We use ecommerce metrics for two purposes: to monitor and to improve performance. Similarly, there are two methods to calculate ecommerce metrics: top-down and bottom-up.
With top-down, we take company-level data like total revenue and number of orders and arithmetically combine them to calculate metrics like average order value. Top-down is okay for monitoring our store’s performance but limited for improving it.
With bottom-up, we instead find accurate, individual measurements — like the revenue for each individual order — and use those measurements to calculate metrics like average order value, create charts, and discover the insights we need to improve our store’s performance.
While top-down is limited to metrics that can be explicitly derived arithmetically (like average order value), bottom-up is not. As shown above, with bottom-up, we can find the median, modal, standard deviation — and, in fact, any possible metric, chart, or statistical method that might help us improve our store’s performance.
Use our AI Powered chatbot to answer questions using your stores data
Top-down leads to mistakes, especially when we try to interact with our metrics to find insights.
Say we want to investigate new and returning users over time. To do so, we create a dashboard where we can change the time frame, like this:
The most common way I see of calculating new and returning customers is to assume that the number of returning customers must be the total number of customers minus the new customers.
returning customers = total customers - new customers
That’s a mistake because a customer can be both a new and a returning customer in the same time frame.
If you ask “how many new customers did we have this year?” you want to know how many people made their first purchase from your store this year. On the other hand, if you ask “how many returning customers did we have this year?” you want to know how many people made their second, third, fourth, etc., purchase this year. So, if a customer made both their first and second purchase this year, they would be both a new and a returning customer.1
If you’re undercounting your returning customers, your customer lifetime value will be wrong, and your business valuation will be wrong. The small errors from top-down can cause big business problems.
All four of these graphs have identical descriptive statistics (mean, standard deviation, linear regression, etc.).
But each is clearly charting a different phenomenon. Limiting ourselves to top-down metrics causes the same problem.
To truly see “what is going on” with our stores, we need to visualize our data. To do that, we need full bottom-up data.
For example, to calculate our customer acquisition cost using top-down, we might divide our company’s total marketing spend by the number of customers. While that does give us a useful customer acquisition cost, we can’t see the distribution of the costs.
With a bottom-up approach, we’d try to assign an acquisition cost to each customer individually (harder but doable with accurate campaign tracking). Then, we can visualize the distribution of our customer acquisition cost in a chart, like this:
The chart reveals what the average hides. In this case, our acquisition cost is bimodal: some customers are incredibly expensive, while others are very cheap. By investigating why, we can improve our acquisition strategy.
Finding accurate, individual measurements forces us to think carefully about what makes our store unique. For instance, to calculate the lifetime value of a customer from the top-down, we might simply divide our total revenue by the number of customers.
customer lifetime value = total revenue / number of customers
However, if we sell subscription products in our store, the lifetime value of a customer might be much greater than the total amount they’ve spent to date. In fact, there’s no definitive way to measure the lifetime value of a subscription customer because we don’t know when they’re going to cancel.
But thinking carefully about how to estimate a subscription customer’s lifetime value gives us a better understanding of our store. In doing so, we might realize that some customers keep their subscriptions for two months, while most others keep them for two years.
(yet another bimodal distribution)
Oftentimes, the most valuable insights come from combining concepts. For instance, to find out which of our marketing strategies are most profitable, we could compare our customer’s acquisition cost to their lifetime value.
With this chart, we can see what high-level, top-down metrics hide. Our most expensive customers to acquire have a much, much higher lifetime value. Instead of trying to lower our acquisition costs with cheaper marketing campaigns, we should double down on the expensive campaigns that have a higher return on investment.
Most of the bottom-up data we’ll need from Shopify can be derived from our order data. You can export your orders directly from the Shopify orders screen like so:
You can also use the Shopify API to pull your orders. Doing so gives you more information than the CSV export but requires development work.
Both of these methods will only give you raw data though. To make sense of it, you’ll have to clean the data, interpreting it in ways that make sense for your store (like finding lifetime value of subscription customers).
But it’s worth it. Taking the time to develop bottom-up data forces us to think carefully about how our store makes money. Ultimately, using that data to create charts and metrics uniquely relevant to our store shows us how to make more of it.
The number of new and returning users in a given time frame can be (and often is) greater than the total number of customers in the time frame, something you’d miss with top-down . [return]