We've talked before about thinking in unit economics, and one of the most common forms of this—which, for reasons we'll get into, is getting even more common over time—is customer lifetime value. The exact mix of calculations varies, but the basic idea is to treat the customer relationship as a stream of cash flows, measure the present value of those cash flows, and then set your marketing budget accordingly.

There are many, many ways to do this wrong. You can:

• Discount at too low a rate.

• Discount the wrong thing—looking at revenue for a product that has a significant marginal cost, for example.

• Underestimate churn, especially if you're spinning up a new marketing channel on the assumption that it will perform as well as your older channels. (The best-performing cohort of Diff subscribers is the very first few people who signed up. A customer cohort performs very well if a material portion of it consists of the founder's parents.)

• Underestimating future growth in revenue per customer—there's an invisible graveyard of companies that died because they didn't spend enough to go after a big market. This is just as big an error as overestimating performance; the difference is that it means wasting founders' and early employees' time, not investors' money.

The fun questions to debate about here are things like the appropriate discount rate. The discount rate is synonymous with the return on investment that you expect from your marketing dollars, but there are a few considerations to keep in mind. One is that since marketing is a variable cost, but there's also a cost to building something worth marketing, an accurate discount rate sets a ceiling on the company's overall economic return on investment. Some fraction of their spending directly drives revenue, but it's at most 100%, so if the company spends until it's hitting a 15% return on CAC, overall return on investment has to be lower than that.

But the actual hard part is figuring out how to compare different kinds of customers, and predicting how churn will evolve. A classic boom-bust story goes like this:

1. New company establishes some model (group buying, ride sharing, meal kits, food delivery). Their cohort numbers are amazing, so they raise a lot.

2. Someone else raises money by either targeting a niche within that broader space, or just by being the biggest business with exposure to it and room in the cap table.

3. That flood of money simultaneously raises customer acquisition cost and, by increasing churn rates and putting pricing pressure on incumbents, lowers customer lifetime value.

So these stories often involve a company raising $50m to invest with incredibly desirable unit economics, and then raising $500m to acquire new customers who will never produce as much cash as it took to acquire them. And that doesn't just happen when new competition shows up. It also happens when there's a new marketing channel: people who find a product through search are different from people who hear about it with social media, or from real-life friends; the people who sign up for a paid product right away have a different profile from the people who need a few iterations of a drip email campaign to make the call. Younger customers have different spending patterns from older ones, and if a business lasts long enough, those younger customers will become older ones. (There's a good reason companies try to habituate young people to using their product even if those young customers are expensive and don't spend much at first. The college student who flew home on Basic Economy may, in ten years, end up being a road warrior working in consulting and sales, and their perception of which airline to spend their employer's money on is affected by how airlines treated them on their own dime.)

In Eugene Wei's fantastic piece on how to make a good chart in Excel, he talks about working at Amazon in the 90s, and seeing that while they were losing money, the dollars they lost were earning them customers who would keep producing more and more profitable spending. Amazon couldn't afford not to spend money as fast as it could raise it, for as long as that was possible.

Customer-level economics are getting more important as customers get easier to track. When retailers mostly transacted in cash or by check, there wasn't a great way to pool data on customers and measure their behavior over time. When those interactions move to credit cards and are mediated by apps, it's much more straightforward for a store to know exactly what a customer's evolution looks like. So, over time, more and more company-level economics will be determined by customer-level numbers. That essentially creates a two-tier system: there are companies selling into an opaque market whose supply/demand dynamics are mostly invisible to them, and there are companies that sell to specific customers, whom they know better with every additional transaction.

Disclosure: Long AMZN.

Read More in The Diff

The Diff spends a lot of time writing about companies in terms of their customer-level economics. Some examples:

• Software valuation multiples are sensitive to their customer-level revenue growth numbers and those numbers are sensitive in turn to how much funding flows into software.

• Airlines can be evaluated in many ways, one of which is based on customer loyalty and the many ways it's monetized.

• Costco has a unique model where their customer acquisition cost is real estate, and they monetize with subscriptions ($).

• But it's not that unique: gyms also monetize the fixed investment in a physical location and exercise equipment through recurring revenue. (But unlike Costco, they have a different view of how much utilization they want: either none at all, or as much as possible.)

Join the discussion!