The standard view in digital health is that engagement is a content problem. Send better reminders, build a nicer app, add a streak counter, and members will stick. Most of the members who leave a programme do not leave because the content was dull. They leave because something boring went wrong: the eligibility check stalled, the follow-up never got booked, the refill paperwork sat unsigned, an insurance question went unanswered for a week. Engagement is mostly an operations problem, and that is exactly the part AI can take on safely.

This guide is for operations and CX leaders at digital health and telehealth companies. It maps where members drop off, what an AI account manager should do at each stage, where a human or clinician must take over, which guardrails matter, and how to prove the effect with a holdout group. Lorikeet publishes this guide and sells an AI account manager, so read the product section with that in mind.

Key takeaways

• Most drop-off in digital health happens in the operational gaps between care: onboarding, scheduling, refills and renewals, paperwork and billing. Those are the steps AI should own.

• AI never gives clinical advice, dosing guidance or symptom assessment. Clinical questions and anything urgent go straight to clinicians, with context attached.

• An AI account manager works toward a retention or adherence goal for one member at a time, and its next step can be a message, a task for your care team, or nothing.

• The guardrails that matter are consent, contact hours, minimal health information in SMS and email, a reason logged for every decision, and fast escalation.

• Measure it against a holdout group on the goal itself, such as follow-ups completed or members still active at 90 days, not on message opens.

Why adherence and retention are hard in digital health

The problem is old and well documented. The World Health Organization's report Adherence to long-term therapies: evidence for action found that adherence to long-term therapy for chronic illnesses in developed countries averages 50%. A review in Mayo Clinic Proceedings, "Medication Adherence: WHO Cares?", puts it the same way: approximately 50% of patients with chronic illness do not take their medications as prescribed, at an estimated cost of around $100 billion a year in the US.

Appointments follow the same pattern. A systematic review of no-shows in appointment scheduling published in Health Policy in 2018 put the average no-show rate across 105 studies at around 23%, and found that the most commonly reported predictors were a long lead time between booking and appointment and a history of previous no-shows. Both of those are operational signals your systems already hold.

Digital health adds its own twist. There is no waiting room, no receptionist who notices a member has not been in for a while, and often no single clinician who owns the relationship. The member's experience is a series of handoffs between an app, a pharmacy, a payer and a care team. Every handoff is a place to quietly fall out.

Where members drop off in a digital health journey

Map your own funnel before buying anything. In most programmes the losses cluster in the same places.

Eligibility and onboarding

A member signs up through an employer or health plan, then hits an eligibility check, an identity step or an intake form. If any of those stalls, they rarely write in. They just stop. Our guide to onboarding eligible members covers this stage in detail.

First and follow-up appointments

The first consult is often booked in the moment. The second one is where programmes leak, because it needs the member to act again, days or weeks later, with no prompt from a person.

Refills, renewals and prescription logistics

The clinical decision is made by a clinician. The logistics around it are not clinical: a renewal form, a pharmacy transfer, a delivery address, a prior authorisation document. Members who hit friction here often drop the programme along with the refill.

Paperwork, billing and insurance

Unexpected bills, unclear coverage and unanswered insurance questions are among the quietest churn drivers. The member does not complain. They cancel.

Going quiet

Some members simply stop logging in. Silence is a signal, but it is an ambiguous one: they may have hit a barrier, they may be doing fine, or they may have moved on. Treating every quiet member the same is how programmes end up spamming the people who are doing well.

What an AI account manager does at each stage

An AI account manager is an always-on AI that your team gives a long-term goal, such as activation or retention, and that then works toward that goal for each member individually. It wakes on a schedule and when something changes, reads what it knows about that member, and decides the next step. In digital health that next step must stay on the operational side of the line.

Two things in that table do most of the work. The first is that "no action" is a legitimate next step. A member who is booked, refilled and active does not need a nudge, and sending one anyway erodes trust. The second is that the handoff column is written in advance by your clinical and compliance teams, not improvised by the AI.

Why reminders alone are not enough

Reminders help. A 2016 meta-analysis in JAMA Internal Medicine of 16 randomised trials found that text-message interventions improved self-reported medication adherence in adults with chronic disease, while cautioning that the trials were short. A fixed reminder still cannot tell that this member's refill is stuck on a pharmacy transfer, or that they already rang support yesterday about the bill. An account manager can, because it remembers the whole member, not just the last message.

