2care vs Aeva AI for Cliniko Clinics

If your clinic runs on Cliniko, Aeva AI and 2care will both come up, and both book into Cliniko: they read availability, book and reschedule, and recognise returning patients. On the happy path, in a quiet demo, they look similar, because almost every voice agent looks similar when the diary does not move and the caller is easy. The difference between them is not the demo. It is the engineering that decides what happens on the twenty per cent of calls where something is not as expected, and that is where a Cliniko practice should look before it chooses.

This is a technical comparison, on Cliniko, of how each agent is built and what that means when a call gets hard.

Aeva is a Cliniko-specific product, and it publishes that it handles up to five actions per call and around fifty concurrent calls. What it does not publish is how its voice pipeline is built, which is the thing that most determines how a call feels and how it fails.

2care runs a native voice pipeline, built in-house rather than assembled from third-party voice services. That is not a branding point; it is the reason for the numbers. Streaming speech recognition transcribes the caller as they talk, an intent layer resolves the reason, provider, urgency and identity from the transcript, and native text-to-speech speaks the reply, all inside a single latency budget of about 480 milliseconds, with the ninety-fifth percentile under 700 and the ninety-ninth under 1,100. Because the whole stack is owned, the read from Cliniko and the reasoning and the reply share that one budget instead of being relayed between separate systems that each add their own delay. An agent wrapped over a chain of external services inherits the latency and the outages of every link; a native one does not.

It is worth seeing that budget stage by stage, because 480 milliseconds is not one number, it is a chain of small ones held together. Streaming recognition returns a usable partial transcript in roughly 120 milliseconds rather than waiting for the caller to finish. The intent step runs while the caller is still speaking, so its cost overlaps speech they were going to make anyway. The model's first token arrives in around 180 milliseconds from a warm pool, and native text-to-speech begins audio in about another 120, with the remainder network and telephony. Each of those numbers is controllable only because the stage is owned; hand any one of them to a third-party service and it becomes a number you watch rather than one you set, which is the whole practical meaning of native.

Where the two differ, on the hard calls

Dimension, on your Cliniko line	2care	Aeva AI
Voice pipeline	Native, in-house, 480 ms reply	Cliniko-specific, custom voice
Turn-taking	Dual-signal, adaptive	Standard
Concurrent calls	1,000+	Around 50
Actions per call	No fixed cap	Up to 5
Languages	50+, multilingual agents	Local-accent set
Patient matching	Scores, escalates ties	Recognises returners
Clinical escalation	3 to 4 sec SLA	Configurable
Write safety	Idempotent, reversible	Standard
Compliance	HIPAA, GDPR, DPDP, BAA	Standard

Competitor figures are estimated from Aeva's public materials; 2care figures are our own.

Every row is a difference in what is engineered underneath, and each one is invisible on the happy path and decisive on the hard calls. On the numbers that decide how a Cliniko call actually feels, latency, turn-taking, concurrency, escalation, 2care leads each one.

Endpointing, barge-in and the calls that stall

The part of a voice agent that separates a good one from a frustrating one is turn-taking, and it is almost never in a demo. A naive agent treats a pause as the end of a turn and cuts a caller off; an elderly patient, or one reading out a Medicare number, pauses constantly. 2care uses dual-signal endpointing: it waits for both the line to go quiet and the sentence to read as complete, with an adaptive silence window of around 550 milliseconds that widens for hesitant or distressed callers. When a caller interrupts, the agent yields within about 200 milliseconds rather than talking over them.

Under load, the pipeline fails over between redundant native model instances before a caller hears a gap, so peak hour behaves like a quiet hour. These are the mechanisms that decide whether a busy Monday on a Cliniko line sounds calm or falls apart, and they are exactly the mechanisms a happy-path demo will never show you. Ask to hear a hesitant caller and a busy queue, not a clean booking.

Patient matching that refuses to guess

Recognising a returning patient sounds simple and is the single most dangerous place for an agent to be careless, because guessing wrong splits a treatment history across two Cliniko records and nobody notices for months. 2care does not match on a single identifier. It generates a small set of candidate records from the strongest key available, usually the phone number or date of birth, scores each on name similarity handled phonetically, date of birth and number, and then decides on two numbers, not one: the top candidate has to clear a confidence threshold, and the margin to the second candidate has to be wide enough. A high score with a close runner-up is not a match, it is two plausible people, and 2care escalates it to a person rather than writing under a guess. That calibration, strict on creating a record and permissive on recognising a person, is the difference between a clean Cliniko list and a slowly duplicating one.

