# Making $100K With Voice AI Agents (The Secrets Nobody’s Sharing)

Source: https://www.youtube.com/watch?v=-X9XILuCwcg
Recap page: https://rapidrecap.app/video/-X9XILuCwcg
Generated: 2025-10-18T10:32:30.567+00:00

---
## Quick Overview

The speaker and guest Brendan Jowett detail a successful strategy for building an AI voice agency, emphasizing the importance of rigorous testing and monitoring using their platform, which has allowed them to achieve positive ROI for clients by systematically improving agent performance across specific test cases and personas.

**Key Points:**
- Brendan Jowett, founder of Inflate AI and Relyable (SaaS), has successfully built his AI agency over two years by focusing on testing and iteration.
- A key metric for success is delivering a positive ROI for clients, which is achieved by systematically testing AI voice agents on specific use cases like appointment booking.
- The testing process involves dedicating two weeks for development and six weeks for testing, divided into internal, live, and automated testing periods.
- Automated testing is crucial because generative AI's unpredictability makes manual testing insufficient; it allows for continuous monitoring, scoring every call, and detecting critical failures.
- The platform generates test cases based on the agent's prompt and specific personas (e.g., Nigel Blackwood, an elderly man, or a young person), allowing for targeted stress testing.
- A specific system achieved an average response accuracy of 87.7% with testing, significantly better than the pre-testing performance, demonstrating the value of rigorous evaluation.
- The goal is to ensure AI systems are controlled within defined limits, ultimately aiming for self-learning agents.

![Screenshot at 07:06: The green-highlighted section detailing the benefits of Automated Monitoring, including instantly detecting critical failures, scoring every live call, and sending periodic test calls to pulse the agent.](https://ss.rapidrecap.app/screens/-X9XILuCwcg/00-07-06.png)

**Context:** This video features an interview between Liam Ottley (host) and Brendan Jowett, a pioneer in the AI voice agent space, who shares the methodology behind scaling his successful AI agency, Inflate AI, and developing his SaaS product, Relyable. The discussion centers on the critical importance of rigorous, automated testing and monitoring to ensure AI agents provide consistent, positive ROI for clients, particularly in regulated industries like real estate where compliance and accuracy are paramount.

## Detailed Analysis

The discussion focuses on the practical steps and philosophy behind building high-performing AI voice agents, particularly emphasizing the role of testing and monitoring, exemplified by the features of the Relyable platform. Brendan Jowett explains that while building an agent seems easy initially, achieving production-grade performance requires extensive, continuous testing to manage the inherent unpredictability of generative AI. The testing pipeline involves two weeks of development followed by six weeks of testing (internal, live, and automated). Automated monitoring is highlighted as essential because it allows for continuous scoring of every call, instant detection of critical failures (like compliance breaches or hallucinations), and component failure detection (Platform, Transcriber, LLM). The results dashboard shows performance metrics like the overall score (e.g., 87.7% for one test), average latency (e.g., 2183ms), and words per minute (e.g., 145 WPM). Test cases are structured around specific objectives (like 'Adopt Agent Persona' or 'Initiate Appointment Booking/Flow') and applied against various personas (e.g., Nigel Blackwood, an elderly former history lecturer) to stress-test the agent's conversational flow and functional accuracy, ensuring a positive ROI for clients.

### Building AI Voice Agents 101

- Platform selection (e.g., Replicate, Vapi, Retail, Vappy)
- Prompting using single or connected prompts
- Functions (API Calls) for external data integration

### The Challenge of Generative AI

- Unpredictable nature leads to potential failures like re-worded greetings, missing disclaimers, hallucinations, and incorrect function calls

### Testing Process Milestones

- 2 weeks development, 6 weeks testing (2 weeks internal, 2 weeks live, 2 weeks automated)
- Initial testing involves manual calls by the team and client, followed by automated testing at scale

### Automated Monitoring Benefits

- Instantly detects critical failures (Emergency Pathways/Compliance), scores every live call, detects component failures (LLM, Platform), and sends periodic test calls for agent 'pulse checks'

### Relyable Dashboard & Test Cases

- Interface shows test case lists prioritized as High, Medium, or Critical
- Test scenarios are generated based on system prompt and specific personas (e.g., Nigel Blackwood)
- Results show overall score, latency, WPM, and per-test case success/failure breakdown

![Screenshot at 00:05: Speaker introducing the concept of delivering positive ROI for clients using AI voice agents.](https://ss.rapidrecap.app/screens/-X9XILuCwcg/00-00-05.png)
![Screenshot at 00:08: Illustration showing a man interacting with a voice AI interface, representing the core technology being discussed.](https://ss.rapidrecap.app/screens/-X9XILuCwcg/00-00-08.png)
![Screenshot at 00:12: Brendan Jowett's credentials displayed, noting his roles as Founder of Inflate AI and Relyable \(SaaS\), alongside his YouTube and LinkedIn follower counts.](https://ss.rapidrecap.app/screens/-X9XILuCwcg/00-00-12.png)
![Screenshot at 00:21: A slide summarizing the key topics to be covered, including Platform, Prompting, Functions, Downsides of Conversational AI Prompts, and Automated Testing/Monitoring.](https://ss.rapidrecap.app/screens/-X9XILuCwcg/00-00-21.png)
![Screenshot at 00:29: The speaker points to the 'Morningside' neon sign in the background while introducing the masterclass on building production-grade voice agents.](https://ss.rapidrecap.app/screens/-X9XILuCwcg/00-00-29.png)
![Screenshot at 00:50: Split-screen view of Liam Ottley \(host, left\) and Brendan Jowett \(guest, right\) beginning the interview.](https://ss.rapidrecap.app/screens/-X9XILuCwcg/00-00-50.png)
![Screenshot at 01:17: Brendan navigates the 'How to build an AI Voice Agent 101' presentation, showing the three main components: Platform, Prompting, and Functions.](https://ss.rapidrecap.app/screens/-X9XILuCwcg/00-01-17.png)
![Screenshot at 02:23: Brendan discusses the difficulty of building production-ready voice agents, noting it's not as easy as simple tutorials suggest.](https://ss.rapidrecap.app/screens/-X9XILuCwcg/00-02-23.png)
![Screenshot at 02:46: The host explains the concept of shrinking the time to qualification/closing using AI voice agents for inbound lead qualification.](https://ss.rapidrecap.app/screens/-X9XILuCwcg/00-02-46.png)
![Screenshot at 04:47: A screenshot of the Relyable platform showing call results with performance metrics like Average Score \(87.7%\), Average Latency \(2183ms\), and Words Per Minute \(145 WPM\). This visualizes the quantitative feedback from testing.](https://ss.rapidrecap.app/screens/-X9XILuCwcg/00-04-47.png)