This is the fifth article in a series on novel ideas for SaaS metrics, which started with The unprofitable SaaS business model trap, COC: a new metric for cancellations, The mistake of 1/c in LTV, and SSEBITDA: Steady-state profit metric.
Metrics summarize tons of processes, causes, and effects into a single number. It’s a two-edged sword.
It’s powerful because it lets you reason about complex systems, especially how it’s changing. It helps you focus on what’s important at a macro level.
But it’s dangerous when it combines so many disparate and disjoint processes and systems that the number loses precision. Because then you think you understand something that you don’t. That’s how bad decisions are made with confidence.
I argue that for many SaaS businesses, the incorrectness of the LTV metric outweighs the value it supposedly confers.
(LTV is defined here if this is greek to you.)
Using LTV in a business decision — like how much we can spend to acquire a new customer — implies that the lifetime gross revenue from a customer is know-able today. Clearly, though, it isn’t.
Every component of LTV changes over time:
- MRR — changes due to how systematic you are at upgrades, your ability to cross-sell, growing/shrinking the install base inside one customer, per-use charges.
- GPM — efficiency of the service, which for small companies can change by 30% in a year and even large stable companies can move by 1% per year.
- Cancellation Rate — hopefully shrinking as the company improves the product and service to address the causes of cancellation, but over a timeframe of years, this can change dramatically with advent of new competitors, shrinking market, different technology, or mixing different customer demographics as you grow into adjacent markets.
If cancellations are under control, then LTV will necessarily be computed assuming 4-6 years of future revenue. But in the timescale of “years,” we know for a fact that all the components of LTV will change! You don’t know how quickly, or when, or by how much.
For example, Hubspot famously had a low LTV, but increased in 3x in 18 months. (Documented in this great SaaS metrics overview by David Skok.) That’s a big swing in a metric that’s supposed to be able to “see four years into the future.”
Three variables, all changing, unpredictably, which you multiply together and…. you expect the result to mean something?
When you believe a number means something solid, when in fact it doesn’t, you make poor decisions. So for example when you read “An LTV:CAC ratio of 3:1 is healthy,” if your LTV metric can’t be trusted, neither can that formula. Your might believe you’re being efficient in acquiring customers, only to find that your SaaS company isn’t profitable even at scale.
What should you do instead?
I don’t think there’s anything LTV is used for that you can’t use other metrics to do just as intelligently, but without the incorrect assumptions.
For example, LTV is often employed to answer the question: “What’s a reasonable CAC?” The typical answer is “LTV/CAC should be at least 3 for healthy companies, and 5 is very good.”
It turns out you can compute the same ratio using GPM and COC (Cost of Cancellations), neither of which pretend to be able to look out years into the future. Here’s the derivation (with p = CAC/MRR defined in that COC article):
Another way to answer the question about “reasonable” CAC is to think in terms of pay-back period p = CAC/MRR — the number of months it takes to earn back in revenue the cost to acquire the customer. Or better in my opinion, CAC/MRR/GPM so that we’re accounting for the costs to serve those customers.
A good rule of thumb with pay-back period is that 6 months is fine, 3 months is fantastic, and 10+ months is poor unless (1) there’s indirect strategic benefit, e.g. branding, (2) efficiency is improving so we want to stick with it, (3) a mature company can plausibly justify 5+ years of revenue per customer.
Another use of LTV is as a general notion of the dollar value extracted by the company, and thus something that ought to be going up over time. True, but in practice I find you always need to know the values of the individual components to truly know whether the company is healthy.
For example, if LTV is steady, is that OK? If all the components are steady, maybe that’s OK. But what if GPM is improving due to investment in cost-cutting measures while cancellations are increasing, and thus LTV is stable. Is that good? Heck no! Your customers are pissed.
Thus, measuring MRR, cancellations, GPM, and CAC individually are always necessary. Sure you can combine them into a number, but I think that only serves to hide data, hide insights, not help “get a handle on the business.”
What else?
What else do we actually use LTV for, besides desiring it to go up in general? Let’s continue the debate in the comments.
25 responses to “Why I don’t like the LTV metric”
Seems like the old question of how do you value a (startup) company? What is “fair value”? Depending on your analyst, you get different valuations :-)
Yeah, no one can agree on that, and it depends on whether you’re the buyer or seller as well. I don’t think LTV can be used to value companies. Maybe you could use it as a sort of double-check to make sure a valuation isn’t completely out of whack.
so awesome! I always look like I have bugs crawling on me when I have to talk about LTV cause I just don’t understand the magnitude of value people give to this number. I don’t use it for anything other than telling investors about it. I use the payback period for all of my asset acquisition decisions, and calculate it for customers, but I’m going to push it for customer acquisition metrics too. Thanks!
In The Black Swan Nassim Taleb, explains how the farther into the future we try to predict outcomes the more accurate the data being used to make our prediction needs to be. The accuracy of future predictions are incredible difficult when you only have two or three variables and practically impossible when using more than 3 variables. LTV is less accurate because you are now several levels further away from pure data calculations (churn) and must make “educated” guesses such as expected months (N=1/c) and discount rates.
