When you think of the demand curve for your business, you imagine a general downward sloping curve.  Generally, economic theory would dictate that as you decrease the price, the quantity sold might increase.  However, decreasing the price might not be the revenue maximizing action, because things would depend on what economists call the elasticity of the demand.  The elasticity of demand is the percentage change in goods demanded for a percentage change in price.  If the demand is highly elastic, then lowering prices will raise revenue, but if the demand is highly inelastic, then lowering prices will lower revenue.

We see a similar dynamic play out in the Customer's Lifetime Value.  Suppose that you were to give a discount to a particular group of customers.  Would that raise customer lifetime values so that total revenue is higher, or would it lower customer lifetime values so that total revenue is lower?  This is a question that is played out over and over again in the relationship between the customers and the seller.

Consider the case with your "best" customer, in terms of generating the highest amount of revenue over their lifetime for your business.  If you gave your "best" customer a discount, would that generate more revenue for you over their lifetime for your business, or less?  Paradoxically, the answer is that it would most likely generate less revenue.  The reason is because at the current prices, their demand is satiated.  If you gave them a discount, it would not increase consumption enough to cover the discount.

Consider the case of your "worst" customer, in terms of generating the lowest amount of revenue over their lifetime for your business.  These are the people who will most likely churn.  If you gave these people a discount would they generate more revenue over their lifetime to cover the discount and have a gain, or would they generate less?  Unsurprisingly, they would most likely still churn and use your discount, so your business would lose revenue.

However, there is a particular class of customers that you currently have, which would most likely grow their consumption given the right incentives.  They are the ones that need to be introduced into to new products, or features, or even to help their own businesses grow so that they could consume more.  They have the potential to grow into your "best" customers.  They are however a very select group that is hard to identify.

With our AI software that forecasts future lifetime values we can identify this group easily and help you avoid incentivizing the wrong groups.  We are able to see what levels of incentives are necessary to induce changes in behavior and the resulting gains or losses.

Because we are able to forecast down to individual future purchase values for each customer, you can target only those customers that bring you a net gain.  We can even automate this feature for you.   Contact us for a demo.

We previously covered average lifetime value by looking at actual past purchase lifetimes.  There is however an alternative formulation that is widely used that doesn't look necessarily at averaging each individual lifetime value, but instead just uses the averages in customer behavior.  Let us elaborate with each formula.

The Average Order size is given by  AO = Total Revenue / Number of Orders. This treats all orders from everyone the same, as if they came from each individual equally.

The Average Frequency of Orders is given by AF = Number of Customers / Number of Purchases.  This doesn't measure true frequency because you don't know the actual time between purchases.

The Average Lifetime is given by AL = sum of Customer Lifetimes / Number of Customers.  However, because you can't necessarily wait 20 years to see the total customer lifetimes, it is sometimes suggested to use 1 / churn rate percentage to estimate this value.

The Average Customer Lifetime Value is then given by AvgCLV = AO*AF*AL.  It is just the previous formulas multiplied with each other to get the "average" in lifetime value.

The above formula, while simple, has a few glaring flaws that are not fixable.  Customers don't behave like the average, and the "average" lifetime value will be terribly misleading.  When you multiple the averages together, you assume that each factor AO, AF,  and AL are statistically independent.   They are not.  When the customer is a high lifetime value customer, the Average Order sizes are larger, the Average Frequency is greater, and the Average Lifetime is greater, for example.  When the customer is a low lifetime value customer, the Average Order sizes are smaller, the Average Frequency is less, and the Average Lifetime is shorter, for example.

Let's work through an example where we can see the effect of correlation on the "average" lifetime value.  For simplicity, assume the correlation is perfect, which won't be too far from the actual case.  These numbers come from a Starbucks case study.  For Starbucks, the average frequency is AF = 4.2 visits per week.  The average order size is AO = $4.05 per visit.  The estimated lifetime is 20 years, so AL = 52 weeks * 20 years.  This gives the Average Customer Lifetime Value as

AvgCLV = $4.05*4.2*52*20 = $17,690.40

We can look at the correlation effect as equivalent to adding an extra term inside the formula for Average Customer Lifetime Value

AvgCLV = (AO+x)(AF+x)(AL+x)

When a customer is high lifetime value customer, x is added to all the base values.  When a customer is a low lifetime value customer, you can think of subtracting x from all the base values.  Now let us look at what happens to the Starbucks customer when they are a high value lifetime customer by assuming it raises all the values by 20%.

