How to use Customer Lifetime Value (LTV) for data-driven transformation

Use LTV to set goals for your business and devolve accountability downwards

If your business is undertaking a data-driven transformation, one of the first things you ought to address is what the purpose of that transformation is. Data-driven transformation should not be about short-term thinking such as squeezing out operational efficiencies. Instead, as suggested in the excellent book Converted by Neil Hoyne, transformation should seek to maximize the full value of every customer relationship. The “full value of each customer” is called the lifetime value (LTV). Data-driven business transformation requires you to orient every aspect of your company — marketing, sales, product, support — around maximizing LTV.

Photo by Alvaro Reyes on Unsplash

Of course, you don’t want to run your business based on how valuable the customer was. Instead, you want to run your business based on how valuable a customer will be. The more valuable the customer is, the more likely you should be to build products they’d buy, to buy ad space in the magazines they read, to provide a dedicated sales person to them, and to proactively address their needs. At the same time, you don’t want to have to make every little decision yourself — instead, you want to empower everyone in your organization to make the decisions you would make. The way to do this is to use LTV.

In this article, I will describe how to use the LTV to drive data-driven transformation across your organization and improve customer outcomes.

LTV provides the levers of your business strategy

Ignore the tools vendors — there is no standard LTV formula. Determining the factors that go into calculating the LTV is something that you, as a business leader, will have to weigh in on — it is one of the key ways that you will set the direction for your organization. The reason is that the LTV calculation is also about business strategy — it won’t be the same for all businesses, and it won’t be the same for your business over time. You cannot simply take the LTV formula that is built into some business intelligence tool and tell your team to use it. Instead, you should treat the LTV as a model that you continually discuss with your leadership team — it is something that you need to periodically revisit.

How to compute LTV is one of the most important strategic decisions that you will make. This article goes into quite a bit of granular detail because it is important that you understand the nuances. These nuances are the levers that you can use to drive your business forward.

The LTV calculation approach that you choose will greatly inform the day-to-day decisions that your leadership team will make. It will guide the investments that marketing makes, the incentives that sales provides, the design of your product, the pricing of your services, and the product mix that you offer. This is not limited to your leadership team — LTV is a way to drive decision making and accountability throughout the organization. You want even the user-interface designer thinking about which buttons to highlight to do so with the goal of maximining LTV. Your business will make hundreds of small decisions that together will drive the LTV upwards. Therefore, it is paramount that everyone in the organization understands not only what goes into the LTV formula but why.

In this article, I will start by dissecting a few widely used formulae. I hope that this will help you employ LTV as a discussion aid so that you come up with an approach that works for your business. Make sure to revisit the LTV approach every planning cycle so that you can adapt it as your business changes. You will make mistakes early on — it’s important to correct them as you go forward and to set the expectation throughout the organization that the factors that go into LTV is something that will be adapted over time.

Retail LTV

Let’s start with the formula for customer LTV thatis proposed in the excellent book Converted by Neil Hoyne. He suggests that you should look at your historical record of purchases and calculate two values:

• Number of transactions per customer
• Average value of each transaction

Then, you need to build a machine learning model that, given everything you know about a customer, will predict:

• The probability that the customer will buy from you

Once you have these three values, multiply them to get the LTV:

This formula works really well for digital marketing — consider the situation where you are a retailer that sells a wide variety of items and want to decide how much to spend on advertising to a customer. You can estimate the number of future transactions and average value of a transaction from your historical database of customer purchases. The probability that this is a customer worth going after will be based on what fraction of people like this person clicked on the ad and became a customer.

There are three wrinkles with this formula that every data scientist should be aware of.

