Medication adherence AI that finds out why, not just whether.
OneDose tracks whether patients are actually taking their medication across an entire panel. It reminds on a schedule built from the real regimen, detects drift across weeks rather than reacting to a single missed dose, asks the patient what went wrong, and routes the clinical reasons to a pharmacist with the history attached.
Whether a patient took a dose is one fact. Why they stopped is the fact that determines what to do about it, and it is the one nobody collects.
Why is adherence so hard to move?
Because it is not one problem. “The patient is non-adherent” describes an outcome, not a cause, and the causes need different responses — which is why interventions that treat it as a single problem plateau early.
It also decays rather than failing. Day-one adherence is usually fine: the instructions are fresh, the clinician is recent, and the reason for the medication is emotionally vivid. All three fade, and what remains is a daily task with no feedback. An antihypertensive does not feel like anything, so taking it produces no sense of benefit and skipping it produces no sense of harm. Unreinforced behaviours decay.
Why do patients actually stop?
Four reasons, and the response to each is different. Treating them as one — as “forgetting” — is the single most common mistake in this category.
They forgot
The only reason a reminder solves. It is real, and it is the one vendors like because it has a software fix — which is also why reminder-only products show early promise and then flatten.
A side-effect
The patient felt worse taking it than not and made an entirely rational trade-off. The fix is clinical: a different drug, dose or time of day. A reminder here actively annoys someone into resenting the channel.
Cost
They could not afford the refill. Patients under-report this systematically, and they under-report it more to software than to a person who is kind about it. The fix is not clinical at all.
They decided it was unnecessary
They felt fine, so they concluded they did not need it — a reasonable inference for an asymptomatic condition, from the evidence available to them. The fix is understanding, and they will not volunteer this unless asked without judgement.
Only the first is a reminder problem. The other three need a person, a different drug, or a different conversation — and none of them get one unless somebody asks the patient a question and records the answer as data rather than as a note.
What does OneDose do differently?
It watches for drift rather than absence. One missed dose is noise. A pattern of missed evening doses, a refill collected four days late, and a patient who has stopped answering is a signal — and it is visible weeks before that patient would show up on any report as non-adherent.
It asks why, and treats the answer as a structured field rather than a sentence in a call note. An intervention that cannot distinguish a cost problem from a side-effect will apply the wrong fix to both.
And it escalates the clinical ones. A side-effect report is a pharmacist conversation, and it is frequently the most valuable call of the month — it is the moment a patient is about to stop, and the only moment anyone can do something about it. An agent that logs it and moves on has wasted the best thing it learned.
Where is automation the wrong answer?
Where the relationship is the intervention. On a small, complex, high-touch panel, a patient who tells a trusted pharmacist the truth about why they stopped is a mechanism that is already working, and replacing that contact with an agent removes it.
Where non-adherence is mostly psychosocial. If patients are stopping because of money, housing or a family situation, they need a human who can act on that — and an agent’s correct move is to escalate, which means a human does it anyway.
And where nobody can act on what you find. Detecting drift across a whole panel produces escalations. An organisation with no one to receive them has bought a better view of a problem it still cannot fix.
Frequently asked
- What is medication adherence AI?
- Medication adherence AI is software that tracks whether patients are actually taking their medication and intervenes when they are not — reminding on a schedule built from the specific regimen, detecting patterns of missed doses or late refills, asking the patient why, and escalating to a pharmacist when the reason is clinical or the patient appears to be stopping.
- Why do reminder apps stop working?
- Because forgetting is only one of four reasons patients stop, and it is the minority one. Side-effects, cost, and a considered decision that the medication is unnecessary all need different responses. A reminder sent to someone who stopped because of a side-effect is not neutral — it is pinging them to do the thing that makes them feel unwell.
- How does OneDose know if a patient took their dose?
- Primarily by asking, and by watching refill and dispensing behaviour where the deployment has access to it. Self-reported adherence has known limits and OneDose treats it as a signal rather than proof — which is why escalation triggers on a pattern across weeks rather than on a single answer.
- What adherence improvement does OneDose deliver?
- OneDose does not publish an adherence figure it cannot attribute to a named deployment, so there is no number to quote here honestly. Published adherence lifts also vary enormously by condition, population and how adherence was measured — a figure from a statin cohort tells you little about an oncology one, which is worth knowing when any vendor quotes you one.
- Does the AI change a patient’s medication?
- No. Any suggestion that a dose, a drug or a schedule should change goes to a pharmacist or prescriber. The agent captures the report, routes it to a human, and tells the patient a human is coming.