HighAvgCLV = ($4.05*1.2)*(4.2*1.2)*(52*20*1.2) = $ 30,569.01

This value is almost twice the baseline. And let's look similarly at what happens when the Starbucks customer is a low value lifetime customer by lowering all the values by 20%.

LowAvgCLV = ($4.05*0.8)*(4.2*0.8)*(52*20*0.8) = $9,057.48

This value is almost half the baseline value.  High and Low value customers might not be a problem with there were an equal number of both types, but if you recall the L-Shaped distribution from the previous post on "Why The Average Customer Lifetime Value is Not Enough", the majority of customers come from the low value of the distribution with a long tail, since roughly 80% of revenue is generated by only 20% of the customers.

The real-world effect of correlation on the terms in the "average" lifetime value AvgCLV = AO*AF*AL, is to make the calculation highly biased and misleading by making the L-shaped distribution have greater extremes in the tails and making it more L-shaped.   When the true lifetime value of customers affects the profitability of your business, you cannot depend on the "average" formulas. Given that there is a cost to acquiring customers (through ads and incentives), then knowing the true lifetime values is a key piece of information that you need for your business.

We can help you with discovering what the true lifetime value is of each and every customer in your business, because we have AI software that forecasts what each individual customer's lifetime value will be, and what their stream of future purchase values will be, with high accuracy, early in their life.  Contact us for a demo.

A lot of e-commerce advice sites will suggest you look at customer lifetime value by their average.  The average value however will be very misleading and may cause you to make terrible decisions on acquiring customers.

Let's examine why the average lifetime value is highly misleading.  Below is a graph of customer lifetime purchase values ordered by customer.  It is a familiar L-shaped distribution with long tails.

This distribution becomes more L-shaped as more customers are added.  It never becomes the shape of a normal Gaussian distribution that everyone is familiar with, which is a symmetric bell-shaped distribution with 2 sides.

There is a rule of thumb for business that roughly 80% of the revenues are driven by 20% of the customers. (In actually it is closer to 73-76%).  Let's see what kind of implications such an extreme distribution has on the average customer lifetime value.

Let's begin with some round numbers.  Say $80 million of lifetime revenue is generated by 20 customers.  The rest of the 80 customers only generate $20 million of lifetime revenue.   This makes the average lifetime revenue generated to be $4m*20% + $0.25m*80% = $1 million lifetime revenue on average per customer (out of 100 customers and $100m revenue).  From the high revenue generating group the average is $4 million per customer.  From the low revenue generating group the average is $0.25 million per customer.  

Now for the sake of simplicity, assume that the Cost of Acquisition of  each Customer (CAC) is $1 million, or close to it because you are basing decisions on the average lifetime revenue generated by each customer.   Then for 20 customers, you are profitable by $3 million, but for 80 customers you are losing $0.75million, each.

As you scale your business, it is more likely that you will add customers who are unprofitable.   The tail of the L-shaped distribution becomes more extreme, and what you thought was the average lifetime revenue of $1 million 6 months ago, is now only $0.5 million on average.  You will not know this is changing because it takes a while to realize that new customer lifetime values are lower than before.  This could be disastrous if you kept the $1 million per Customer Acquisition Cost.  However, using our AI forecasting software you can change the distribution of lifetime values by going after only the more profitable customers or changing unprofitable customers into profitable ones.

Suppose that you are able to cut 40 out of the 80 unprofitable customers.  Then the profit goes from zero in the previous example to +$30million.  We can do this for you because we can forecast customer lifetime value for each and every customer and forecast individual future purchase values for each and every customer, early in their lifetime.  Contact us for a demo.