"We sent less and sold more, and it still felt wrong for two weeks."
A DTC coffee-subscription brand came to Sendnexa with a cart-recovery flow converting at 1.9% and a spam-complaint rate of 0.19%, nearly double Google's 0.1% guidance. Eight weeks later the flow converted at 3.1% on 22% fewer sends. Cole Bergeron ran the flows; Dana Okafor ran the models. The interview is lightly edited.
| Metric | Before | After (8 weeks) |
|---|---|---|
| Cart placed-order rate | 1.9% | 3.1% (category avg 3.33%) |
| Campaign click rate | 1.2% | 1.8% (Klaviyo avg 1.69%) |
| Send volume | baseline | −22%, revenue flat |
| Spam-complaint rate | 0.19% | 0.06% |
What was actually broken when you opened the account?
Cole: The cart flow itself was fine on paper: three touches, decent copy, a discount on touch two. The problem sat around it. Campaigns went to the full list twice a week, including a large block of profiles that had not bought in months, and that pushed the complaint rate to 0.19%. Cart abandonment runs at 70.2% on Baymard's numbers, so the flow mattered, but the reputation damage from campaigns was taxing everything the flow tried to do.
Where did the models come in?
Dana: Klaviyo's churn-risk score was already computed on every profile; nobody had built anything on it. We created a suppression segment of 18,900 high-risk profiles and pulled them out of routine campaigns in week one. That is the entire origin of the "−22% sends" number. Revenue did not move, which told us those sends had been producing complaints, not orders.
What did you change in the cart sequence itself?
Cole: We went from three touches to four: 2 hours, 20 hours, 48 hours, and 5 days. The catch is that late touches generate most of the complaints, so touch four only sends to profiles the model scores as low churn risk. High-risk profiles get three touches and stop. That let us add sending pressure where it was safe and remove it where it was not.
What did not work?
Dana: The first fortnight looked bad. Cutting a fifth of send volume drops your topline dashboard numbers immediately, and the conversion gains arrive later. We had modelled that lag, and it still took a tense weekly call to hold the line. There was also a failed subject-line test in week five: the sample could only resolve a 1.5-point difference and the result came in at 0.4, so we reported it as inconclusive rather than shipping a fake winner.
What should someone copy from this engagement?
Cole: Gate your late-sequence touches on churn risk. It is one segment and one flow filter in Klaviyo, maybe an afternoon of work, and it attacks complaints exactly where they concentrate. The full method is what ouraudience-analysis service systematizes, and the process page shows where it lands in an engagement timeline.
Client identity anonymized. Figures are representative engagement data, consistent with Klaviyo's published benchmark ranges (cart flows average a 3.33% placed-order rate and $3.65 revenue per recipient).
Your cart flow has a version of this problem
Most do. Bring the account to a working session and we will find it in 45 minutes.
Book a working session