Adrian, Reggie — this is Brian from ScrappyLabs. I want to give you a five-minute tour of what twelve months of Better Way Health Zendesk data is telling us, and what I think it means for revenue you're either making or leaving on the table. The full written report is attached, but if you're driving or between meetings, this version covers the same ground. Let's start with the headline.

Fifteen point eight percent of your real customers file a cancel signal in any given twelve months. Once they cancel, only one point two percent ever come back. So the leverage is upstream of the cancel — and you have a remarkably predictable upstream signal. When a customer's recharge fails — that "outbound charge errors" call your team makes — that single event is two point eight times more likely to appear in the sixty days before a customer cancels than at random. And the median time between that call and the cancel is twenty-seven days. That's a four-week intervention window, repeating month after month, with a known starting gun.

Now, here's what surprised me. We classified all eight hundred and three voicemail-only tickets your agents leave — Whitney, Michelle, Zach, Sharlene, Taylor — and we found something specific. Every one of them uses two completely different voicemail scripts, and they switch between them deterministically based on what the call is about.

When the call is a charge-error — the highest-leverage moment in the customer lifecycle — the script is purely transactional. Quote, "we had an issue processing your order, please call us back." Zero percent of those voicemails ask for feedback. Zero percent frame the account as inactive. Seventy-six percent ask for a callback. Three percent offer an SMS link to update the card.

But when the same agent leaves a voicemail on a cancellation save call, the script is completely different. Eighty-one percent ask for feedback. Sixty-four percent frame the account as inactive. They mention pricing, they offer reactivation, the voice is warm.

The relational frame is already in your playbook. It's just reserved for the wrong moment. Charge-error voicemails get the cold script. Save calls — which usually come after the customer's already mentally checked out — get the warm one.

On the save calls themselves we found something equally surprising. Of the forty-nine save calls that ended with a customer satisfied or agreeing to stay, the largest bucket of wins — fifty-four calls — came from agents offering nothing concrete. Just a friendly check-in. "I noticed your account went inactive, any feedback for us?" Forty-four percent win rate at scale.

Discount offers were a smaller bucket — twelve calls, sixty-seven percent win rate. Price-lock reminders, where the agent reminds someone they're grandfathered into older pricing, were nine for nine. Undefeated, but small sample. The pattern: low-pressure check-ins outperform aggressive discounting. Price isn't the leading objection in your data. Reaching out warmly is.

So here's what I'd test, single experiment: rewrite the charge-error voicemail. Three changes. One — explicitly name the failed payment instead of the soft "issue processing." Two — fire an SMS link to update the card in parallel with the voicemail, so the resolution path doesn't depend on the customer calling you back. Three — borrow the warm closer your agents already use on save calls.

The baselines are real, not estimated. Recurring charge-error within thirty days is seventeen point seven percent across eleven hundred and forty-four customers. To detect a drop to twelve percent we need about six hundred customers per arm, which is four to five months running parallel. The SMS click-through is the immediate signal — control is structurally zero, treatment is anything positive.

One constraint to call out — most subscription cancellations probably happen via the Recharge UI without producing a Zendesk ticket. So the cancel rate we measure from the support data is a floor, not the truth. Getting Recharge subscription state by date would let us measure actual revenue retention rather than ticket-trail proxies.

Bottom line — the script library is the cheapest lever you have. The training is already in your team. We have the experiment ready. I'm happy to walk through any of this on the next call.