An estate agent in Bristol was taking 40 inbound calls a week. Eleven went unanswered — school runs, viewings, lunch. Average callback time: 3.5 hours. Their competitor across the road had an automated SMS going out within 90 seconds and an AI voice agent calling back within four minutes. The Bristol agent's conversion rate on inbound enquiries was 19%. The competitor's was 31%.
The same budget, the same market, the same number of missed calls. The only difference was whether those missed calls triggered a recovery sequence or sat in a voicemail queue until someone had time to look.
This is how to build the recovery sequence.
The maths of the missed call: what it actually costs
MIT research published in the Harvard Business Review found that leads contacted within five minutes of enquiring are 21× more likely to convert than leads contacted after 30 minutes. The same pattern holds for inbound calls: the lead who rang you is at peak intent at the moment of the call. Every minute after that, intent decays.
For a business missing 10 calls a week at an average deal value of £2,000 and a 20% conversion rate on recovered leads, that's £4,000/month in recoverable revenue — before any spend on new lead generation. Missed-call recovery is one of the highest-ROI automation investments an SME can make because the leads are already warm; you're just being faster than your competitors at reaching them.
The three failure modes: (1) no callback at all — the voicemail sits unlistened; (2) a callback 3+ hours later when the lead has called three competitors; (3) a generic automated SMS with no personalisation that gets ignored or triggers an unsubscribe.
The three-layer recovery sequence
The pattern that consistently outperforms alternatives across our UK SME deployments:
| Layer | Timing | Channel | Purpose |
|---|---|---|---|
| Layer 1 — Immediate SMS | ≤90 seconds | SMS | Acknowledge the missed call, set expectation |
| Layer 2 — AI voice callback | 3–5 minutes | Outbound voice | Re-engage, qualify, book or transfer |
| Layer 3 — Email follow-up | 30–60 minutes | Backup if SMS/call not answered |
Each layer is conditional on the previous: Layer 2 only fires if Layer 1 got no response within 3 minutes; Layer 3 only fires if Layers 1 and 2 produced no conversion. The full sequence is self-cancelling — if the lead responds at any layer, the remaining layers stop.
The SMS text for Layer 1:
Hi [Name], sorry we missed your call — we'll call you back in the next few minutes. If it's easier, reply here. STOP to opt out.
Short. Personal where possible. Sets a specific expectation. Includes an opt-out. No exclamation marks.
Building the sequence in n8n or Make
The flow starts with a Twilio webhook on missed-call events. Twilio fires a webhook to your n8n or Make instance when a call ends with CallStatus=no-answer or CallStatus=busy. From there:
{
"trigger": "twilio_webhook",
"condition": "CallStatus IN ['no-answer', 'busy']",
"steps": [
{
"id": "crm_lookup",
"type": "http_request",
"url": "https://api.hubspot.com/crm/v3/objects/contacts/search",
"method": "POST",
"body": {"filterGroups": [{"filters": [{"propertyName": "phone", "operator": "EQ", "value": "{{caller_number}}"}]}]}
},
{
"id": "send_sms_layer1",
"type": "twilio_sms",
"to": "{{caller_number}}",
"body": "Hi {{contact_first_name | 'there'}}, sorry we missed your call — we'll call you back shortly. Reply here or STOP to opt out.",
"delay_seconds": 0
},
{
"id": "wait_for_response",
"type": "wait",
"duration_seconds": 180,
"cancel_on": "sms_inbound_from_{{caller_number}}"
},
{
"id": "voice_callback_layer2",
"type": "retell_outbound_call",
"to": "{{caller_number}}",
"agent_id": "{{recovery_agent_id}}",
"dynamic_variables": {"lead_name": "{{contact_first_name}}", "enquiry_context": "{{last_deal_stage}}"}
}
]
}
The crm_lookup step runs first to personalise the SMS and the voice callback. If the contact isn't in the CRM, the flow falls back to "there" for the name field and uses a generic recovery agent script.
The wait_for_response node is the key architectural piece: it holds the flow for 3 minutes, listening for an inbound SMS reply. If a reply arrives (any reply — not just specific keywords), the voice callback is cancelled. If no reply, the voice agent calls.
The voice callback agent: what it says and what it doesn't
The recovery voice agent has one job: acknowledge, qualify briefly, and convert to a booked slot or transfer to a human. It is not a sales agent. It does not pitch. It does not try to close in a two-minute callback. The framing matters: this is a service call following up on a contact the lead initiated, not a cold outbound call. Design the agent's persona and script to reflect that distinction — recovery agents that sound like outbound sales agents produce complaint rates 3× higher than those that sound like a responsive business calling back.
The opening: "Hi [Name], this is Aria calling back from [Company] — you tried to reach us a few minutes ago. Have you got 60 seconds? I wanted to make sure we didn't miss anything important."
This framing — "I wanted to make sure we didn't miss anything important" — performs significantly better than any sales opener in the recovery context because the caller already initiated contact. You're following up on their intent, not creating a new one.