LTV, EBITDA, Market Cap, Credit Score, Birth Weight, Life Expectancy. We are all looking for that single metric to tell us how awesome we are doing.
Customer viewpoint:
I was actually thinking of moving my sites to WP engine.
Your last few posts turned me off entirely.
I don’t want to be part of your COC or LTV calculations. Some factor in your equations. Some nameless cog in your expected churn.
I thought perhaps WP engine was about delivering a good WP experience. Giving a shit.
Nope. It’s about algebra.
No interest in signing up for that. Changed my mind.
I’m truly sorry you feel that way.
Actually, I believe you have it backwards. *Because* we care about churn, we focus on customer service.
For example, we reach out to every single customer that leaves to ask why, so we can see whether we can improve the service somehow. Not to “trick” folks or because they’re cogs, but exactly the opposite — so that we can learn to be better and treat people like people.
We also ask for feedback at the end of every support ticket, which goes straight to managers.
What you should be afraid of is: (1) companies who don’t care if customers churn out, (2) companies who are more concerned with top-line growth than in making sure customers who do sign up are happy.
Great response to a “hard to hear” comment, Jason. Very well said :)
I actually moved my site TO WP engine because I like your blog and know that you’re numbers and customer focused. I help SaaS companies grow for a living, and helping them become data driven in their decision making is a key part of that. The ones that do often become customer focused and try to get the deep insights to understand why customers exhibit behaviors so that important metrics can be moved in the desired direction.
The same happened with me. I like your blog and your presentation at MicroConf (vimeo.com/74338272) really changed my life (sorry for the cliché, but it’s true). I thought that the minimum I could do to pay you back was becoming a client. And I still own you a lot!
Thanks, this was a great read. I’ve really enjoyed this whole series.
May I ask, if you don’t use LTV (net or gross profit based) for estimating the customer profitability for different plans/segments, what do you use for that?
I thought people were using LTV as a profitability metric, to find out the most/least profitable customer segments. The full view is explained in this pic from my latest blog post: http://www.happybootstrapper.com/wp-content/uploads/2013/10/gross-profit-based-cltv.png
This chart is a great example of my point, because it’s wrong.
Look at the “-downward migration.” It removes some arbitrary amount from… what? Gross revenue? But the cost to serve might change too right? And it depends whether that downward movement happened early or late in the contract. It might happen early if sales people are over-zealous, or late if your retention-professionals are good at saving. One of those is a healthy thing, the other is unhealthy.
I like asking what’s happening with all this during this month. Not 5 years from now, but now. I think cohort-based analysis is good for evaluating things over time.
All that said, sure, you could use something like this to compute a CLTV and see what happens with it. But look at all the variables in your chart. Whatever happens to CLTV, you don’t know why, or whether that’s good or bad, or what to do next. All you know for sure is that over time, all those variables will change anyway.
So what’s the point?
Thanks for answering. I totally agree with what you said, CLTV does hide factors.
But I was trying to ask about historical CLTV, not about future-looking. I just wasn’t expressing myself clearly, sorry.
Should I take what you said to mean that you don’t calculate historical CLTV either?
Or do you assign relevant costs (e.g. customer support usage) to right segments in your cohorts and just skip calculating the total/average profit for them?
Sure you could be backward-looking, but of course if things are trending differently now — and hopefully they are! — that’s more important.
For example, if now your GPM is better, then although it’s too bad it was worse last year, that fact doesn’t effect anything that happens this year or the next.
One reason why many startups fail is not seeing that all variables are dependent on each other. For example, a startup has $100 user LTV, and 50% y/y growth. So using rules of thumb they compute optimal cost per acquisition at $20-$30. They also think “we are improving product and adding upsells, so LTV will grow” and “we will improve our funnel, so CPA will go down”, so with these assumptions they happily pay $50+ for user acquisition. But both statements are true in isolation – ONLY if the customer base stays the same, which is by definition impossible because of growth. Maintaining 50% growth rate will require casting a rapidly expanding net instead of skimming the cream, which in turn pushes LTVs down and CPAs up (diminishing returns). At the same time, potential new entrants do the same faulty math as the incumbent and jump in this “super-profitable and quickly growing market”, making everything turn to sh!t even faster.
I couldn’t agree more; thanks very much for writing this. People don’t realize that CPAs get worse with scale; they only assume they’ll improve with funnel optimization.
It is true that, for beloved products, it’s not unusual for word-of-mouth marketing to increase in proportion to the customer base, although not enough to offset cancellations. If you include these in your blended CPA, then CPAs can be more steady.
There’s a debate around whether you SHOULD blend everything into your CPA. Some people say you should only measure the marketing that’s “paid.” Others say WoM counts because it takes work/cost to generate that kind of fanaticism, and if that partially “pays for” some acquisition, what’s the harm in that? And in fact, isn’t that exactly what you should do to stave off competitors by affording higher CPAs (e.g. with AdWords where it’s a zero-sum game)?
I think intelligent people can reasonably disagree on that point. I tend towards the latter.