How a four-location IV hydration chain cut month-eight churn from 31% to 12%

This is a case study, written with the permission of the practice owner, with the brand and exact numbers obscured enough to protect them from competitive scrape. The substantive findings — the churn shape, the intervention, the result, and the failures along the way — are reported faithfully. The practice is a four-location IV hydration brand in a major US metro, running 1,100 active members across the four sites at the start of the engagement. Their membership product was a $99/month all-access drip plan with one drip per month included and discounted add-ons. Healthy by industry standards. Quietly bleeding revenue at month eight of every cohort.

The setup

The owner came to us in late summer 2025 with what they described as a "growth problem" — new member acquisition was strong, but total active members had been flat for two quarters. The first thing we asked for was the cohort churn curve, broken out month-over-month from signup. They had not run that report. The clinical coordinator pulled it inside of two days, and the shape was immediately clear: month-one churn around 14%, a smooth glide path through months two through seven (1-3% per month), then a sharp spike at month eight to roughly 31% churn within a single month, then a gentler decline curve from there.

The month-eight cliff was not anomalous — we have seen versions of it at every IV brand we have audited. The novelty period of having a membership wears off around six to eight drips in, the perceived value plateaus, and the recurring charge stops feeling like a treat. Without an intervention in that window, you lose roughly a third of your eight-month cohort. The math compounds: every cohort loses a third of its members at month eight, then the curve continues to decay from a smaller base. Steady-state member counts are fixed by acquisition velocity rather than by retention.

The hypothesis

The hypothesis we built with the owner was that month-eight churn is reactive — members do not "decide" to cancel at month eight; they fail to find a reason to re-engage and they drift, and then a quarterly credit card review surfaces the recurring charge and they cancel reactively. The implication is that the intervention needs to land before the drift starts, ideally between months six and seven, and needs to do two things: revive perceived value and recommit to a forward-looking treatment plan rather than a transactional drip.

The version of this hypothesis that did not survive contact with reality was that a discount or a freebie would do the work. The owner had tried that the previous summer — "free add-on with your next drip" pushed at month seven via email blast. The response rate was sub-5%, and post-hoc analysis showed it accelerated cancellations rather than reducing them. The mechanism: a discount frames the membership as transactional, which is the opposite of what you want at the moment the patient is questioning whether the value is there.

The intervention

The intervention we built was a four-touch outreach loop, timed to month seven of membership, designed to land before the month-eight churn window. The four touches were structured around clinical perceived value rather than discount or promotion. Each touch was orchestrated through the unified inbox so the patient experienced one continuous conversation across channels rather than four separate marketing pushes.

Touch 1 was a personal voice call from the clinical lead — actually placed by a voice agent with the clinical lead's voice cloned (with explicit consent), delivering a check-in framed around the member's drip history. Specific protocols they had received, specific add-ons they had liked, and an open-ended question about how they were feeling. The voice agent's job was to make the call happen at scale; the actual response handling routed back to the clinical lead's queue for any non-trivial reply.

Touch 2, three days later, was an in-thread iMessage with a proposed protocol upgrade — a new add-on or a seasonal protocol — framed as "I wanted to suggest this for the next time you come in" rather than as a sales push. The proposed protocol was actually generated by a small recommendation model running over the member's drip history and presented to the clinical lead for review before being sent.

Touch 3, one week later, was a calendar invite to a pre-scheduled next drip slot — the slot generated by the system, not chosen by the member. The invitation framing was "your next drip is on the books for {date}, want me to lock it in or swap it?" The mechanical commitment to a future slot was the highest-impact lever in the entire sequence.

Touch 4, two weeks later, was a slow email recap from the clinical lead — short, signed personally, summarizing where the member was in their wellness arc and what was coming up next. The email had no call-to-action; it was deliberately not a marketing email. The reason for including it was that the clinics where this worked best were the ones where the patient felt seen as a person across multiple touchpoints, not as a customer being asked to act at each one.

The build

Mechanically, the loop was built on the unified inbox tier of the Tality platform with three external integrations: the practice's existing Mindbody member management for cohort dating, ElevenLabs voice cloning for Touch 1 (with the clinical lead's recorded consent), and Twilio Messaging for the iMessage and SMS legs. The model stack for the voice agent was Deepgram Nova-3 for STT, Claude Sonnet 4.6 for the conversational handling, and the cloned voice for synthesis. Total implementation time from kickoff to first member receiving Touch 1: 18 business days. Most of that was getting the voice cloning consent paperwork through the clinical lead's attorney.

The result

The intervention ran on the September 2025 cohort first, with the August cohort as a held-out control. Month-eight churn on the September cohort (the treatment group) came in at 12% versus the August control at 31%. The absolute reduction held across the next three cohorts measured (October, November, December 2025), settling into a steady-state churn at month eight of 11-14% versus the historical 28-33% baseline.

The financial framing on this is uncomfortable when you actually run it. The retained members from the four treated cohorts represented roughly $310,000 of annualized recurring revenue that would have churned. Net of the implementation cost and the ongoing operational cost of running the loop, the practice booked roughly $278,000 of incremental ARR off an intervention that took 18 business days to build. That is the most under-discussed economic shape in the wellness vertical.

The failures along the way

Three things did not work in the first three weeks of running the loop, all of which are worth documenting because they are the kinds of things that get glossed over in a clean case study and end up being most of the work.

First, the voice cloning was rejected by roughly 8% of members on the first call. They could tell. The cloning quality was high but not perfect, and members who had had a long real-voice relationship with the clinical lead caught the synthesis. We made two changes: we re-recorded the source audio with a longer training sample, and we changed the opening line to be a deliberate AI disclosure ("Hi {first_name}, this is the AI assistant calling on {clinical_lead}'s behalf") which paradoxically increased trust because it framed the call honestly rather than letting members feel deceived.

Second, the calendar invite in Touch 3 was initially generating slots that conflicted with the member's known preferences (we were defaulting to the next available slot rather than the slot that matched their historical drip timing pattern). The fix was to weight the slot recommendation against the member's last six drip slots, which doubled the slot acceptance rate on the calendar invite.

Third, and most consequentially, we initially ran the loop on the entire cohort at month seven. After two cohorts we discovered that members in the lowest-engagement quartile (fewer than 4 drips in their first 7 months) were churning at a higher rate when treated than when left alone. The intervention activated dormant accounts and surfaced the recurring charge in a way that triggered cancellation rather than re-engagement. We now suppress the loop for the bottom quartile of engagement and route those members into a different conversation entirely — a clinical re-onboarding rather than a retention push.

What this transfers to

The specific intervention here was built for IV hydration membership at month eight. The architecture pattern transfers cleanly to any wellness business with a recurring revenue model and a predictable engagement decay window. The two transferable elements are: identify the churn-cliff month in your cohort data rather than guessing, and intervene before it with a multi-touch loop that frames the relationship clinically rather than transactionally. Discounts do not work; perceived clinical engagement does.

The owner of this practice has since rolled the same loop across all four locations and is now running a similar architecture at month fifteen to defend against the second smaller churn wave that appears around that time. We will write up the month-fifteen extension if and when we have a year of data behind it. For now, the working answer for any IV hydration operator who is reading this and thinking "do I have a month-eight problem": run the cohort report this week. The number will tell you whether the intervention is worth your evaluation cycles or whether your churn shape lives somewhere else.