
An answering service, an in-house on-call rota and an AI voice agent can all pick up your phone at 11pm. They behave very differently when the call is hard. This article compares them on the tests that matter to an Irish home care or hospice team, including the places where each one is weaker than the others. It does not name suppliers, and it does not quote prices, because we have not found reliable public comparisons.
If you have not yet mapped your options, start with our overview of five ways to cover your phone after hours. This piece goes deeper on three of them.
Answering service, rota or AI voice: what you are comparing
An answering service is a call centre that answers in your agency's name, takes a message and sends it on. An in-house on-call rota is your own nurse, coordinator or manager carrying a phone. An AI voice agent is software that speaks with the caller, records what is said and routes the call by rules you set.
Each can be run alone or in a mix. The right choice depends on your risk, your size and how much sleep you want your senior people to keep.
Nine tests
1. Speed and capacity
Does every call get answered at once, even when several arrive together? A person answers one call at a time. A rota has one phone. An AI voice agent can take many calls in parallel. That matters on a bad night, when one fall report and two worried families call within minutes.
2. Knowledge of the client
Who does the caller get? The on-call nurse may know the client. An answering service operator usually does not. An AI voice agent knows what you load into it, which might be a client list, a rota and your rules, but it knows nothing beyond that. Decide how much each caller needs to repeat.
3. Handling distress
This is where people are strongest. A calm, warm human voice can settle a frightened family. An AI voice can be courteous and patient, but it can misread distress and can sound flat at the wrong moment. For hospice and end-of-life calls, put a human close behind the first layer. Our article on hospice out of hours calls explains why.
4. Judgement
Judgement belongs to your clinicians. No answering service operator and no AI voice agent should give clinical advice. In an Irish survey of out-of-hours palliative care phone advice, 73% of staff giving advice had no specific training in doing so by phone. That shows how hard it is even for clinicians. Design the first layer so it never has to judge.
5. Live transfer and escalation
What happens with an urgent call? The best answer is a live transfer to a named person, then a second person, then a third, with the call logged. A weaker answer is a text message that sits unread. Ask each supplier to demonstrate its chain, including what it does when nobody picks up. Our guide to on-call escalation rules shows how to define the chain.
6. Quality of the record
A record that your office can read at 8am and act on without ringing anyone is worth more than a good call that leaves no trace. Look at a sample: Is the caller named? Is the client named? Is the issue clear? Is the action recorded? Are quotes in the caller's words?
7. Data protection
All three handle health data. Your agency is the controller, and any supplier is a processor. Article 28 of the GDPR requires a binding contract that sets out the processing and requires the processor to act only on your documented instructions. An AI voice agent adds the AI Act duty to tell callers they are speaking to an AI system. Our detailed guide to AI voice and GDPR covers both.
8. Staff impact
An on-call rota puts load on your own people. The Department of Health's Strategic Workforce Advisory Group report records stakeholder concerns about unsociable hours, working alone and burnout among care workers. If your on-call nurse is also your best nurse, protect them. A first layer that filters routine calls can help.
9. Failure mode
Ask how each option fails. A rota fails when a phone is off or a person is asleep. An answering service fails when messages are not read. An AI voice agent fails when it mishears or meets a situation outside its rules. The best systems fail safely: when unsure, they hand the call to a person, and they say so in the record.
The comparison at a glance
| Test | Answering service | In-house on-call | AI voice agent |
|---|---|---|---|
| Speed and capacity | Good, limited by operators | One call at a time | Many calls at once |
| Knowledge of the client | Low unless scripted | High | Only what you load in |
| Handling distress | Warm, varies by operator | Strongest | Needs a human close behind |
| Judgement | None, should not give advice | Clinical, if a nurse | None, must not give advice |
| Live transfer | Possible, must be agreed | Direct | Possible, must be agreed |
| Quality of record | Varies, often a short message | Depends on the person | Consistent, with a summary and recording |
| Staff impact | Low | High | Low |
| Typical failure | Message waits for morning | Phone off, person asleep | Mishearing, unusual situations |
Read the table as a starting point. Your own tests will matter more.
