S******y
发帖数: 1123 | 1 How to compute Life Time Value for a customer?
There seems to be so many ways to do that. Can
somebody provide some guidelines?
For example, I have 500K customers spending data for
the past 2 years month by month. I would like
compute LTV for my customers. How could I do that? | b*******y
发帖数: 30 | | b*******y
发帖数: 30 | 3 by 'value', you mean revenue, profitability, sale, or some other
characteristics? is this used within finance, risk, or marketing dept? -- it
all depends on the context, i guess | S******y
发帖数: 1123 | 4 Thanks!
It is based on Revenue (i.e. spending from customers)
It is used in marketing.
Based on LTV, we can decide which customers we go after. | s*****n
发帖数: 2174 | 5 LTV is rarely calculated for each individual customer, but usually
calculated for a certain type(group) of customer.
1. First you may want to classify the customers into certain groups. Field
knowledge needs to be used to do this, since there is no general guideline.
Sometime, only one (overall) group is considered, so you can skip this step.
2. Model the value per time span, for example, revenue per month per user.
This is usually as simple as doing an average.
3. Model the median/mean life spa | S******y
发帖数: 1123 | 6 Thanks so much!
That is very helpful in helping me connect the dots. | S******y
发帖数: 1123 | 7 Regarding -
>3. Model the median/mean life span for each group. You can use complex
model such as fancy survival models, or simple approaches such as doing a
arithmetic mean of empirical life spans.
For suvival models, does that mean -
compute probability for each month, then compute weighted average for life
span?
Besides, what about cencored data modeling?
Thanks. | x****u
发帖数: 262 | 8 我觉得你就拿survival的model套一下就可以了，什么censored 都解决了
【在 S******y 的大作中提到】
: Regarding -
: >3. Model the median/mean life span for each group. You can use complex
: model such as fancy survival models, or simple approaches such as doing a
: arithmetic mean of empirical life spans.
: For suvival models, does that mean -
: compute probability for each month, then compute weighted average for life
: span?
: Besides, what about cencored data modeling?
: Thanks.
|
|