The phonetic step is what lets it be permissive without being reckless. It maps a name to how it sounds rather than how it is spelled, so a nickname reaches the full name and a transliteration reaches its variant, without the loose character match that would also pull in genuinely different people who happen to spell alike. Most agents that claim to recognise a returning patient do the easy version of this, a lookup on a phone number, and a shared household or clinic number is exactly where the easy version quietly creates duplicates. The margin check is what a careful matcher has and a careless one does not.

Escalation with a boundary and an SLA

When a Cliniko call turns clinical, the question is what the agent does in the next two seconds. 2care resolves urgency on every call and routes it through an Escalation Engine with real targets: a routine call resolved on the line, a described emergency to a person in about three seconds, a clinical question to a nurse queue in about four, each carrying the transcript and the resolved patient so the clinician does not start from nothing. The agent never assesses or advises; it recognises that a description needs a human and hands it over, and every escalation records the trigger that raised it, so the boundary is auditable and tunable. The write itself is idempotent, so a retry after a network stutter never books a Cliniko session twice, and every write is logged and reversible. The detail behind the boundary is on the platform and the compliance page.

Compliance, and the agreement behind a booking

A booking that transcribes a call, resolves a patient and writes to Cliniko is creating and maintaining protected health information, which under HHS guidance makes the vendor a business associate rather than a conduit. 2care signs a BAA, encrypts patient data with AES-256 at rest and TLS in transit, never trains shared models on it, and covers UK and EU deployments under GDPR and Indian ones under DPDP with data held in region. For a clinic that takes patient data seriously, what a vendor signs and how it stores data is not a footnote, and it is a fair thing to ask of any agent on your Cliniko line.

Where 2care is right for a Cliniko practice

The question a Cliniko practice should ask is not which tool can book a call, both can, but how much the calls that are not simple matter to it. For a very small single site with light call volume, only the happy path may ever be exercised, and the depth described here rarely comes into play. 2care is built for the practice where it does.

It is the right choice for a clinic that cares about the failure paths: the busy morning that must not give a busy signal, so a thousand or more concurrent calls are answered at once; the returning patient who must not become a duplicate, so matching scores candidates and escalates an ambiguous one rather than guessing; the clinical call that must reach a person in seconds with context, so the Escalation Engine routes it on an SLA with the transcript attached; and patient data that must sit under a signed BAA and hold to more than one framework. The more a practice's real calls test those edges, the more clearly 2care is the agent engineered for them, on the same Cliniko line either tool would book a simple appointment into.

Frequently asked questions

Both book in Cliniko, so what actually differs?

The engineering on the hard calls. 2care's native pipeline replies in about 480 milliseconds, its endpointing does not cut off a hesitant caller, its matching escalates an ambiguous patient rather than guessing, and its escalation routes a clinical call to a person in seconds with SLAs. Those are the things that decide a busy or difficult call, and they are invisible in a happy-path demo.

How many concurrent calls can each take?

Aeva publishes around 50 concurrent calls; 2care handles 1000 or more with no busy signal. For a small single site both are ample; for a busy or multi-clinician practice, the ceiling is the difference between answering a Monday rush and dropping half of it.

How does 2care avoid duplicate Cliniko records?

It resolves the caller against the Cliniko record before writing, scoring candidates and requiring both a confidence threshold and a wide margin to the second candidate. A close call is escalated to a person rather than guessed, so a treatment history is not split across a duplicate.

What happens on a clinical call?

2care recognises it, leaves the booking flow, and routes it through the Escalation Engine: an emergency to a person in about three seconds, a clinical question to a nurse queue in about four, with the transcript attached. It never assesses or advises, and it logs the trigger so the boundary is auditable.

Is 2care compliant beyond Australia?

Yes. It signs a BAA in the US, encrypts data with AES-256 at rest and TLS in transit, and covers UK and EU deployments under GDPR and Indian ones under DPDP with data held in region, all without using patient data to train shared models.

The calls to judge it on

Do not judge either agent on a clean booking, because both pass that. Judge them on a hesitant elderly caller, a returning patient with an ambiguous record, a call that turns clinical, and a busy Monday queue. That is where the engineering shows, and it is where 2care is built to hold.

Hear 2care handle exactly those calls on your Cliniko line, live, when you book a demo, or read how the connection is built in the Cliniko guide.