The metric you choose decides which strategies survive.
A finished controlled experiment makes a single point more persuasively than any claim we could make about ourselves. Category and figures are altered to protect the client; the design, the discipline, and the direction of every result are exactly as run.
From the full book to the trial pool
Counts approximate · widths not to scale
The setup
A DTC apparel brand was scaling acquisition hard. Nearly four in five of its customers had bought exactly once, and roughly three in ten units came back as returns. Dashboards looked healthy. Contribution was flat. Their reporting answered 'how many did we acquire?' but never 'were they worth acquiring?'
Score, then select
We scored roughly 120,000 customers by 12-month forward value, net of returns and discounts, and in apparel the netting is the whole story. The top decile of one-time buyers justified real investment before any email was sent; the model separated them from the majority who would not repay a dollar of spend.
The experiment
An eight-week randomized controlled trial on the selected pool, three arms: a control, a modest targeted offer, and the same email with no offer. Read dates and stopping rules were registered before launch, and every read carried customer-level confidence intervals.
The verdicts
The targeted offer built measurable forward value and rolled out inside the client's existing Klaviyo account with zero new build. The no-offer email was statistically zero for eight straight weeks and was retired. Selection alone beat untargeted marketing several times over.
Two emails. One built value. One only looked like it did.
Same customers, same window, same dashboard. On engagement they were hard to tell apart. Forward CLV told them apart in every single read.
The reminder without an offer
net forward value built per customer
95% CI spans zero · all eight weekly reads
On the engagement dashboard it looked healthy: opens, clicks, conversions within a point of control. Forward CLV read it as noise in every single week. Our recommendation was to stop sending it and reclaim the budget.
The targeted offer
net forward value built per customer
95% CI +$1.30 to +$9.10
A modest offer, sent only to the one-time buyers the model selected. Confidently positive across the full window, and worth roughly a quarter million dollars a year at the client's rate of new high-value customers.
We’re deliberately careful here: this is one brand’s result. We report it as a range, not a guarantee, and half of what it proved is that we will tell you to stop doing something. The only way to know what it means for you is to run the same measurement on your customers.
Find out which of your strategies the wrong metric is hiding.
The free diagnostic scores your customers on forward value and shows you where an engagement-based read is quietly steering you wrong, on your data, against a back-test on your own history.
Read-only access · No deck required · You keep the analysis either way