Managers and investors typically use “excel-based models” which assume an “average customer”. While these models can be useful for certain managerial decisions, they are not appropriate for making big bets (e.g. on channel spend, which customer segments to target, and the valuation of the company).
Your best customers are 100x more valuable than your worst
Customers are extremely heterogeneous in their behavior: how frequently they purchase within a category, how loyal they are to one brand within a category, how long they remain in the category, the percent of items they return, how much they spend on an average visit, &c.
And of course, all the variation in individual dimensions compounds as they interact with each other, such that the best customers for a company are often (at least) 100x more valuable than the worst. Something like an 80/20 rule exists for every company we’ve studied (80% of revenue coming from 20% of customers).
Armed with the knowledge of which customers are most valuable, investors can accurately value the options available to managers: which segments the company can target going forward, which channels are the most profitable and which the company should exit — all things that an “average customer” can’t tell you.
Statistical CLV models are (much) more accurate
CLV calculations are highly sensitive to assumptions. If your model uses an annual churn rate of 10%, when in reality the churn is 15%, your customer valuation will be 50% too high.
Since most of the inputs to CLV are something that happens “in the future”, most managers use shortcuts such as a “finite horizon” LTV, that only covers a 1-, 3-, or 5-year period. This can massively understate LTV and lead to underinvestment in building the customer-base.
Statistical methods are capable of estimating the unobserved parameters with limited data, and modern methods can do so very accurately. We routinely estimate revenue from existing customers within 5% accuracy over a six month or one year holdout period.
Advanced methods allow investors to ditch coarse heuristics and create more accurate forecasts.
CLV changes over time (and usually not in the right direction)
As companies grow, their customer-base expands outward from the early adopters, who are typically the highest spenders and the most loyal customers the company will ever acquire.
This means that CLV for new customers tends to fall over time. Not only that, but the cost of acquiring new customers (CPA) tends to increase over time for the same reason. Since companies typically build their “average customer” models on those customers with the longest tenure, they build their models based on their most valuable customers and overstate how much equity they’ve built in their customer base.
This delta can be especially pronounced in fast-growing companies, as the majority of their customer base has a short tenure (with 100% growth, 50% of the customer base was acquired within the last year), but their models are built off their earliest cohorts. The delta can mean as much as a 50% or more swing in overall valuation.