The short version
In a subscription health product, almost every message a patient gets after signing up is triggered by something the product recorded: they stopped at identity verification, a doctor prescribed, a card was declined, a refill is two days away. Lifecycle marketing (LCM) turns those moments into the right email or text, and should stop when the moment has passed.
This guide covers the funnel those messages work on, which product events mark each stage of a patient's journey, how the events reach the messaging tool and come back, and what was built on them. It is a sibling to building click-funnel events you can trust and event properties: what matters and how to construct them, which cover the events and properties themselves.
The examples come from a telehealth subscription company in weight-loss and longevity treatment, where I was Head of Product and led the product and engineering team. The lifecycle team owned the messages; we owned the events. Figures are percentages or ratios from production analytics, mostly from August to October 2026, with windows noted where they differ. Event and property names are simplified from the real schema to follow the conventions in the sibling guides. Throughout, "checked out" means completed the purchase, and "started checkout" is the step before.
Three ideas run through what follows: events, not schedules, start and stop messages; the messaging tool needs the moment, not the medical details; and a message has only worked if you can show what would have happened without it, which most programs, ours included, cannot yet do.
Events, attributes and the messaging tool
A lifecycle tool such as Customer.io or Braze holds two kinds of data about a person, and getting the split right matters more than any template.
- Attributes are state. Consent flags, the onboarding version seen, the subscriptions held and when each next bills. An
identifycall overwrites them; Customer.io templates read them as{{customer.X}}. - Events are moments. Checkout completed, a payment failed, a refill is due. A
trackcall sends them, they can start a journey, and templates read their properties as{{event.X}}. - Journeys need both, plus an exit. An event starts the journey, attributes and properties choose the branch and fill the copy, and another event ends it: a failed-payment journey starts on
Payment Failedand should exit onPayment Succeeded. In Customer.io, exit conditions are separate from conversion goals, so a journey can count a conversion and keep sending. Set both.
The funnel the messages work on
The funnel is the patient journey seen as shares. Each stage is also an audience: the people who stopped there are who the next message is for, and the event that moves them on is that message's exit. For new accounts created in August 2026, followed for 60 days:
| Stage | Share of new accounts | Lifecycle messages that work on the gap |
|---|---|---|
| Account created | 100% | Abandoned-onboarding email keyed to the last step reached |
| Questionnaire completed | 87.1% | Treatment-choice and identity-verification stage emails, aimed at the biggest leak: only 20.9% of questionnaire finishers start checkout |
| Started checkout | 18.2% | Checkout-stage email |
| Checked out | 7.9% | Receipt, welcome and prescription updates; failed-payment sequence |
| First payment | 4.8% | Follow-up reminder, refill heads-up, failed-payment sequence |
| Second payment | 3.7% | From here on: refill heads-ups, cross-sell and cancellation journeys |
After checkout, the patient no longer controls the pace. Of those who checked out, 99.0% were accepted as patients and 80.8% were prescribed. Clinicians and billing set the timing from here, so the messages in this stretch are mostly about explaining what happens next.
The patient journey, stage by stage
Each stage of the journey, the event that marks it, the message it triggers, the properties that message needs and how to judge it. The last two columns are recommendations.
