Growth rates

The most basic way to look at revenue data is to calculate growth rates. Usually investors look at historical annual growth rates and monthly or quarterly growth rates in LTM (Last 12 Months).

For annual growth rates investors look at the longest historical period possible. This may be 2-3 years for a recently established company or as long as 10-15 years for a mature business. Longer period shows business resilience through the macroeconomic cycles.

Monthly or quarterly growth rates are calculated for at least the LTM period. The longer period is available for analysis the better. The monthly data shows seasonality as well as current growth of the business.

Let’s look at an example. You consider investing in an online retailer BrandCo with $13 million in sales in LTM. The sales of this company increased by 20% in LTM compared to PTM (Previous 12 Months). As you can see on Exhibit 1, in the last 6 years BrandCo’s revenue was growing quite consistently at 13% CAGR (Compounded Average Growth Rate).

Exhibit 1: Annual growth rates

revenue chart
Source: Quantacap analysis.

According to Exhibit 2, in the LTM the sales were almost two times higher in each month from January to April 2023 compared to the same months of 2022, but in August and October 2023 the revenue declined.

Exhibit 2: Monthly sales

Source: Quantacap analysis.

However, total revenue figures do not provide insights into the quality of the growth and what is driving the topline. Therefore, we should look further into the product vintages to understand the sources of growth.

Product vintages

One of the key revenue drivers is the launch of new products. The product life cycle depends on the product, category, competition and can last from a few months up to a decade. If the product life cycle is short, the company has to launch a lot of new products to sustain its revenue. If the product life cycle is long, then sales can be generated by mature products.

Launching new products is a high risk, therefore investing in a business driven by regular product launches requires a deep understanding of the category and strong execution capabilities in new product development.

Let’s return to our example with BrandCo. The revenue split by product vintage is presented on Exhibit 3. Around 67% of its sales is generated by products launched in 2019 and the rest comes from products launched in 2020-2021.

For an eCommerce retailer this is a good result because products keep generating revenue for over 5 years with no visible decline or cannibalisation among vintages. Nevertheless, we should continue our analysis to uncover customer behaviour and revenue drivers.  

Exhibit 3: Revenue by product vintage

Source: Quantacap analysis.

Customer cohorts

The revenue is driven by individual customer purchases. Each customer discovers a product, makes the first order and later may return to buy the product again.

The usual practice is to split the customer base into cohorts by month of first purchase. For each cohort investors calculate the evolution of the number of unique customers, average order value and order frequency over time.

You can see an example of cohort analysis in Exhibit 4. In August 2022, BrandCo acquired 764 new customers in the Shopify channel. This cohort made 1 order with a value of $76 on average. In the following month only 27 customers from this cohort made purchases again spending $60 per order. In addition, in September 2022, BrandCo acquired 742 new customers, who made their first purchases.

We can conclude that only 4% of the BrandCo customers make a repeat purchase in the months after their first purchase. It means that there is almost no sticky customer base, instead BrandCo has to re-acquire new customers constantly to make new sales.

Now the question becomes what actually drives the revenue of the business and how to manage it post acquisition?

Exhibit 4: Customer cohorts

Source: Quantacap analysis.

Machine Learning

Advanced statistical analysis can help to quickly uncover revenue drivers and use that knowledge to explain historical revenue dynamics or predict future sales.

There are multiple ways to model revenue dynamics starting from basic linear regression to decision trees and neural networks. We refer to these methods as Machine Learning (ML). Each method has its own balance of accuracy, training speed and interpretability.

The factors explaining the product sales quantity include product price, discount, advertising spend, product age (since launch), dummy variables for months and weekdays. External factors, such as keyword trends and prices of competing products, should be considered as well if data is available. So, the first step in applying ML is to prepare structured and clean input data.

The next step is to train models and assess the quality of their fit based on historical data. There are no universal models suitable for all products. The best model is selected based on its fit to actual data, which usually means the model with higher R-squared, lower Mean Squared Error (MSE) or Mean Absolute Percentage Error (MAPE).

Then the selected best model can be used for analysis and forecasting.

The analysis means understanding the impact of each factor on product sales. This can be estimated by slightly changing the factor up and down to compare how much the product sales will change in return according to the model. For example, try changing the price by 10% up and down to see how much the sales quantity will respond.  

To predict product sales, analysts need to forecast the explaining factors first. The baseline for this can be the most recent values, for example, current product prices and advertising spend. Then the forecasted inputs are fed into the model to receive projections of product sales.

In Exhibit 5 you can see the result of analysing one product of BrandCo with ML. Sales of this product are mainly driven by the price. In June 2023, BrandCo tried raising product price from $39 to $45, but this resulted in monthly quantities sold dropping from 4,000 to 600. This relationship was captured by the ML model. Because the current price for the product is high the model predicts that sales of the product will remain low at $30,000-32,000 per month compared to around $140,000 at lower price levels historically. For investors this indicates an opportunity to increase sales of the product by experimenting with lower pricing after the acquisition.

Exhibit 5: ML for product sales analysis

Source: Quantacap analysis.

Conclusion

Detailed revenue analysis helps to understand what drives the business. It is crucial to learn how to manage the business after the acquisition and make good investment decisions.

Almost all investors look at growth rates. But more advanced methods, including product vintages and customer cohorts, Machine Learning for sales drivers analysis and forecasting, can help sophisticated investors to gain an edge in evaluating opportunities and creating real value-add as operators.