1. The first wrinkle is what “value” means. Is it profit? Is it revenue? Is it gross margin? This depends on your business. Stable businesses usually choose profit. Businesses in fast-growing sectors prioritize market share, and therefore choose revenue. Companies that are readying themselves to be publicly listed often choose gross margin. You will have to set the value in a way that discourages short-term efficiency improvements and pricing changes that increase profitability to the detriment of long-term customer retention. You might, for example, choose a weighted sum of revenue and gross margin as your measure of value where the weight changes over time as you get closer to a planned IPO.
2. Second, your business will change over time precisely due to the actions that your organization makes. The number of transactions per customer might fall if your advertising funnel becomes larger. If you advertise only high-value products, the average value might increase. Therefore, you need to predict the future # of transactions or value based on each actions you undertake. In such a case, ask your data science team to build ML models to predict the first two variables also. Alternatively, you might choose to set these as annual goals for your teams to hit. For example, the product team can be measured on how well they increase the average number of transactions while the the marketing team can be measured on the effectiveness of ads targeting to increase the probability of purchase.
3. Third, your historical record will underestimate newer customers because they haven’t been around long enough to build out a full record. An easy way to think about this is to imagine that you want to predict the number of books read in high school. The model will be good for 12th graders, and terrible for 6th graders. So you’ll have to calibrate newer customers based on the behavior of your oldest customers. In the books analogy, you’d find the 12th graders most similar to each 6th grader and calibrate — a 6th grader who read 13 books will end up reading 124 books, for example, where the 124 is the average lifetime book reading of 12th graders who read 13 books in 6th grade. This is often called a cohort analysis. If you have enough data, you can use just the history of the 12th graders in several time-windows and train the model. So, train the model to predict the actual reading habit by using 12th graders’ history from 6, 5, 4, 3, 2, 1 years ago.

The way you approach these three issues will fundamentally change how you calculate and use LTV.

Fulfilled-By-Amazon LTV

Another LTV formula that you will see referenced quite a bit is the LTV calculation used by merchants that use Fulfilled By Amazon (FBA). If you fulfill through a 3rd party such as Amazon, you may not even have customer information. Also, if you have not been in business for a long time, and your margins are quite low (or negative, as is true for many startups), then you don’t have the historical information that you need to use Hoyne’s method. You have to take a simpler approach.

If you can somehow estimate the number of times a customer will come back and buy, you can get the lifetime value of a customer using:

In other words, you take the price of an item, subtract the costs of the sale (shipping, Amazon fees, etc.) and multiply by the total number of times you expect the customer to buy from you. Subtract the customer acquisition cost (CAC) such as any ads or search placement fees you pay either Google or Amazon to get the LTV.

This immediately highlights the problem for any retailer who’s relying on Fulfilled by Amazon — with no brand-loyalty, the N_repeats is 1 and the CAC will eat up your profit margin. So you have to invest in brand loyalty to get customers visiting you directly — the LTV is a great way to focus the attention of your organization on the existential threats to your business.

Subscription Product LTV

Suppose you have a a fixed-fee service. Perhaps you are selling a magazine for which subscribers pay $30/year or a life insurance policy with an annual premium. For such products, you can compute the LTV based on the expected number of years for which the customer will renew their subscription:

This formula uses a few concepts. Let’s walk through them one-by-one:

• The S_year is the survival rate by year. For example, S_3 would be the fraction of customers that maintain their subscription into year 3. Estimate it from the historical data by cohort (or customer segment). For example, customers in rural areas might have a different renewal rate than those in urban areas.
• The V_subscription is the value of a subscription. This could be the profit, computed by substracting the cost of providing the service and the customer acquisition cost.
• r is the discount rate. Because subscriptions last into the future, you have to compute the net present value of a customer using compound interest. This is not purely financial — the discount rate also captures the network effects of a long-term customer. If you are not sure of what this could be, start with a discount rate of about 10%, but be aware that incentivizing word-of-mouth referrals will improve the r, and hence the LTV.
• N is number of years you can reasonably estimate these values. Typically, you can compute it based on the historical record and can extrapolate the curve. So, N tends to be the year at which S_year drops below 1%.

As with retail LTV, use the increase of survival rate, value of subscription, and discount rate as a goal-setting exercise for various teams.

Cross-sell Product LTV

The subscription product calculation assumes that you have only one product to sell. However, suppose you have several magazines. It is quite likely that you will be able to cross-sell a new magazine to current subscribers. Your LTV calculation has to take this into account by considering the likelihood of such cross-sell.

The way you do this is to compute the LTV of the ith product and then add in the likelihood of being able to cross-sell the jth product to subscribers of the ith one:

Then, add in the likelihood of being able to cross-sell the jth product to subscribers of the ith one:

If you have the data to do so, you could do this by year (i.e. estimate the probability of being able to cross-sell in year 1, year 2, etc.)

