A voice AI agent that resolves 70% of support calls without a human

The challenge

QuickBooks Accounting runs a small in-house support team that was getting overwhelmed every tax season, with call volume spiking 4x and wait times stretching past 20 minutes. Hiring seasonal staff wasn't scaling — training took weeks and quality was inconsistent.

What we built

We designed and shipped a voice AI agent that answers first-line support calls directly:

• Real-time speech-to-speech pipeline with sub-second response latency
• Tool-calling into their existing ticketing and billing systems to pull account status, invoices, and payment history
• Guardrails and escalation logic that hands off to a human agent the moment a call needs judgment, refunds, or gets emotionally charged
• Call transcripts and outcomes logged automatically for QA and compliance

The agent went live in a phased rollout — shadow mode first, then live for a subset of call types, then full first-line coverage.

The results

Within the first month, the agent was resolving 70% of inbound calls without any human involvement, cutting average handle time nearly in half and giving the team 24/7 coverage for the first time. Human agents now spend their time on the calls that actually need them.