives should be tested against each subsegment.
Non-Americas customer base seems to be the potential for gaining “new customers”. It is important to investigate which countries do they come from and channels they used to come into Google Merchandise Store. The advertising should be focused on no drinkware categories.
How accurately can we predict a buyer persona?
An effective marketing strategy is all about the personalization in branding and offerings.
Imagine having a business process that estimates the type of buyer persona from the buyer’s first purchase and targets him/her with the right branding message and offerings at the right time.
The use-case reveals that without having any machine learning model we would have a 33% confidence of classifying buyers to one of the three buyer personas correctly.
Testing the 6 different models the one called Gradient Boosting Classifier can classify buyer to buyer persona with an accuracy of 68%.
Imagine how your marketing ROI could increase if you would improve targeting efficiency by 35%!
Call to action
Have you ever done customer journey mapping? If not, maybe this is the time to start. “Why” was justified, “How” can be found here.