VZU Portfolio · voice-agently · adaptusgroup
Voice Agently.
AI voice agents. 180ms p50. 14 languages. Your brand's voice.
Inbound + outbound voice agents that actually pick up the phone. 1M+ minutes a month in production. 200+ operators deployed. A streaming pipeline that runs end-to-end in 180ms p50. The Vancouver dental chain case study: 22% abandon to 1%, $21k/mo staffing to $4k/mo, 6-month payback.
For the small business
Three simple steps. Live in a week.
Step 01
Forward your number.
Port your number to Voice Agently or set up call forwarding in five minutes. The existing line stays live. The agent takes the after-hours volume the moment you switch it on.
Step 02
Pick a voice and a script.
14 languages with regional accents. Pick the voice. Type the script in plain English. The agent handles the FAQ, books the appointment, takes a message, escalates the exception — out of the box.
Step 03
Watch the calls.
The dashboard shows every call, the transcript, the booking, the escalation. You pay per call after that. The pilot is fixed-fee for the first 30 days. The architecture is real, the ship date is real.
The product
What Voice Agently does.
Voice Agently is the SMB-facing brand of VZU's voice-agent product. Inbound + outbound. 14 languages with regional accents. Production-grade, in the operator's brand voice, on the same NetWit Agentic OS runtime as every other brand in the VZU family.
The MCP integration story is the brand's superpower. Every action the agent takes is an MCP server — book the appointment, fetch the customer record, write back the call summary, escalate the exception. The agent sees what the agent is allowed to see. The audit trail records every call. A new capability ships in an afternoon.
Cloud, hybrid, or on-prem — same code, same data, operator-controlled. The cloud deployment is the default. The hybrid is the right answer when the operator's data can't leave the perimeter. The on-prem is the right answer when the regulator requires it.
The architecture
180ms p50. Streaming pipeline. Edge-deployed.
Voice Agently runs a streaming pipeline — audio in → STT → LLM → TTS → audio out — fully parallel. The first 200ms emits a "Got it." filler while the LLM thinks, so the caller never hears a dead line. End-to-end p50 is 180ms. Sub-200ms means the conversation feels like a conversation, not an IVR.
The agent runtime — Vega — orchestrates turn-taking, intent classification, entity extraction, writeback, and escalation. The operator picks the LLM and the voice. Vega handles ~80% of calls end-to-end; the remaining 20% escalate to a human with the full context, not a half-transcript.
Edge-deployed in five regions — Vancouver, Dubai, Mumbai, London, Bangalore. The call hits the closest region, the round-trip is short, the latency is the same everywhere. The audit trail is on from the first character.
The proof
14-location Vancouver dental chain. 22%→1% abandon. 6-month payback.
22% → 1%
call abandon rate, pre to post
$21k → $4k
monthly staffing cost, pre to post
+18%
booking rate uplift
6 mo.
payback on a 4-month pilot
The chain was running 14 locations with a shared after-hours answering service. 22% of inbound calls were abandoning on hold. The team was re-keying the next morning. The booking rate was flat despite growing call volume.
Voice Agently took the after-hours volume on day one. The agent answered in 180ms, in English with the regional accent, with full access to the practice-management system via MCP. 1% abandon on the new volume. The next-morning re-key dropped to zero — the agent wrote the call summary to the patient record before the call ended.
The dental chain kept the front-desk staff. It redeployed them to the in-clinic patient experience. The 6-month payback was on the staffing delta alone; the +18% booking rate paid for the second year.
Reference deployment
F50 Telecom. $150M annual handle-cost savings.
$150M
annual handle-cost savings
14M+
monthly calls handled
9
languages
4.4 / 5
CSAT, agent-handled
The same Vega runtime that powers the dental chain powers a Fortune-50 telecom. 14M+ calls a month, 9 languages, sub-200ms p50, CSAT 4.4 out of 5. The same operator-controlled deployment model — same binary in cloud, hybrid, or on-prem.
The problem
The phone is the leakiest bucket.
The 67% no-voicemail stat
Two-thirds of callers won't leave a voicemail.
The caller rings, gets an IVR, gets a queue, hangs up, calls the next provider. The "missed call" line item on the P&L is the line item nobody owns. The after-hours volume is invisible to the owner until the month-end close.
The IVR tax
Five menus, zero resolution.
The legacy IVR pushes the caller through five menus, dumps them into a queue, plays the same hold music for 11 minutes, and resolves 40% of nothing. The CSAT is in the basement. The agent on the other end has no context when the call lands.
The compliance gap
The audit trail is missing.
Healthcare, financial, and government operators need the call recorded, the transcript retained, the consent captured, the access controlled. The legacy answering service gives them a spreadsheet. The cloud AI vendor gives them a dashboard. Neither gives the regulator the trail.
Migration
Switching from a competitor. 2-4 week migration.
Week 1
Discovery
Map the existing call flows, the CRM, the calendar, the escalation list. Identify the calls you can hand off to the agent on day one.
Week 2
MCP wiring
Stand up the MCP servers for the calendar, the CRM, the knowledge base. Vega learns the workflows. The audit trail is on from the first test call.
Week 3
Parallel run
The agent handles a portion of the volume; the legacy IVR or answering service handles the rest. You compare transcripts, CSAT, resolution rate, abandon rate, in real time.
Week 4
Cutover
The competitor is retired. The agent is on 100% of the volume. The migration is the proof — the architecture is real, the ship date is real, the on-call is real.
The migration is a fixed-fee engagement. The architecture is real. The ship date is real. The audit trail is on from the first character. The competitor's contract is the only thing that has to wait until the end of the term.
Key stats
Voice Agently, in numbers.
180ms
p50 latency, end-to-end
14
languages with regional accents
1M+
minutes / month in production
200+
operators deployed
~80%
Vega auto-resolves; 20% escalates with full context
5
edge regions — Vancouver · Dubai · Mumbai · London · Bangalore
MCP
every action is a typed MCP server call
Everything it does
What Voice Agently ships with.
- 01Inbound + outbound voice agents in 14 languages with regional accents
- 02Streaming pipeline architecture — 180ms p50, "Got it." filler, no dead-line silence
- 03Vega agent runtime — turn-taking, intent classification, entity extraction, writeback, escalation
- 04MCP-native integration with CRM, EHR, ERP, calendar, knowledge base, and every other VZU brand
- 05Audit trail on from the first character — every call, every prompt, every tool call recorded
- 06RBAC at the agent level — the agent sees what the user is allowed to see
- 07Cloud, hybrid, or on-prem — same code, same data, operator-controlled
- 08SOC 2 Type II, HIPAA, ISO 27001, PIPEDA, ISC²/CISSP-aligned by default
Built for
Who runs on Voice Agently.
Get in touch
Brief VZU on Voice Agently.
Email the team. Reply in one business day. The pilot is fixed-fee, the architecture is real, the ship date is real. Run trace on the dental chain case study at vzu.ca/services/voice-agents/.