Looking for ways to boost the profitability of your e-commerce store?
We work with hundreds to retailers to convert their data to actions and sales using data science.
This articles covers the data science principles and shows you how you can do the same with your own business without writing a single line of code or hiring a data science team.
Throughout this article, we'll use Rachel's cosmetics online store as a case study. Rachel runs a Shopify store that sells cosmetics and skincare products in the US. For confidentiality reasons, I've distorted the numbers so these insights are purely illustrative.
Alright, let's get started.
Overview of data science in e-commerce
Before jumping in, let’s talk about how data science is used in e-commerce. Here's a schematic of how a data science engine for e-commerce looks like:
Strategies fall under two categories:
- Customer acquisition: Knowing how much it costs to acquire a customer from a given channel and how much the lifetime value of that customer is.
- Customer retention: Figuring out how to make customers buy more and how to keep them as a happy customer for the longest period of time.
Data science insights (green boxes in diagram above)
Customer acquisition and retention strategy stem from several insights that we can draw using data science:
- Customer lifetime value predictions
- Persona analysis
- Churn detection
- Customer segmentation
- Cohort analysis
- Trend analysis
We'll be going through a bunch of them later and show you how you can apply them to your own business.
Marketing Actions (blue and red boxes in diagram above)
An online retailer can automate the implementation of an acquisition and retention strategy with an engine that integrates with its database, email marketing, and ad platforms.
How much is a customer worth: predicting lifetime value?
Knowing how much a customer is worth helps you figure out how much you're willing to spend to acquire each customer.
Many businesses calculate customer lifetime value by looking backward and calculate how much a customer has spent on average. We all know dwelling on the past is not how we look at businesses. But businesses cannot really know how much a customer will buy in future.
This is where data science comes in. It works by using statistics to model how a customer buys. We assume that each time a customer buys a product from our store, there is a chance that he will buy again. We also know how much the value of that next buy will be by looking at the historical average. With this information, we model and predict how much we can expect to earn from a customer over his/her lifetime. Some key inputs to the model are the size of the first purchase, the size of repeated orders, the time between orders and a discount factor.
Thankfully, you don't have to do any of this data science to figure this out. Let's look at Rachel's business. With Metisa Acquisition Insight, you can see each customer is worth $10,058 to her business (wow, that's a lot, but that's because these numbers are not real). This means for every customer that Rachel acquires, she now can expect to receive $10,058 (in present value terms) of cash from the customer.
Let's think about this for a second. Rachel's average first order size (see the Acquisition by Channel table) is less than $5,000. Why is her customer lifetime value that high? The answer lies with the value proposition. The better your value - your brand, your service offering, your product curation - the more likely your customers are to make repeated purchases and the higher their lifetime value. This is why before you even go into optimizing for customer acquisition by channel, focus on getting your product offering right.
Customer lifetime value tells you how much each customer is worth to your business. The overall metric is a measure of how good your value proposition is: your customer service, brand values, products and more. This is probably the single most important metric that determines how easy or hard it will be to implement customer acquisition and customer retention strategies. It's always easier to sell a product that sells itself!
How much should I acquire a customer for?
Principle: customer acquisition cost (CAC) < lifetime value (LTV)
To run a profitable store, you should acquire a customer for no more than the present value of his/her future profits to you.
To express this more succinctly, we introduce the concept of customer acquisition cost and lifetime value. You have a profitable store when the customer acquisition cost is less than the lifetime value of that customer. In reality, stores have to be a little more conservative than that. In the case where lifetime value is based on the present value of sales, stores typically target a customer acquisition cost to lifetime value ratio ranging from 0.2 - 0.33.
Customer acquisition cost (CAC) = Cost per click (CPC) / conversion rate from click to sale
Going a step further, e-commerce stores acquire customers from many channels - AdWords, Facebook, social media, bloggers, partnerships etc. You usually pay per click for these advertising channels. How cost per click translates to customer acquisition cost depends on the conversion rate.
Let's look at Rachel's store. From Metisa Acquisition Insight dashboard, she sees that customers from Google have a much higher customer lifetime value than the ones acquired from Facebook. So far, Facebook has been her main channel for acquiring customers. She knows her customer acquisition costs for both channels from the Google Analytics and Facebook Ads dashboards. She does a quick ROI calculation and finds that Google gives a higher return. As a result, she decides to shift more of her marketing spend on Google Ads.
Takeaway: Set appropriate cost-per-click targets
Set appropriate cost per click targets on your channels based on estimated lifetime value from each channel. Assuming a sales based lifetime value, the rule of thumb is CAC/LTV = 0.2-0.33. If your conversion rate of 1%, you'll need CPC = LTV/500 to LTV/300.
Who are my most valuable customer groups?
What your customers buy can give you valuable insights as to who they are. For instance, John is a customer who shops for gifts for his female friends. Sophie is a shoe enthusiast who only buys shoes and related accessories from your store. There could be a large number of Johns and Sophies in your dataset. Customer personas help you identify who they are.
Metisa Customer Personas help you identify which groups of customers are most valuable to you. Let's look at Rachel's personas. In the diagram below, you can see that her customers naturally fit into the 6 personas.
The persona with the most valuable customers is Cluster 5. Each customer is worth $1,553 and there are 82 of them. It contains customers who buy lipsticks and creams. Finding more customers of this persona would help increase the average customer lifetime value in the business.
Takeaway: Focus your marketing efforts on high-value and potential customer personas
One of the things Rachel can do is focus on finding more customers from the high-value personas. She can do that by targeting customers with similar interests and demographics as them in her advertising campaigns.
She should also think about how she can get personas with a large number of customers but lower customer lifetime value to spend more. For instance, this will be Cluster 3 and 4. She can do this by sending tailored emails to customers in each of these personas with a specific offer to get them to buy more.
The process of setting up these emails is often specific to your customer base and is determined on a trial-and-error basis. We can also use A/B testing to test out the effectiveness of certain email marketing campaigns or rules.
Which customers are at risk of leaving my platform?
Another key concept in customer retention strategy is figuring out which customers are at risk of leaving your platform. Knowing this information is valuable because you are able to take preventive steps to win them back them.
The Metisa Retention Insight breaks your customers into their lifecycles. We use a statistical model to determine if each customer is at-risk. This model is 25% more accurate than how businesses usually look at churn, which is to consider a customer who hasn't bought for a period of time (say 12 months) as at-risk.
Takeaway: Focus on winning-back at risk and lost customers
You can setup email autoresponders that go out when a customer becomes at-risk or lost. These are usually the best moments to win-back a disengaged customer. Experiment with different offers to win your customers back. Keep in mind that the best offer is not necessarily the one with the deepest discount. You have to test it.
Want to get predictive analytics on your data?
If you run a Shopify, Magento, BigCommerce store, you can get instant predictive insights about your business with Metisa. Just install our extension and you're ready to roll. No code required and free to start. Visit our website and sign up for a free account. We can do the same with other and even custom built stores, just reach out to firstname.lastname@example.org.