Where the line sits: operations, not clinical care

This is the rule that makes AI safe in digital health, so state it plainly in your configuration and your member-facing copy. The AI handles the operational side of engagement: eligibility and onboarding, scheduling and rescheduling, refill and renewal logistics, paperwork, billing and insurance questions, and re-engaging members who went quiet. It does not give clinical advice, does not discuss dosing, does not assess symptoms and does not tell anyone whether to start, stop or change a treatment.

Clinical questions go to clinicians. Anything urgent goes to your urgent pathway immediately, and the member is told how to get help now. The AI's job in those moments is to recognise the question, hand it over with the context attached and make sure the member is not left waiting without knowing what happens next. For out-of-hours coverage specifically, see our guide to after-hours patient messaging.

Guardrails for proactive patient outreach

An AI that contacts members on its own needs firmer rules than one that only replies. Five matter most. None of this is legal advice: set the rules with your own counsel and privacy officer, then make the system enforce them every time.

Consent

Check consent for the channel and the purpose before every message, not once at sign-up. In the US, the FCC explains that the TCPA requires consent for many autodialed or prerecorded calls and texts. If a member opts out of SMS, that has to hold across every workflow, not just the one they replied to.

Contact hours and frequency

Contact hours in the member's own time zone, and frequency limits across all channels combined. A member who got a billing email this morning does not need an appointment text this afternoon unless it is genuinely time-sensitive.

Minimal health information in SMS and email

HHS guidance says covered providers may communicate with patients by email provided they apply reasonable safeguards, such as limiting the amount or type of information disclosed. HHS's minimum necessary requirement asks covered entities to take reasonable steps to limit uses and disclosures of protected health information to the minimum necessary for the purpose. In practice, write outreach that works if someone else reads the phone: "You have a booking to confirm" rather than the programme name, condition or medication. Keep the detail behind a login.

An audit trail of every decision

Every message sent, and every message deliberately not sent, should be logged with the reasoning behind it in plain language. When a member, a payer or a regulator asks why someone was contacted, "the model decided" is not an answer.

Escalation that actually lands

Handoffs should create a task in the queue your team already works, with the member's history attached, and the AI should stop outreach on that topic until a person has closed it.

How to measure engagement and retention with holdout groups

Open rates will flatter anything. Retention is the number that pays for a digital health programme: as Harvard Business Review notes, acquiring a new customer costs five to 25 times more than retaining one. So measure the retention goal directly.

1. Pick one goal with a clear finish line, such as follow-up appointment completed within 30 days, or member still active at 90 days.

2. Hold out a random share of eligible members who get no AI outreach at all. Keep everything else the same.

3. Compare the goal rate between the two groups over the same period.

4. Track the costs alongside the lift: opt-out rates, complaints, handoffs to your team and how many of those handoffs were useful.

5. Read a sample of decisions every week for the first month, especially the ones where the AI chose to do nothing or chose to escalate.

If the held-out members complete follow-ups at the same rate, you have learned that cheaply. If the AI group does better and opt-outs stay flat, you have a result you can take to the board.

How Lorikeet approaches this

Lorikeet Dreamweaver increases customer lifetime value by giving every customer their own account manager. You set long-term goals, and every member gets their own Dreamweaver: an always-on account manager that works toward those goals over weeks and months, on every channel. Goals, not campaigns: you set the outcome, such as retention or activation, and you do not build a segment or a sequence.

A Dreamweaver wakes on a schedule and when something changes, and its next step is a message on the channel the member prefers, a task for your team, or no action at all. Most AI agents remember a conversation. Dreamweavers remember the whole person. Contact hours, limits and consent are checked before every message, every decision is logged with the reasoning behind it, and a holdout group is built in. Lorikeet is built for regulated industries including digital health, and Dreamweavers are built into the Lorikeet Concierge, so they share context with support conversations. Outreach runs on voice, SMS and email.

There is no separate charge. Dreamweavers are billed as touches and conversations at your existing rates (see pricing). If you are standing up the support side first, our guide to deploying AI support in weeks for digital health covers the rollout.

Getting started

Start with one stage from the table, usually follow-up booking or refill logistics, because the finish line is clear and the clinical boundary is easy to write down. Have your clinical lead write the handoff rules before anything is switched on. Run it against a holdout, and read the decisions, not just the dashboard.

If members are signing up and quietly drifting off before their second appointment, we should talk. You can get a demo, or try Dreamweaver with a 30-day free trial.