A bit suffocating, but important intro

The fact is that most companies are concerned with improving their financial performance. Some believe that the key to success is customer focus. Every day, employees of the headquarters and regional offices evaluate the economics of the decisions made and figure out how to increase customer loyalty.

In fact, we are constantly faced with many different questions about how to achieve growth in the company’s profitability. Decisions made can be of the management level:

• How much to invest in a business direction and when will the investment pay off?

• How much are we making on product A and is it cannibalizing revenue from product B?

• How much has the capitalization of our business changed over the past year?

In addition, we make (mainly in automated mode) many operational decisions:

• Information about which product to send to the client in an SMS?

• With what priority to serve customers under conditions of limited resources?

• What is the optimal price to offer the client for the services consumed?

To answer these questions, companies come up with many different metrics and measure their changes from a potential solution. Inevitably, there are many metrics, some of which may be inversely correlated, and it is often not obvious which action is optimal.

Try to answer the question: is it worth increasing the cost of a service by 10% if it will lead to the fact that after 3 months 6% of customers will stop using it?

We know how to facilitate the decision-making process – introduce another metric in the company! Nowadays it is fashionable to call it CLTV. In this and the following articles, we will tell you why and how we implemented CLTV in the company.

Disclaimer: We really tried to make the articles as compact as possible, but the experience of implementing CLTV in two companies made it clear that this product is not only a cool ML development, but also a culture that should be defined and accepted by all participants. The articles will contain both the technical developments that we have implemented and the principles that we consider important for achieving success.

What is CLTV? It is also LTV, CLV

CLTV (customer lifetime value) – a metric used to estimate the profit that the company can receive from its client during his use of the company’s products and services.

Importantly: this value is the prediction of ML models. In some companies, they are limited to calculating the profitability that the client has brought to the current day.

The idea is simple: make decisions based on their impact on long-term profitability, not on arbitrary factors. To do this, you need to take three steps:

1. Learn to predict the income and expenses that we will incur in the future.

2. Learn to answer the questions: how does the predicted return change as a result of our actions.

3. All this should be implemented only at the level of an individual client.

When you go through all three steps, you will get a universal integral metric that simultaneously takes into account many factors. Since it is calculated at the client level, you can use it personally, in addition, you can look at aggregates in the segments you are interested in and monitor its dynamics.

How to apply:

1. Monitor business performance

2. Evaluate the economics of products and customer segments

3. To rank customers in the conditions of restrictions on the provision of a certain level of service

4. Contrast the costs of the business initiative with the potential increase in profitability

In the argument to increase ARPU by 10% and lose 6% of customers, you will have the understanding that in this way you will be in the profit for 5-6 months, but starting from the 7th month, the result of your decision will be unprofitable. It is optimal to increase the cost of the service by 4%.

If you’ve done it right, implementing a CLTV culture will make you more customer-centric.

How to calculate CLTV

If you dare to implement CLTV in the company, then be prepared for the fact that you will be constantly asked for a formula! I will upset you, there is no universal methodology, you need to sharpen the concept for your business. When studying the experience of other companies, we came to the conclusion that CLTV is most often implemented in companies with a subscription model of monetization (online games, streaming services, etc.). Beeline can be attributed to such companies.

In our case, the main factors that should be present in the formula were:

• Survival

• Proceeds

• Costs

It is necessary to take into account the maximum list of financial components – otherwise, when making decisions, you will be guided by a distorted picture of the world. Special attention should be paid to expenses:

• More often than not, it is they who kill the winnings, which we are happy to reflect in presentations.

• There are many indirect costs among the costs (for example, co-workers) — you will have to develop a methodology for allocation to the client, so finance colleagues will become your best friends.

Finally, let’s move on to the formula

Basic calculation methods

CLTV = Выручка / Уникальное число клиентов

CLTV = Среднее время удержания * Средняя стоимость заказа

Too simple, because these formulas can average completely different customers, and also forget about costs. Let’s go further.

CLTV = (Доходы — Затраты ) / Количество клиентов

But I would like to be more precise, because customers have a tendency to flow away, and there are costs

CLTV = (Средний платеж * Количество повторных продаж или платежей * Среднее время удержания) - Затраты

But we would like a bigger and more accurate horizon, as well as inflation.

CLTV = (Маржа / (1 + Ставка дисконтирования - Коэффициент удержания)) - Стоимость привлечения клиента

I would like more detail, which does not take into account the behavior of the client after engagement

The options are more complicated

The calculation of CLTV can be approached as to RFM analysis. By segmenting customers by lifetime, frequency and amount of purchases, we will get a more accurate forecast (link).

