This question is just as much of a math one as it is a valuation/business one, but here it goes:
Say I have a heavily right skewed distribution (cumulative order amounts from customer over their 'life' with a business) and I want to assign an "average" value to any given new customer that will be the most representative (likely) of how much they will purchase (in dollars or units).
I have roughly 2 years of purchase data and obviously the newer customers have shorter 'lives' (measured in months) and less cumulative purchase amounts, simply due to the fact they haven't had the chance to buy as much (time-wise). I'm not 100% sure how to deal with this other than subtracting off some data.
I see 3 options here:
- Take the mean of customer lifetime values from the last 2 years (maybe subtract ~6 months?)
- Take the median of customer lifetime values from the last 2 years (maybe subtract ~6 months?)
- Take a weighted average based on how many customers fall into each 'life' bucket (in months)
Note that the mean is around 30% higher than the median. Median survival time of a customer is ~4 months but again, a small % will last 2 years+ due to the very long tail. There's also a slight complication in that average retention (life) has declined since the last year, so I fear using that data may overestimate the true average lifetime order amounts going forward.
My general question is just how to go about assigning a good 'average' customer lifetime value for any given customer. The 3 methods all have pros & cons to me. I have looked into some survival analysis methods but I'm not sure if customer data meets those assumptions and it still leaves out how to allocate dollars/units to customers. Any advice on how to approach this would be greatly appreciated.