| Stage | Event that marks it | Lifecycle message | Properties it needs | How to judge it |
|---|---|---|---|---|
| Lead or waitlist | Lead Captured, Waitlist Joined | Lead offer; launch email when a new territory opens | Capture method, first campaign | Onboarding starts from the message's links |
| Started onboarding, did not finish | Last step reached: Questionnaire Completed, Treatment Chosen, Identity Verification Started, Checkout Started; should stop at Checkout Completed | One email per drop-off point, then a follow-up at 15 to 45 days | Last step reached, onboarding version | Checkout within 7 days, against unsubscribes |
| Purchased, awaiting prescription | Checkout Completed, Patient Accepted, Prescription Issued | Receipt (with an expedited-shipping branch), welcome, post-prescription series | Product class, billing cycle | Prescribed, then paid, within 30 days |
| Verification problem | Identity Document Rejected | "Upload rejected", asking for a new document | Rejection reason | Re-upload and completion |
| In treatment | Order Shipped, Provider Message Sent, Follow-up Requested | Shipping notice, "your doctor sent you a message", follow-up reminder | Product class at most; the message itself stays in the portal | Follow-up completed within 14 days |
| Refill or dose change | Refill Reminder Due, Dose Change Due | Heads-up two days before the charge or change | Refill or change date; dose details shown in the portal | Charged, failed or cancelled within 10 days |
| Payment problem | Payment Failed, Card Expiring; recovery is Billing Updated then Payment Succeeded | Failed-payment email and SMS; card-expiry email | Amount, invoice reference | Paid within 7 days |
| Cancelling | Cancellation Started, Subscription Cancelled, Subscription Paused, Subscription Reinstated | Cancellation email, reason-targeted follow-ups, a one-time recapture link | Reason, who initiated it, product category | Reinstated within 30 days |
| Ready for more | active_subscriptions attribute, Add-on Purchased | Cross-sell to another product line around day 30 or 45 | Active subscriptions, product class | Purchase of the second product |
| Advocate | Referral Sent | Referral invitation | Referral link | Referred accounts that check out |
Where messages have room to act
Timing tells you what kind of message belongs at each stage. We estimated it by re-running funnels with windows from one hour to 60 days.
- Leads are mostly not patients yet. Most captured leads came from shop and newsletter sign-ups, and only 4.4% created a patient account within 30 days. Lead messages have to earn the next step, not assume it.
- Account to checkout is mostly fast. Of new patients who checked out within 60 days, 74% did so within an hour, 82% within a day and 91% within a week. Abandoned-onboarding emails, the first arriving about two hours in, can only work on the slower quarter.
- Prescription and first charge are system-timed. 95% of prescriptions came within a day of checkout, and the median first charge about a day and a half after it. A meaningful share of prescribed patients had still not paid 30 days later, so messages here should set expectations, not push.
- Rebills have a rhythm. The first rebill lands about 21 to 22 days after the first payment, later ones every 27 to 28 days, so refill messages have a predictable slot.
- Payment recovery is slow. After a failure, 20% had paid within a day, 37% within a week and 52.5% within 30 days, so there is room for a sequence.
How the data flows
The app never sent a patient an email or text itself. Templates, delays, branches and exits lived in the messaging tool; the codebase owned the events.
- The product records the moment. Browser and backend send
trackandidentifycalls to Segment. Server calls go through a durable outbox that retries four times with exponential backoff and sends identify calls ahead of track calls. - Segment fans it out to Customer.io for email and SMS, to Mixpanel, and to any other destination that needs it. A few calls, such as an identify client and profile deletion, went to Customer.io directly.
- Engagement comes back. Customer.io's delivery and engagement events return through their own Segment source into Mixpanel, on the same user timeline as the product events.
- Links close the loop. Every link carries UTM tags, so a checkout after a click is recorded with the message that preceded it.
- One ID everywhere. The account's user ID, minted at lead capture or signup, is the ID in Segment, Mixpanel, Customer.io and the patient record, and the identify call at that point ties earlier anonymous browsing to it. 91% of server-side accounts traced back to a browser onboarding start.
- A deliberate delay. Checkout and onboarding-complete events wait at least five minutes, so attribution and the patient record have settled before any journey reads them.
- Deletion reaches the messaging tool. Deleting a patient in production hard-deletes their Customer.io profile too. Every other destination needs its own check.
Delivery shows whether it works: 92.6% of new accounts got at least one email within 14 days, and 99.0% of checkouts got one within 7 days, usually the receipt, about six and a half minutes after purchase. The other 7.4% of new accounts are the gap to monitor (example 8).
A checkout event, simplified, with illustrative names and invented values. It names the product class outright, because a journey should never have to work it out (example 3 shows why).
{
"type": "track",
"event": "Checkout Completed",
"userId": "3f9c2a7e-51b4-4c0e-9d1a-7b2e8c40f611",
"timestamp": "2026-10-14T18:27:05Z",
"properties": {
"product_class": "glp1",
"billing_cycle": "monthly",
"onboarding_version": "v2",
"utm_source": "customer.io",
"utm_campaign": "abandoned_onboarding_identity_1"
}
}
A profile update, with trait names in the same style and all values invented. The subscription list is re-sent whenever a treatment is created, started, prescribed or paused. The backend builds the referral link and passes it as a trait, so no patient ID appears in a URL; the design for recapture links follows the same pattern.
