Behind the scenes: how Robin AI handles a real emergency call

2:47 AM, Tuesday, 16 January 2025

The phone rings at Oakwood Care's emergency line. No human is awake to answer - but Robin is.

What happens in the next 8 minutes demonstrates why AI in homecare isn't just about efficiency - it's about saving lives.

This is a real call (client details anonymized). Let's step through it, second by second.

---

⏱️ 02:47:03 - Call connects

Robin: "Hello, you've reached Oakwood Care's emergency line. This is Robin, your AI assistant. I'm here to help 24/7. May I have your name and the client you're calling about?"

Caller (distressed, speaking quickly): "It's Emma Johnson-my mum, Mary Thompson, I think she's fallen. Her alarm went off and I-I don't know if I should call 999 or-"

What's happening behind the scenes:

Robin's AI is processing multiple data streams simultaneously:

1. Voice analysis:

   • Tone: Anxious, elevated speech rate
   • Keywords detected: "fallen," "alarm," "999"
   • Urgency score: 8/10 (high priority)

2. Client database lookup:

   • Name: Mary Thompson
   • Age: 82
   • Address: 14 Willow Lane, Bristol BS4 2QQ
   • Medical conditions: Osteoporosis, diabetes
   • Emergency contacts: Emma Johnson (daughter), Dr. Sarah Patel (GP)
   • Last care visit: 21:30 (5 hours ago)

3. Decision tree activated:

   • Fall + elderly + medical conditions = Potential emergency
   • Protocol: Assess severity - Dispatch carer/ambulance - Notify family/on-call

Time elapsed: 12 seconds

---

⏱️ 02:47:15 - Information gathering

Robin: "Emma, I understand the alarm company contacted you about Mary. I'm pulling up her details now. Are you with Mary at the moment?"

Emma: "No, I'm 20 minutes away. The alarm company called me. I don't know if she's hurt, if she can get up-"

Robin: "I understand this is frightening. I'm going to help you. First, do you know if Mary is conscious?"

Emma: "I-I don't know. Can you send someone? Should I call an ambulance?"

Behind the scenes:

Robin assesses:

• Caller is not on-site (limits immediate information)
• No confirmation of consciousness (escalates priority)
• Caller asking for guidance (needs clear instructions)

Robin's decision:

1. Alert care team immediately (visual assessment needed)
2. Simultaneously notify on-call manager
3. Provide caller with actionable steps
4. Keep 999 option ready but avoid unnecessary ambulance dispatch

Care team alert sent:

• Emergency email dispatched to on-call team (02:47:42)
• On-call manager David Chen notified
• Details include: Client name, location, medical history, urgency level

Time elapsed: 38 seconds

---

⏱️ 02:47:41 - Dual-track response

Robin: "Emma, I've just alerted our care team about Mary's situation. Our on-call coordinator is being notified right now and will dispatch a carer to Mary's home immediately. While we wait, I want you to try calling Mary's mobile-can you do that for me?"

Emma: "Okay, yes, I'll try now."

Robin: "Stay on the line with me. I'll wait while you call."

Behind the scenes:

Parallel actions in progress:

1. On-call manager (David Chen):

   • Emergency email received (02:47:42)
   • Reads situation details (02:49:08)
   • Dispatches carer Jessica via SMS (02:49:33)

2. Carer Jessica:

   • SMS received from David: "URGENT: Client fall. Mary Thompson, 14 Willow Lane. Please respond."
   • Response: "On my way, ETA 6 mins" (02:49:47)

3. Emma attempts to call Mary:

   • Result: No answer (3 rings, voicemail)
   • Reports back to Robin (02:48:31)

Time elapsed: 1 minute 28 seconds

---

⏱️ 02:48:31 - Escalation decision point

Emma: "She's not answering her phone. Oh God, what if-"

Robin: "Emma, listen carefully. Our care team is being dispatched to Mary now. Because Mary isn't answering, I'm going to recommend we also alert the ambulance service as a precaution. Would you like me to guide you through calling 999, or would you prefer to do that yourself?"

Emma: "Can you do it? I need to drive there now."

Robin: "Yes, I'll make sure 999 is contacted immediately and I'll pass on all of Mary's details. You start heading to Mary's home and drive safely. Our carer will arrive and will call you the moment they're with Mary. Stay on the line if you want updates, or you can hang up and focus on driving."

Emma: "I'll drive. Thank you. Please just-please make sure someone gets there."

Robin: "Our team is responding right now, I promise. Drive carefully, Emma."