But this approach does not take much into account, for example, combinations of products that a subscriber can use at the same time: a subscriber just with a SIM card or a subscriber with a SIM card and Internet at home.

Markov chains

The matrix of subscriber states is drawn:

• Each subsequent state depends on the previous one

• Each future state has the possibility of its occurrence

How can there be a change in the tariff/connecting home Internet and combining all this into one contract/buying a ringtone from Valery Leontyev, etc. Transitions from any segment to the rest are considered.

We distribute income and expenses by state, multiply by the transition probabilities and, voila, we got CLTV. But at each moment of time, our next step depends only on our current situation, that is, after one change of state, we will forget that Casanova broke into your life last month (example of the weather).

Some methods are discussed in more detail with examples and pictures (links to articles: here, here, here and here)

The listed approaches pretty much limit you to the amount of data you can consider. The resulting result describes the segment rather than the customer. And companies usually have a lot of useful information:

• How regularly does the Client pay?

• How is your interaction going – is it an app or does it come to the office?

• If he comes to the office, how often?

• Did he buy anything from your office?

CLTV is in the beeline

There may be an assumption that everything is simple in Beeline – customers pay monthly, it is enough to multiply this amount by the average life time of the customer, and CLTV is ready.

But it is not so.

Beeline has hundreds of client statuses, from the simple option “the client just bought a SIM” to “the client who has a family tariff with several recipients connected, also has Beeline TV + Internet, he pays for parking from his mobile account, reads horoscopes by subscription every day, and yesterday I bought a phone in the salon for my daughter, and he was credited with bonuses.”

All this is rather difficult to describe with a formula.

A change in any variable affects CLTV. ARPU increased -> outflow increased; competitors launched a new tariff -> outflow increased; wanted to attract new users -> advertising costs increased; sold sims in the subway -> customers with low loyalty

Any business has 100,500 metrics and a dashboard for each. Most of them describe past events, the rest are plans for a quarter or a year. Typically, a plan contains averages per customer that are multiplied by the number of customers. CLTV, on the other hand, builds a forecast on a client-by-client basis, and has an even longer horizon of interaction with clients. And secondly, CLTV takes into account all income and costs of interaction with the client when attracting, servicing, and maintaining it. Based on this information, you can decide whether this client is worth your investment or not, and it is also possible to determine which company’s product is causing losses.

Let us recall our requirements:

• Profitability forecast for each client for a long time horizon with monthly detail.

• Forecast of costs that each client will bring (administrative, advertising, employee salaries, logistics, recruitment costs, etc.).

• Accounting for all used goods and services.

• Taking into account all behavioral characteristics: how often and to whom he calls, how often he uses the Internet and sends memes, when he changed the tariff and from one to another.

• Accounting for not only data that comes in once a month, but also daily features.

• Be able to evaluate the impact of marketing campaigns on the change in CLTV.

• Be able to use CLTV as a KPI.

Based on the above wishes, CLTV Beeline believes that:

where:

N – The horizon of forecasting, in Beeline, is equal to 5 years (the number derived by experience, on average, almost all products pay off on this horizon)

t— a discrete prediction interval, the current version is equal to a month

L_t – The probability of the client’s “survival” until the specified period t

R_t – Revenue per client in the specified period t

C_t – Costs for the client in the specified period t

Further articles from our team will tell you how we forecast all this, how we started from excels to machine learning models, how we got to releases and monitoring, analyzing complex products and evaluating their impact on CLTV.

I would like to thank the team for their help in preparing this and the following articles of the series about CLTV:

And here she is

Product Owner:

Salavat Mullabaev @insanmisin

Team Lead:

Vyacheslav Batanov

Data Analyst:

Maxim Burkatskyi

Daria Eskova @DariaES

Tatyana Slesareva @deckerar

Oksana Suvorova @ovsuvorova

Nikita Lopatkov @nicklpv

Katya Boykova @ Kaatun

Ilya Ishchenko @xbbt

Data Scientist:

Anton Melnikov @a_melnikov

Natalia Kultygina @nataliiiiya

Yaroslav Sapronov @ya_sapr

Oleksandr Svetlichnyy @a_svetlichnyy

Vladlen Severnov @vsevernov

Oleg Vavulov

Semyon Kushelov @sml997

Data Engineer:

Sergey Reutov @s_reutov

Roman Poluhodkin @Ropol