{
"type": "identify",
"userId": "3f9c2a7e-51b4-4c0e-9d1a-7b2e8c40f611",
"traits": {
"first_name": "Alex",
"onboarding_version": "v2",
"sms_opt_in": true,
"active_subscriptions": [
{
"product_class": "glp1",
"status": "active",
"next_billing_date": "2026-11-04",
"billing_cycle": "monthly"
}
],
"referral_link": "https://portal.example.com/r/k7Qm2xR9"
}
}
What comes back from Customer.io uses its own Title Case event names (Email Delivered, Email Link Clicked, Email Converted, SMS Failed) with snake_case properties, and trigger_event_id points to the product event that started the journey. Values invented:
{
"event": "Email Converted",
"properties": {
"campaign_name": "abandoned_onboarding_identity_1",
"journey_id": "jrn_example_01",
"action_name": "Email 1",
"template_id": 1234,
"delivery_id": "dlv_example_01",
"trigger_event_id": "evt_example_01"
}
}
Applied examples
Aggregate patterns from the same company, not individual patients. Figures are mostly from 30-day windows in September and October 2026, with other windows noted.
1. Abandoned onboarding, rebuilt by stage
Situation. Everyone who created an account without checking out got the same abandoned-onboarding flow, wherever they stopped. In October 2025 the lifecycle team proposed flows keyed to the exact drop-off point.
What was built. Six stage emails keyed to the furthest onboarding step reached and meant to stop at purchase (current versions date from December 2025), then a 15-to-45-day email and a long-tail one.
Result. These emails accounted for 63% of checkouts carrying a lifecycle link tag. The identity-verification email had the biggest stage audience, matching the weakest step in the onboarding funnel: just over a third of those who started identity verification went on to start checkout. About 7% clicked, 7% converted on the tool's goal and 5% unsubscribed. The treatment-type and treatment-choice emails did worse, with unsubscribe rates (7.8% and 6.7%) more than double their 3.2% click rate. Among August 2026 accounts followed for 60 days, clickers converted at 21.6% against 2.3% for recipients overall, but clickers select themselves, so that gap cannot be read as the email's effect.
Lesson. Splitting by stage makes sense, because the stage tells you what to say, though we had no before-and-after to prove it. Every stage email costs list, and some stages lose more than they win. These emails are separate from the redesign in the onboarding case study, which moved more people to start checkout without a clear increase in purchases; they work on the people the funnel still loses.
2. The long tail that cost more than it earned
Situation. A reactivation email went to people 120 to 150 days after abandoning onboarding. Nothing had happened to them; only the clock moved. Its audience was about eight times the biggest stage email's.
Result. 0.27% clicked, 1.0% unsubscribed and about 1 in 10,000 recipients checked out: roughly 100 unsubscribes per checkout. Broadcasts looked similar: only 2.8% of onboarding starts from October's newsletters became checkouts, against 13.6% for lifecycle-tagged traffic overall.
Lesson. A message triggered only by the clock is a broadcast in disguise. Report unsubscribes per checkout, and reactivate only with a real reason, such as a new product line or a price change.
3. A new product line broke "if not A, then B"
Situation. Journeys decided a patient's product line by elimination: not line A, so it must be line B. That broke when a third product line launched.
What was built. Engineering and the lifecycle team agreed to add one explicit property, product_class, to the checkout event and leave every existing property alone, so live journeys kept working.
Result. In the rollout week, 12.5% of checkout events lacked the property; from the next week, none did. Part is still open: the messaging tool still recognized one line only by a version tag, so a patient buying an add-on to that line could look like a patient on another, and the lifecycle team has asked for an explicit product category.
Lesson. Send categories explicitly, add properties rather than changing them, and check coverage weekly until it hits 100%.
4. Failed payments across email, SMS and automatic retry
Situation. Over the past year, about 6% to 9.5% of paying patients had a failed payment each month.
What was built. The lifecycle team's email and SMS sequence starts on Payment Failed (amount, invoice reference); recovery shows up as Billing Updated then Payment Succeeded, and those are the events it should exit on. The current version dates from January 2026. In October 2026 we added automatic retries: up to six over about seven weeks, front-loaded in the first three days, before an invoice is marked uncollectible.