The agent's decision tree is deliberately short: acknowledge the original enquiry, ask one qualifying question (is this still something you're looking into?), then either (a) book a slot with a specific team member using a calendar integration, or (b) transfer live to a human if one is available. Three outcomes maximum: booked, transferred, or "not right now — I'll follow up by email."
See Inbound Lead Routing with AI for the broader architecture of handling all inbound channels, not just missed calls, with a unified routing layer.
Compliance: PECR consent and timing rules
Three PECR points that apply specifically to automated missed-call recovery:
Legitimate interest as the basis for the SMS. An inbound caller has expressed intent to contact you; following up on that contact within a reasonable time window is defensible under legitimate interest, provided you conduct a legitimate interest assessment and document it. The ICO's direct marketing guidance is the authoritative reference.
Opt-out must be honoured. Every automated SMS must include a STOP mechanism. If a contact replies STOP, they must not receive further automated outreach. Wire this into your CRM as a flag — not just a list in your SMS tool — so it's enforced across all channels.
Timing of the automated callback. An AI voice callback at 7pm on a Sunday will annoy rather than convert, and may constitute a breach of PECR's requirement that direct marketing be conducted at reasonable times. Restrict your automated callbacks to 9am–6pm on weekdays and 9am–1pm on Saturdays. Wire this as a schedule condition in your n8n or Make flow before the Retell outbound call node fires.
What changed in 2025–2026: AI voice agents in recovery flows
Until 2024, the standard pattern was SMS + human callback. The constraint was that AI voice agents weren't reliable enough for the recovery context — they struggled with ambiguous intents, non-standard UK accents, and the open-ended nature of a callback where the lead's original enquiry might be anything.
That changed with the generation of LLM-backed agents that can handle open intents gracefully. A recovery agent that can say "What were you looking to discuss?" and then route the response to the correct qualification path — mortgage, insurance, IT support, estate agency, whatever the business handles — is now buildable with off-the-shelf components. Retell AI's multi-intent routing and VAPI's conditional branching both support this pattern without custom code.
The conversion difference: AI voice callback versus no voice callback in our deployments adds 8–12 percentage points to recovery rate on missed calls that don't respond to the initial SMS. The calls that would have been lost to voicemail backlog are being re-engaged.
Measuring recovery success: the metrics that matter
Most teams measure missed-call recovery by "callback completed." That's the wrong primary metric — completing a callback that doesn't convert doesn't recover the lead.
The metrics worth tracking:
| Metric | How to measure | Target |
|---|---|---|
| Recovery rate | (leads re-engaged within 24h) / (total missed calls) | >50% |
| Recovered conversion rate | (deals closed from recovered leads) / (recovered leads) | ≥ baseline inbound conversion |
| Layer 1 response rate | SMS replies / SMS sent | >15% |
| Layer 2 completion rate | Voice callbacks completed / calls initiated | >60% |
| Time to first contact | Median minutes from missed call to first recovery touch | <5 minutes |
| Opt-out rate | Opt-outs / total contacts | <2% |
The opt-out rate deserves particular attention. A rate above 2% signals that your recovery sequence is too aggressive — either the timing is wrong (too late, so they don't remember calling), the tone is off (sounds like a cold call, not a callback), or the frequency is too high (second recovery attempt hitting someone who already opted out). Any of these can be fixed by adjusting the sequence parameters.
Wire all six metrics into a single dashboard — we use a lightweight Postgres table that n8n writes to at each sequence step, with a Metabase dashboard pulling from it. Two hours to set up; saves the ops manager from manually counting outcomes in a spreadsheet each week.
Good / Bad / Ugly: three approaches, three different outcomes
Good: Twilio missed-call webhook → n8n flow → CRM lookup → personalised SMS within 90 seconds → 3-minute wait node → AI voice callback with agent trained on the business's specific service context → calendar booking integration → email backup at 45 minutes. Compliance: documented legitimate interest, STOP handling in CRM, time-window gate. Recovery rate on missed calls: 34%.
Bad: A Zapier automation that sends a generic "sorry we missed you" SMS 20 minutes after the missed call, from an unrecognised number, with no opt-out. Response rate: 4%. Because the SMS arrives 20 minutes late, the lead has already called a competitor. The opt-out omission is also a PECR breach.
Ugly: An automated voice callback that calls at 8pm, opens with "This is an automated message from…", plays a pre-recorded sales pitch, and offers to "press 1 to speak to an agent" who isn't available. Three complaints filed. One went to the ICO. The campaign was shut down.
Missed-call recovery sits at the top of the funnel. For the full picture of how inbound leads are routed, qualified, and handed off once re-engaged, see From Missed Calls to Money: Routing Inbound Leads with AI. The timing constraints of recovery connect directly to the broader research on speed-to-lead windows for UK SMEs — the same five-minute window that determines conversion on new enquiries applies equally to missed-call follow-up. The Voice AI and Document Analysis case study shows how voice agents handle both inbound recovery and outbound qualification in a single deployment.
Book a 30-minute scoping call — we'll map your current missed-call volume and design the recovery flow before you leave.