AI Phone Calls in Aged Care: An Honest Guide for Boards and Executives

Written by a vendor, deliberately against interest. If this guide makes you harder to sell to, it has worked.

Start with what AI calls cannot do

• • They cannot replace human relationship. The consistent position of ethicists, peak bodies, and the major faith-based frameworks is that AI in care must augment the human relationship, never substitute for it. Any vendor promising an "AI companion" that makes visits unnecessary is selling against the settled view of the sector.
• • They cannot make clinical judgements. A call can notice that a person sounds flat, mentioned dizziness, and skipped lunch. Deciding what that means and what to do is a human's job, and must stay one.
• • They cannot be the last line of defence. If the entire safety case rests on software never failing, there is no safety case. The design question is always: what happens when the call does not connect, and who notices?

Where the case for them is real

Coverage and documentation. No workforce in Australian aged care can affordably hold a warm fifteen-minute conversation with every client every day, write it up, and act on what was heard. Software can hold the conversation and produce the record; humans keep the judgement and the follow-through. The result is not fewer humans in care; it is human attention spent where a conversation found something, rather than spread thinly across everyone by roster. The days between visits stop being silent.

The envelope that makes it defensible

• • Informed, revocable consent. The person knows what the service is, agrees to it, and can stop it with a sentence.
• • Disclosure on every call. Nobody should ever be unsure whether they spoke to a person.
• • Human review of everything that matters. Concerning conversations reach a qualified person within a defined service level, with receipt confirmed.
• • Data boundaries in writing. Where conversations are stored, who can access them, and a contractual commitment that they are not used to train third-party models.
• • Low-risk scope. Wellbeing conversation and social connection, not diagnosis, medication advice, or crisis response.

External reference points exist and are worth using: the Rome Call for AI Ethics, which providers can ask their vendors to sign, and the transparency and dignity principles in the strengthened Quality Standards. A vendor who welcomes being held to a published ethical framework is telling you something; so is one who changes the subject.

Ten questions for any vendor

• • 1. What exactly does the AI decide on its own, and what does it hand to a human?
• • 2. Walk me through your worst failure to date and what changed afterwards.
• • 3. What happens when a call is unanswered? Show me the trail.
• • 4. How do you obtain and record consent, and how does a person opt out?
• • 5. Where is our data stored, who can see it, and is it used for model training?
• • 6. Which published ethics framework are you accountable to, and can we audit against it?
• • 7. What is your escalation service level, in writing, with what remedy?
• • 8. What do the older people themselves say? Show unfiltered feedback.
• • 9. What happens to our records and continuity if we leave you?
• • 10. What will you not build, even if we ask?

Frequently asked questions

Is any of this required by law?
The Aged Care Act 2024 and the strengthened Quality Standards require dignity, transparency, and evidence of monitoring and response; they do not prescribe or prohibit AI. The envelope above is how providers adopt the technology inside those obligations.

Should we pilot before committing?
Yes, and structure it: a defined cohort, consent obtained properly, success measures agreed in advance, and an evaluation report you would be comfortable showing your regulator.

Where does careplans AI sit in all this?
We provide wellbeing check-in calls with human escalation, built inside the envelope described here; see how the service works and our trust commitments. This guide is the standard we expect to be held to.