Context
Miracell Homecare Limited delivers domiciliary care. The office team coordinates rotas, care visits and enquiries, and the telephone is the main front door. Calls arrive continuously through the working day and compete directly with care coordination.
The operational need
Unanswered calls had real consequences: potential new enquiries went elsewhere, families could not reach the office promptly, and staff absorbed long callback lists. Outside office hours nothing was answered at all. The requirement was not novelty. It was fewer lost conversations and a written record of what each caller needed.
What was implemented
An AI receptionist answers incoming calls, greets the caller, captures who they are, how to reach them and the nature of the enquiry, and logs the call so that the office team works from a structured record rather than memory or a scribbled note. Urgent matters are flagged for human follow-up rather than resolved automatically.
Where it sits in the workflow
The system sits in front of the office team rather than in place of it. Enquiry handling is a defined process: the call is captured, categorised and passed to a person for decision. The AI does not make care decisions and does not give clinical advice.
Administrative support, not care decisions
The system handles administrative contact only. It does not make care decisions, give clinical or nursing advice, assess eligibility, allocate care workers, or change a care plan. Those decisions stay with the registered manager and the care team. Its function is to capture information accurately and route it so that a person can decide what happens next, which is why the escalation route and the human review of every logged enquiry matter more than the call handling itself.
The implementation experience
Work focused on configuration rather than installation: writing the call flows around the questions the office team actually asks, diverting the line, testing with staff before go live, and revising wording after listening to real calls. The adjustments that mattered were small and practical, such as handling a caller who is upset before collecting their details.
Introducing AI into a regulated environment
- Human oversight and a defined escalation route for anything urgent or ambiguous.
- Confidentiality decisions taken before go live, including what is captured, what is retained and for how long.
- A clear statement to callers that they are speaking to an automated system and how to reach a person.
- Particular care with callers who may be vulnerable or distressed.
- Staff able to see every captured call, so automation never hides information from the people accountable for care.
Status and my role
This is a live deployment in operational use, not a pilot or a proof of concept. My role was founder and operator oversight: defining the requirement, deciding how it should behave in a regulated setting, and overseeing implementation and use within the business. I did not develop the underlying AI technology.
What is not claimed
No quantified administrative savings, return on investment, reduced staff time or improved care outcomes are claimed. The evidence here is a working deployment in one business, described honestly, and its value lies in the implementation experience rather than in measured financial benefit.
Lessons
- The failure path matters more than the demonstration. What happens when the caller is distressed or the request falls outside the script decides whether the system is safe.
- Staff trust decides adoption. If automation creates extra work for the office, it gets switched off.
- Call logging becomes governance evidence, not just administration.
- Confidentiality and escalation are design decisions, not afterthoughts.