Result. For failures from September 1 to October 20, 2026:
- Fast, and followed by recovery. 87% of failures got the email within 7 days, about 30 minutes after the failure on average. On the tool's own conversion goal, the email converted about 29% and the SMS about 28%. Within a week, 72.4% of clickers paid against roughly 20% of non-clickers.
- Recovery clusters after the first email. Later emails were followed by recovery in only 3% to 15% of cases, though by then they reach the harder cases.
- A test that proved nothing. A new first email recovered 48.4% against 43.5%, not significant (p about 0.2) on an 81/19 split.
- Retry. In its first week, retry recovered invoices for patients who never touched their card, nearly 4 in 5 of them on the first attempt.
Lesson. Exits need as much design as messages. Ours did not stop on cancellation at first, so check that every journey exits on success, on cancellation and on anything the system does by itself, such as a retry. A simple check catches leaks: compare the people a sequence reaches each week with the people who entered it, and look into any gap the sequence's length does not explain. Once the system recovers payments by itself, the messages must know what it is doing.
5. Inside treatment: follow-up, refill and cross-sell
After the first payment, lifecycle messages mostly support care rather than sell, and the three kinds behaved very differently.
- Follow-up reminders. The clinical follow-up reminder was the most-acted-on message we had: about 64% clicked and 48% converted on the tool's goal. 83% of patients completed their follow-up within 14 days, half of them after day one, which is where the reminders act.
- Refill heads-ups. After the two-day notice, 85.5% of patients were charged, about 2.2 days after the email; 4.3% had the charge fail and 3.6% cancelled. A notice that close to the charge mostly prepares the patient rather than changing the outcome, and the lead time is a product decision, not a messaging one.
- Cross-sell. The 30-day check-in and upsell email drew 10.4% clicks but about 0.1% conversion, and the 45-day upsell 2.4% clicks. After an in-portal card for one product line launched, portal purchases of that line rose by about 70% week over week: a small, early sample, but a better sign than the email.
Lesson. Messages tied to something the patient already means to do, such as finishing a follow-up, get acted on. A new decision, such as adding a product, does better where the patient already is, in the portal. Cross-sells also raise the HIPAA marketing question covered below.
6. Cancellation: the save happens in the product
Situation. Cancelling triggered a cancellation email, reason-targeted follow-ups and, for some, a one-time recapture link.
Result. Cost was the most common reason given, at 33.7%. 83.4% of people who opened the cancellation flow cancelled within a day, and only 1.4% of them applied the save discount offered in the flow. Afterwards, the cancellation email converted 1.8% on its own goal and no checkout in the window carried its link tag; 4.9% of cancellations were reversed within 30 days. One reason-targeted follow-up lost 12.7% of its recipients to unsubscribes.
Lesson. The save moment is inside the product flow, while the patient is deciding, not in the email that follows. Spend effort there first, and keep post-cancellation messages few and clearly useful.
7. Upstream changes that broke, or nearly broke, messages
- Lowercase names. The tracking layer lowercased first and last names, which suits matching in other destinations. The same attribute fed email, so greetings went out in lowercase. The fix shipped in October 2026, with a backfill planned for existing profiles.
- A survey format change. A message was triggered when a satisfaction score crossed a threshold. When the survey switched from a 1 to 10 score to thumbs up or down, the trigger silently stopped firing until the change was reverted.
- Moved pages. Template links to specific portal paths would have returned 404s after a portal rebuild without redirects, so the rule became that every lifecycle call to action uses the portal's root URL.
Lesson. A journey is a contract with every upstream system. Write down what each trigger depends on (event and property names, value formats, link target) and treat changing any of them as changing the journey.
8. Messages nobody could trace
Situation. A patient's messaging profile can fail to sync without an error anyone sees, and then every message for that patient silently stops.
The repair tool. An admin button re-sends a patient's full profile: useful, and exactly where unneeded fields accumulate, so review it field by field.
A second case. One journey, an onboarding email sent after an in-person purchase, was triggered from outside the product, and nobody could quickly say what fired it or why it sometimes failed. About half the onboarding starts it drove ended in a checkout, so it mattered.
Lesson. Keep a register of every journey (trigger, exit, goal, owner, source system) and monitor the share of new accounts with a messaging profile and a first email. Nobody notices when an untraceable message stops.