Behind the scenes:

Robin's escalation logic:

• No response from client after fall alert = Medical emergency protocol
• Care team alerted (visual assessment needed)
• 999 guidance provided (paramedic backup)
• Family informed and en route

999 call coordinated:

• David (on-call manager) contacts 999 immediately
• Provides complete details from Robin's email:
  • Address: 14 Willow Lane, Bristol BS4 2QQ
  • Nature: Elderly fall, non-responsive to phone
  • Medical history: Osteoporosis, diabetes
  • Carer en route to scene

Ambulance dispatch confirms: Non-urgent ambulance (Category 3) dispatched, ETA 18 minutes.

Time elapsed: 2 minutes 51 seconds

---

⏱️ 02:50:19 - Carer arrives

Jessica texts David: "On scene with Mary now"

Behind the scenes:

David logs Jessica's text update in the system:

• Care team notified of on-scene status
• Status update shared with Robin for caller follow-up

Time elapsed: 3 minutes 16 seconds

---

⏱️ 02:52:04 - Situation assessment

Jessica (calling Robin): "Robin, it's Jessica. I'm with Mary. She's conscious, sitting on the floor next to her bed. Says she tried to get up to use the loo, lost her balance. No visible injuries, but she's shaken and can't get up on her own."

Robin: "Thank you, Jessica. Can Mary answer questions coherently?"

Jessica: "Yes, she's alert. A bit embarrassed, to be honest."

Robin: "Understood. Our care coordinator has contacted 999 as a precaution given the fall and her osteoporosis. Ambulance ETA is 15 minutes. Can you stay with her until they arrive?"

Jessica: "Absolutely. I'll get her a blanket and some water."

Robin: "Perfect. I'll notify David immediately so he can update Emma."

Behind the scenes:

Robin logs Jessica's assessment in the system:

• Client status: Conscious, alert, no immediate danger
• Carer on-scene: Staying until ambulance arrives
• Alert sent to David: "Status update from Jessica - Mary conscious and alert"

David calls Emma (02:52:31):

David: "Emma, good news. Jessica is with your mum. Mary is conscious and alert - she fell trying to get to the bathroom but has no visible injuries. Jessica is staying with her, and an ambulance is on the way as a precaution."

Emma (relieved, tearful): "Oh thank God. Can I talk to her?"

David: "Jessica is with Mary now. I'll have her call you directly in just a moment. You can meet them at the house or at the hospital if Mary is taken in for assessment."

Emma: "I'm 10 minutes away. I'll go straight there. Thank you so much."

Time elapsed: 5 minutes 28 seconds

---

📋 What the care team received: real-time alert, zero admin

Within 35 seconds of Emma's call, the Oakwood Care team received this email:

---

FROM: Robin - your Oakwood Care Assistant robin@oakwoodcare.co.uk TO: oncall@oakwoodcare.co.uk, coordinator@oakwoodcare.co.uk SUBJECT: EMERGENCY message from Emma Johnson TIME: 02:47:42

---

Dear Care Team,

We have received an EMERGENCY message from Emma Johnson. Here are the details for you:

• Urgency level: CRITICAL
• Caller name: Emma Johnson
• Caller phone: 07700 900123
• Caller email: emma.j@email.com
• Post code: BS4 2QQ

Message:

[EMERGENCY] Mother Mary Thompson (82) has fallen at home. Personal alarm company alerted daughter. Mary not answering phone. Location: 14 Willow Lane, Bristol BS4 2QQ. Medical history: Osteoporosis, diabetes. Daughter Emma is 20 minutes away. [999 status: caller advised to call immediately]

Caller called from this number: +447700900123

Please prioritise this matter accordingly.

Best regards,

Robin - your Oakwood Care Assistant Oakwood Care Ltd. +44.XXXX.XXXXXX robin@oakwoodcare.co.uk https://oakwoodcare.co.uk

---

Everything David needed in one email:

✅ Urgency clearly flagged: "CRITICAL" in the subject line ✅ Caller contact details: Phone number for immediate callback ✅ Full situation context: Fall, medical history, 999 status ✅ Location: Complete address for carer dispatch ✅ Action guidance: "Please prioritise this matter accordingly"

Plus, available to the care team:

• 🎙️ Call recording: Secure link in the system (2 min 58 sec)
• 📝 Call transcript: Full conversation, searchable
• 📊 Client record: Direct link to Mary's care file

Total time for David to understand the situation and take action: 20 seconds

No phone tag. No voicemail. No missing information. Just clear, actionable intelligence delivered instantly.

