Most of the targets I’ve worked to in my career have been some version of cost per acquisition. Spend this much, get a customer, keep the ratio under a line. It’s simple and everyone in the room understands it, which is why it survives. In luxury I’ve come to think it’s often answering the wrong question.
The problem is what “acquisition” means. A CPA target, or a ROAS target for that matter, usually looks at one order: the first one. It asks whether the money spent to win that sale was covered by that sale. If the answer is yes, the channel is working; if not, cut it.
That works reasonably well for a business where customers buy often and the first order looks a lot like the tenth. It works badly when the first order is only a small part of what that customer will be worth to you.
What a customer is worth over time
Think about how someone actually behaves with a business like Watchfinder. They might sell us a watch they’ve owned for years, and come back a few months later to buy something else. They might buy one watch, and then, three years on, part-exchange it for another. They might never buy at all, but bring the business a watch to sell twice in a decade. And satisfied customers in this category talk to each other; watch people are a fairly chatty bunch.
None of that shows up in a cost per acquisition calculated on a single transaction. The channel that brought that person in is judged on one order, and everything the customer is worth after that is invisible to the target.
So the channels that are good at finding new people tend to look expensive, and the channels that are good at recapturing people you already had tend to look cheap. Optimise to that picture for long enough and you shift money away from the activity that brings new customers in, towards the activity that harvests the ones you’ve got. On paper that looks efficient. My worry is what it does to the size of the customer base two or three years later.
It sounds similar to the television and digital problem I wrote about earlier this month, but that one was about which channel gets credit for a sale. The issue here is time: how long you wait before deciding what a customer was worth.
What I’d rather measure
The obvious answer is lifetime value, and in principle everybody agrees. In practice it’s hard, for a few honest reasons:
- The time frames are long. If a typical customer’s second transaction comes two or three years after the first, you need years of clean data before you can say much about lifetime value with any confidence.
- The data lives in different places. Buying and selling are often handled by different systems, and a single person can easily appear as two or three records.
- Nobody wants to wait. Budgets are set monthly. Telling someone that a channel’s true worth will be visible in 2019 is not a satisfying answer in a September budget meeting.
None of that is a reason to give up on it. What I’m trying to do instead is a middle way:
- Separate new from returning customers in the reporting. Even this basic split changes the picture a lot. A channel delivering a high share of genuinely new customers deserves a looser target than one mostly reaching people who already know you.
- Set targets by channel role. Prospecting and remarketing shouldn’t be held to the same ratio. One is meant to find people, the other to bring them back.
- Build a rough value estimate per customer type and use it, even if it’s wrong. A sensible estimate based on the history you do have is better than the implied assumption that every customer is worth one order, which is certainly wrong.
- Look at cohorts. Take the customers won in a given month and track what they’re worth six, twelve and eighteen months later, by the channel that first brought them. It’s slow, but it’s the closest thing to a fair test.
The awkward conversation
The hardest part of this is the conversation it creates. A lifetime value view nearly always says you should be willing to pay more to acquire customers than a single-order view allows. That sounds, to a finance director, a lot like marketing asking for permission to spend more money on worse-looking numbers.
The only way through that I’ve found is to bring the evidence: the cohort data, the split between new and returning, the share of business coming from repeat customers. It doesn’t make the conversation easy, and I don’t always win it, but it moves the argument on from one month’s ratio.
I still use the ROAS calculator for the single-order view, because it’s useful to know. I just don’t let it be the only number that decides where the money goes.