The rules for health products
The analytics side is in the privacy section of the properties guide; these rules are specific to messaging. This is not legal advice, so have counsel confirm each one. Vendor terms are as documented in October 2026.
- Keep health details out. The messaging tool needs to know the moment, not the reason. Diagnoses, test results, weights, doses, questionnaire answers, doctor and patient messages and card details do not belong in it. A coarse product class is about the most sensitive thing most journeys need, and it still belongs under a business associate agreement (BAA).
- Link to the portal. Customer.io's HIPAA guidance says to keep protected health information out of messages and link to a secure portal where the patient logs in. "Your doctor sent you a message" plus a link is the pattern, and about 57% of recipients clicked it.
- Know what each BAA covers. Customer.io signs one on request, and your SMS provider needs its own. Braze's applies only to a designated account on its HIPAA cluster, says the service is not to be relied on for patient treatment and bars submitting diagnoses or test results. Check every vendor in the chain.
- Separate care from marketing. Under HIPAA, a message encouraging a purchase is marketing and generally needs authorization. Refill reminders for a currently prescribed drug are excluded (with conditions on payment), as are messages about treatment and the provider's own services when no third party pays; a related product the patient is not prescribed does not fit the refill exception. Tag every journey as care, transactional or marketing, and let counsel place each cross-sell.
- Get consent per channel. Record SMS consent separately at signup. Marketing texts sent by autodialer need prior express written consent under the TCPA; there is a health care message exception from the written-consent requirement for covered entities and their business associates, but don't assume a promotion fits it. Honour "stop" and similar replies within 10 business days and keep solicitations between 8 a.m. and 9 p.m. local time. For email, CAN-SPAM needs a postal address and timely opt-outs, Gmail and Yahoo require one-click unsubscribe for bulk marketing, and CASL needs consent and a working unsubscribe for Canadian recipients.
- Keep marketing off the transactional path. Customer.io's transactional API sends to unsubscribed people by default: right for a receipt, wrong for a promotion.
- No identifiers in URLs. Our recapture discount links (for follow-up completion, cancellation and failed payment) are locked to the account and usable once per type. The design kept the patient ID out of the URL and treated the eligibility check at redemption as the real guard against double discounts.
- Review copy like clinical content. If the email system cannot receive replies, never invite them. Outcome claims such as weight lost need clinical sign-off, and prescription product imagery must meet FDA rules. Templates live outside the codebase, so add them to your copy compliance review explicitly.
- Delete everywhere. Deleting a patient should delete their messaging profile, in this tool and the next.
Not a HIPAA covered entity? Health data is still regulated. If you run a health app or personal health record, the FTC's Health Breach Notification Rule counts an unauthorized disclosure as a breach, and the FTC has used its general powers against health advertisers too. Washington's My Health My Data Act requires opt-in consent and lets individuals sue.
Measuring whether lifecycle messages work
The real question is whether the patient did something they otherwise would not have. First, the reporting has to be trustworthy, and ours had gaps.
- Opens are not a metric. Use clicks for engagement and product events for outcomes.
- Know what "converted" means. Each campaign defines its own goal; one promoting a live Q&A showed about 74% of recipients "converted". Customer.io counts conversions per campaign, so summing campaigns can count one checkout several times. Use a server-side product event as the goal.
- Clicks are self-selected. A 72% recovery rate among clickers does not mean the email caused 72% of anything.
- Link-tag attribution is not lift. The 8.1% headline counts checkouts carrying a lifecycle tag; it includes people who would have bought anyway and misses those who came back without clicking.
- Campaign names are permanent keys. Typos became analytics values for good, and inconsistently stripped commas split one campaign into two rows.
- SMS is a blind spot. Delivery figures for texts disagreed between sources, clicks were recorded on well under 1% of deliveries, and almost every SMS recipient was getting shipping notices, with almost no SMS in abandoned onboarding.
Holdouts that are hard to see
Several journeys already had holdout or control arms. Customer.io still logs a delivery for a held-out message, routed to an internal trap, so in analytics these arms look sent. The tell is that not one of those held-out deliveries was opened or clicked, while ordinary deliveries showed opens on about 95%. They are real no-send groups, but recent, small and identifiable only by step name, so they could not yet show lift.