Multi-tier Product LTV

The cross-sell calculation assumes that the products are independent and that customers will buy multiple products. However, many subscription services (think of tax software) have several tiers of services. Perhaps there is a basic tier for which people pay $10/year, a medium tier priced at $20/year, and a premium tier that costs $30/year.

With multi-tier products, we can assume that customers will be exclusively using only one service. Someone who bought a premium-tier tax preparation software will not also buy the basic one that year.

Each of the tiers can be considered a separate subscription product and you can simply add up the LTV of the three products. But in reality, people will not just churn out. They may move to the higher-level service as their income increases, or as a result of cross-sell. They may move to the lower-level service because of price pressure.

To capture this, we need to incorporate the probability that a customer moves to a different tier of service. This, too, can be estimated from historical data, affected by marketing campaigns and product changes, and predicted through machine learning. Regardless of how we come up with the Probability, you need the probability that the customer moves from another tier to this tier, and we need to add that to the survival rate:

Ad-Supported Product LTV

This is all great if your product is one that users pay for. What if your product is free, and you make money through in-app ads? In such cases, you may not know anything about individual customers. Instead, compute the LTV through a cohort consisting of people who installed the app on a specific day:

You can get the value of that cohort as the ads revenue from ads displayed to that cohort (plus any in-app purchases you support).

How to Implement LTV in your business

In this article, we have discussed why LTV is important and the nuances that go into calculating LTV. How do you use LTV to set direction for your business? Getting LTV to be taken up by your organization involves several teams working together.

Your data science team needs to come up with a formula that mixes many of the elements discussed in this article. They can use this template of questions to ask the business and determine how to calculate and report LTV:

• How many separate SKUs do we sell?
• Do customers buy multiple SKUs?
• Will customers buy the same product multiple times?
• Do customers subscribe to a service and pay monthly or annually?
• What’s the best measure of value?
• How do we know when a customer has churned and won’t come back?
• Do we upsell customers?
• Do we cross-sell to customers?
• Do we own our fulfilment channel?
• Do we know who our customer is?
• What are some cohorts within which LTV is likely to be the same?
• How do we set sales incentives?
• How do we evaluate the efficacy of marketing offers?
• How do we set Product OKRs?

Your sales, product, marketing, customer support, and other teams should take on OKRs that will positively impact the parts of the LTV formula over which they have control.

• The product team can be measured on how well they increase the average number of transactions (retail LTV), or the number of years that a typical customer stays (subscription LTV), or the likelihood that a customer upgrades (multitier LTV).
• The ops team can be measured on how well they improve the operational efficiency and thereby increase the value per transaction. They could do this by reducing shipping times, reducing manufacturing costs, etc.
• The marketing team can be measured on the effectiveness of ads targeting to increase the probability of purchase and the likelihood of cross-sell and up-sell.
• The support team can be measured on the effectiveness of customer service to reduce customer churn

Your COO should take on the task of measuring and managing to the LTV. This will involve analytics teams across the company to take on measurement tasks and send that information to an enterprise data warehouse that powers a reporting dashboard.

At a company all-hands, you (the CEO) should explain their vision to the company. Go through the goal you have for data-driven transformation, and how you will use the LTV to devolve responsibility downward into the organization. Make sure to explain that the method of calculating LTV will change over time, but that the purpose of the LTV is to ensure that everyone in the business keeps the customer front-and-center.

The LTV is how many of the most successful consumer businesses such as Amazon and Google manage their customer-facing businesses, and I hope this article motivates you to join their ranks.

Sources/Additional Reading

• Converted: The Data-Driven Way to Win Customers’ Hearts. https://www.amazon.com/Converted-Data-Driven-Way-Customers-Hearts/dp/0593420659
• Increasing LTV on Amazon. https://www.sellerapp.com/blog/best-ways-to-increasing-customer-lifetime-value-on-amazon/
• Survival Analysis in Telecom. https://support.sas.com/resources/papers/proceedings/proceedings/sugi28/120-28.pdf
• Ads LTV. https://support.google.com/analytics/answer/6182550