---

🗂️ How it's logged for CQC compliance

Every call Robin handles is automatically documented:

Email record stored:

• Sent timestamp: 02:47:42
• Recipients: On-call team and care coordinator
• Urgency classification: CRITICAL
• Complete caller details and situation summary
• 999 status tracked

Call data logged:

• Call start/end times (02:47:03 - 02:48:01)
• Caller identity (Emma Johnson, relationship: daughter)
• Client referenced (Mary Thompson, ID: 4782)
• Keywords detected ("fallen", "not answering phone", "osteoporosis")
• Actions taken (999 advised, care team alerted)
• Full audio recording (stored securely for 7 years)
• Complete transcript (searchable by keyword)

Staff response tracked:

• Email opened by David Chen at 02:49:08
• Jessica dispatched at 02:49:33
• Carer arrived on scene at 02:54:47
• Outcome recorded: Client safe, ambulance attended

For CQC inspections:

When an inspector asks "Show me how you handled emergency calls in January", the care coordinator can:

1. Search by urgency: Pull all "CRITICAL" priority emails
2. Filter by date range: January 1-31, 2025
3. Review response times: Call received → carer dispatched → resolution
4. Access complete audit trail: Email, recording, transcript, outcome

Example search results:

• 3 CRITICAL emergencies in January
• Average response time: 8.2 minutes (call to carer on-site)
• 100% had care team response
• 100% documented with full audit trail

Documentation time required from care staff: 0 minutes

Robin captures everything automatically while the team focuses on what matters: responding to the emergency.

For a complete guide to using AI audit trails for CQC inspections, see our article on CQC compliance documentation.

---

🧠 What just happened? (the AI breakdown)

1. Natural language processing (NLP)

Robin understood:

• Emotional distress in caller's voice
• Key urgency indicators ("fallen," "alarm," "999")
• Context clues (time of night, caller's location)

2. Multi-system integration

Robin accessed:

• Client database (medical history, contacts)
• Care team notification system (email alerts to on-call staff)
• Emergency escalation protocols
• Call recording and documentation system

3. Decision-tree logic

Robin followed protocols:

• Assess severity - Dispatch resources - Inform stakeholders - Document

But adapted in real-time:

• Client not answering phone - Escalate to 999
• Carer available nearby - Dispatch before ambulance (faster visual assessment)

4. Human-in-the-loop safety

Robin didn't replace humans-it coordinated them:

• Carer provided on-ground assessment
• Paramedics made medical decisions
• Family stayed informed

5. Automatic documentation

Zero admin burden for the care team.

The email alert included:

• Complete situation summary
• All caller details and contact information
• Medical history flagged
• 999 status tracked
• Links to call recording and transcript

David didn't need to write notes at 3 AM. Everything was captured automatically.

This emergency response is just one example from Soma Healthcare's experience. Read the full case study showing 1,500+ calls handled.

---

❓ What if Robin hadn't been there?

Let's replay this scenario with a traditional setup:

02:47 AM - Phone rings

• Voicemail picks up
• Emma leaves panicked message
• On-call manager asleep, phone on silent

03:15 AM - Manager wakes up, checks phone

• 28 minutes later
• Listens to voicemail, calls Emma (no answer-she's driving)
• Calls Mary (no answer)
• Tries to reach nearest carer (Jessica's off-duty, doesn't answer)

03:35 AM - Manager calls 999 directly

• Ambulance dispatched
• Emma arrives at Mary's house, finds door locked, panics
• Ambulance arrives (03:52 AM)
• Paramedics break door lock to access Mary

Total time: 1 hour 5 minutes (vs. Robin's coordinated 8-minute response)

Additional costs:

• Broken door: £200
• Unnecessary ambulance dispatch (could've been assessed by carer first): £250
• Family trauma from 1-hour uncertainty: Priceless

Calculate your own agency's risk exposure from missed emergency calls: cost of missed calls analysis.

---

🎯 The key takeaway

Robin isn't magic. It's protocol automation done right.

Every step Robin took - alert care team, coordinate 999 response, notify manager, document everything - is what a perfect human coordinator would do.

The difference?

• Robin does it in seconds, not minutes
• Robin never sleeps, panics, or forgets steps
• Robin documents everything automatically

But here's the crucial part: Robin knew when to involve humans.

• Carer assessed the client in person
• Paramedics made medical decisions
• Family provided emotional support

AI didn't replace the care team-it made them faster, more coordinated, and better informed.

---

💬 What Emma said later

In a follow-up satisfaction call:

"I was terrified. It was the middle of the night, I was 20 minutes away, and I didn't know if my mum was lying there hurt. But Robin didn't just take a message-it felt like someone was handling it. By the time I got there, Jessica was already with Mum, the ambulance had checked her, and everyone knew what was happening. I can't imagine what that hour would've been like if I'd just left a voicemail and waited."

Net Promoter Score (likelihood to recommend Oakwood Care): 10/10

---

🚀 Ready to see Robin in action?

Every homecare agency will face emergency calls. The question is: how fast will you respond?

Or explore more AI use cases:

• How Soma Healthcare cut calls by 70% →
• The real cost of missed calls in homecare →