- Customer.io: per message. Holdouts are set per message (variant B of an A/B test, or a random cohort branch), and held-out people can still register conversions. Name holdout steps consistently, and judge on conversion, since the sent variant always wins on clicks. For a program-level holdout, assign a random bucket attribute once and filter on it in every workflow.
- Braze: one global group. A Global Control Group can be 1% to 15% of users; Braze recommends no more than 10%, at least 1,000 users and at least a month, ideally a quarter. Control users still receive API campaigns, which suits transactional health messages.
- Size it and seal it. Small lifts need very large groups, so run a power analysis first, and hold out every channel or the SMS contaminates the control.
Windows and the stratification lesson
Set attribution windows from time-to-convert data, not vendor defaults. Braze falls back to a 3-day window when no conversion event is set (configurable up to 30), while our checkouts after an abandoned-onboarding email came about 10 days later on average, and most payment recoveries came after day one.
Then compare like with like. In the onboarding case study, the new version had about a fifth of the weakest week's new accounts and about three quarters of the strongest week's, while conversion moved for both versions. Pooling the weeks credited the new version with that difference, and comparing within each week removed most of the apparent lift. Lifecycle messages fall into the same trap when a holdout's share changes midway or a journey launches in a good month. Keep the holdout share fixed, stratify by lifecycle stage and by week before comparing, and count only people who have had their full window.
Migrating messaging tools without breaking journeys
If you ever move lifecycle tools, for example from Customer.io to Braze, most of the risk sits outside the tool. What to plan for, with vendor details as documented in October 2026:
- Make events the interface. When the product holds no message logic and nearly everything flows through a pipeline such as Segment, the product side of a move is mostly a change of destination, and Segment can rename events and properties per destination. The journeys and templates still have to be rebuilt. List every direct call the product makes to the old tool; ours were few, such as an identify client and profile deletion.
- Inventory every journey. Trigger, branches, exit, goal and source system, including anything fired from outside the product, like the retail onboarding email in example 8.
- Keep the ID, and test signup. Use the same UUID as Braze's external ID; Braze advises against guessable IDs and email addresses. Anonymous history behaves differently: Customer.io merges up to 30 days of it on identify, but only events from the last 72 hours can trigger a campaign, while Braze merges an anonymous profile only into an external ID that does not exist yet.
- History stays behind. Event history rarely moves with you, so re-send the state journeys need (subscriptions, next billing date, consent) as attributes at cutover.
- Check limits against payloads. Braze caps string values at 255 characters and, by default, allows 20 segmentable event properties per workspace.
- Warm the domain properly. SPF, DKIM and DMARC with alignment first. Leaving a shared IP means warming a new domain and IP on a conservative schedule, with low-engagement lists, like the long-tail email in example 2, kept out of the warm-up.
- Plan the overlap. Decide what is frozen in the old tool, because anything built there during the move gets built twice.
- Give it an engineering owner. Even with little code to change, events, identity, links and deletion belong to engineering, and a move that starts as a marketing project needs an engineering owner assigned early.
A checklist
- Mapped. Every funnel stage has an event that marks it and an event that ends it.
- Explicit. Product class and onboarding type are sent, never inferred by elimination.
- Exits. Each journey exits on success, on cancellation and on what the system does by itself, such as a retry.
- One ID, settled data. The same UUID everywhere, and purchase events sent only once attribution and the patient record are ready.
- Registered. Trigger, exit, goal, owner and source system written down for every message.
- No health details. Nothing clinical, no doses and no card details in the messaging tool, with the full-profile sync reviewed field by field.
- Covered. BAAs across pipeline, messaging tool, SMS provider and analytics.
- Classified. Every journey tagged care, transactional or marketing, and SMS consent recorded separately.
- Measured on outcomes. Server-side goals, unsubscribes per conversion, and a fixed, all-channel holdout compared within stage and week.
The takeaway
Lifecycle messaging is only as good as the events beneath it, and the funnel tells you where each message belongs. The messages patients acted on most were tied to a precise moment: a follow-up due, a doctor's message waiting, a payment that had just failed. Several of the ones that cost the most, or went wrong, ran on the clock alone, guessed what the patient had bought, or lacked an exit. Get the events, the exits and the identity right first, then use holdouts that can prove the messages matter, because a click rate cannot.