Pick the option whose failure you can live with, then add a second layer for the failure that remains.
When each one makes sense
An answering service fits when your call volume is modest, most calls are routine messages and a human voice matters to your clients. Make sure live transfer for urgent matters is part of the contract.
An in-house rota fits when you are small, your nurses are willing and the call volume is low. Build in a backup, and watch for fatigue.
An AI voice agent fits when you need every call answered, recorded and routed the same way at any hour, and you have people ready to take the urgent ones. Hettie is designed for that role. It keeps four escalation levels, sends urgent and emergency calls to a person at once and leaves clinical decisions to your team.
A mix fits most agencies as they grow. Use the first layer for everything, and a named human for what the rules escalate.
Contract points to insist on
Whichever option you choose from outside your team, put these in writing.
- Service description. Hours, call types handled, what is escalated and how.
- Escalation chain. Names or roles, order, timing and what happens when no one answers.
- Records. What is kept, where, for how long, and how you get it.
- Data protection. An Article 28 agreement, a list of sub-processors and the location of data.
- Testing. The right to run test calls and to audit the results.
- Change control. Who can change the rules, and how changes are approved and logged.
- Exit. How you get your data and your numbers back if you leave.
Two scenarios to test against
Imagine two nights and ask how each option copes.
In the first, a daughter rings at 10.40pm to say her father has not been out of his chair since the evening carer left, and he sounds confused on the phone. The call needs to reach a nurse quickly, with the facts already written down, and the daughter needs to hear what will happen next.
In the second, three families, a carer and a hospital ward ring within the same twenty minutes after a storm has caused delays. The call needs to be answered every time, sorted by your rules and summarised for the morning, with the urgent ones reaching a person first.
A rota struggles with the second night. A message-taking service may struggle with the first. A first layer with a live human behind it is built for both. Write your own two scenarios, based on your worst recent nights, and use them in your trial.
Questions to ask on any demo
- Show me an urgent call end to end, from ring to named person.
- What do you say to a caller who asks whether to call an ambulance?
- What happens when the first person does not answer?
- Show me last night's records from a real account, with names hidden.
- Who can listen to recordings, and is access logged?
- How do I change the rules, and who approves the change?
- What happens when your service is down?
Run a two-week trial
Do not decide on a demo alone. Run a short trial with test calls at awkward hours, one for each situation you fear. Score each on speed, accuracy of the record and the route the call took. Ask two members of staff who were not involved in the purchase to review the results. If you are also preparing for regulation, keep the results. They are exactly the kind of evidence described in our piece on the Home Support Providers Act 2026, and a record of how you tested your cover shows an inspector that you take it seriously.
For details on pricing structure and what is included, see our pricing page.
Sources
- Coffey S et al.: Specialist palliative care out of hours telephone advice in Ireland, a national survey (BMJ Supportive & Palliative Care, 2024)
- GDPR Article 28: Processor
- Department of Health: Report of the Strategic Workforce Advisory Group on Home Carers and Nursing Home Healthcare Assistants (September 2022)
- EU AI Act, Article 50: Transparency obligations
Frequently asked questions
Is an AI voice agent cheaper than an answering service?
We have not found reliable public price comparisons for Irish providers, so we will not claim it. Compare total cost, including the cost of fatigue in your on-call staff and the cost of a missed urgent call, as well as the monthly fee.
Can we keep our current answering service and add an AI voice agent?
Yes, in a mixed model. Decide which layer answers first, what it passes to the second layer and how the handover is recorded. Test the handover with real calls.
Does an AI voice agent have to say it is a machine?
Yes. Article 50 of the EU AI Act requires that people are informed when they interact with an AI system, unless it is obvious from context. Check that your supplier does this on every call.


