Newgate Systems Start a conversation

Case study · AI systems

Voice agents that hold
a forty minute call.

Air AI runs inbound and outbound phone calls with agents that sound human, remember previous conversations, and act inside the business systems behind them.

Client
Air AI
Industry
Conversational AI
Region
United States
Platform
Real-time call orchestration, CRM integration

01

The client

Air AI replaces or augments call-centre work — outbound sales, inbound support, lead qualification, appointment scheduling — with agents that can carry a genuine conversation rather than walk a caller through a phone tree.

The bar is uncomfortably high. People forgive a chatbot; they hang up on a voice that stalls, talks over them, or forgets what they said ninety seconds ago.

02

The problem

A long voice conversation is a real-time systems problem. Latency has to stay below the threshold where a pause reads as a malfunction, state has to persist across a call that may run forty minutes, and memory has to carry across separate calls weeks apart.

On top of that the agent has to actually do things — log the call, update the record, book the follow-up — which means it is only as useful as its integration into the CRM behind it.

03

Requirements

  • R1Real-time call orchestration with latency low enough to feel natural
  • R2Conversation state maintained across calls of 10–40 minutes
  • R3Memory of prior interactions carried between separate calls
  • R4Integration with CRMs and business tools to log calls and update records
  • R5Round-the-clock availability at volume, without human presence
  • R6Reliability and cost per call kept inside commercial limits

04

The solution

We built the infrastructure underneath the conversation — orchestration, state, memory and integrations — so the agent is a working part of the customer's operation rather than a demo.

The unglamorous truth about voice AI: the model is the easy part. Latency budgets, session state, reconnection handling and CRM write-backs are what decide whether it survives contact with real callers.

05

The work

  • 01Call orchestration. Real-time infrastructure managing inbound and outbound calls at volume.
  • 02Session state. Conversation state maintained reliably across long calls.
  • 03Memory. Recall of prior interactions carried across separate conversations.
  • 04CRM integration. Connections into CRMs and business tools to log calls, update records and schedule follow-ups.
  • 05Scale & reliability. Availability, latency and cost per call held within operational limits.
Real-time systemsVoice AICRM integrationOrchestrationLatency engineering

06

Outcome

10–40 min

Conversation length supported with full recall

24/7

Agents handling volume without human presence

Integrated

Calls logged and records updated in the systems behind them

More work

Other records.

Next

Facing something like this?