We Built an AI VP of Customer Success That Replaced Hundreds of Human Hours. Here's Exactly How.

This podcast episode features Amelia and Jason discussing their development and implementation of an AI VP of Customer Success tool, named QB, built using no-code/low-code platforms like Replet. They emphasize the transformative power of building custom solutions versus relying on off-the-shelf software, particularly in enhancing customer success and operational efficiency. The core message revolves around the "magical" ability to rapidly iterate and customize software based on direct customer feedback. Unlike traditional software where feature requests face long roadmaps or outright rejection, Amelia highlights that if a customer asks for something not overly complex, they can build and ship it the same day. This agility is presented as a significant advantage, allowing businesses to adapt and deliver precisely what their customers need. Jason introduces the context of their "agentic journey," starting with AI SDRs and then developing their own AI agents, including an AI VP of Marketing and now the AI VP of Customer Success. He notes that QB is the best tool they've built so far, with fewer bugs and greater value generation per unit of time. He stresses that with each iteration, their AI agents become better, encouraging listeners to embrace this continuous improvement process. A key takeaway for the audience is the universality of their approach. They aim to share solutions that are not niche to their specific business (SAS, media, community, events) but are applicable to any B2B company. The central problem QB addresses is the difficulty for humans to manage a large customer base with numerous deliverables and follow-ups. The solution, they argue, is an AI agent that can scale to manage this effectively. They urge listeners to think about how they can build their own customized version of such tools, rather than dismissing the presented solutions as specific to their industry. Amelia details the evolution of QB. It began as a simple project management tool for Saster Annual sponsors, intended to replace an existing paid tool that lacked AI capabilities. The initial spec was basic: task assignment, single sign-on, and light automations for reminders. However, as QB was deployed and used by actual customers, its potential as a more agentic tool became apparent. A significant "unlock" was the level of insight QB provided into customer engagement. Unlike their previous portal, which offered no data on logins or activity, QB provided granular visibility into who was logging in, what they were doing, and what they were not. This data allowed them to automate personalized communication and proactive follow-ups. The AI VP of Customer Success, QB, functions as a sophisticated onboarding automation tool. It manages approximately 13 core tasks with subtasks for each sponsor, proactively reminding them, identifying gaps, and analyzing uploaded assets. This process is managed daily, with Slack and email updates provided. This is contrasted with the unreliability of human teams or third parties to consistently perform these tasks. The tool ensures that customers complete their onboarding and deliverables efficiently, reducing stress and improving the overall experience. Amelia illustrates the dramatic reduction in human hours and billable costs achieved by QB. Comparing Q1 of the previous year to Q1 of the current year, they saw a 70% decrease in billable hours from agencies and production teams, translating to thousands of dollars saved monthly. The cost of building QB with no-code platforms is presented as minimal compared to these savings. Regarding costs, Jason states that across all their AI applications, including QB, their total token usage has remained under a $200 monthly cap. He emphasizes that while there are soft costs associated with building and maintaining these agents, the hard costs for AI usage are often surprisingly low for most use cases. The evolution of QB from a static portal to a fully agentic tool is detailed. The "before" state involved manual, newsletter-style communication with generic links, leading to low customer engagement. The "after" state, with QB, involves highly personalized, customized emails sent in minutes, with unique links and information tailored to each sponsor's specific needs. This has led to a tenfold increase in customer engagement. A key example is TikTok, a sponsor who was previously slow to submit deliverables but, with QB's personalized guidance, completed almost all tasks within a day. The process of building and deploying QB is outlined: 1. **Write a Spec:** Amelia shows both an initial, basic spec and a more advanced, current spec developed with QB's assistance. She advises using AI tools like Claude to help generate and refine these specs. 2. **Load into Vibe Coding Platform:** Using platforms like Replet, the spec is fed to the agent. Amelia emphasizes giving the agent design preferences and examples of desired functionality. 3. **Test Every Function:** Thorough testing of all features is crucial before deployment to ensure functionality and prevent breakage. 4. **Hook up Email Capabilities:** Integrating email allows for personalized communication and database creation. 5. **Agent Hopping for Sensitive Data:** For security, sensitive data like customer contracts is not stored directly within the agent. Instead, it's accessed via integrations with systems like Salesforce or Clerk. This "agent hopping" adds a security layer, preventing all sensitive data from residing in one place. 6. **Deploy to Production:** The agent is initially deployed to a few select customers to identify and fix bugs. Amelia shares an example of a critical bug related to login timeouts that was discovered and fixed through real-world deployment. 7. **Iterate and Add Features:** Based on customer feedback and observed usage patterns, new features are continuously added. Examples include email marketing copy, networking information, and booth graphic submissions. QB's ability to manage speaker time slots and prevent conflicts is highlighted as a significant improvement over manual processes. 8. **Leverage Data for Agentic Actions:** Understanding where data can be leveraged enables the transition from a portal to an agent. This involves automating tasks like sending personalized tips, reminders, and proactively managing customer needs, even escalating to collections when necessary. Jason reiterates the importance of maintenance. He states that there is no "set and forget" with these agents. Daily checks, status updates, and proactive problem-solving are essential. He advises having agents send daily email status reports to catch issues that might otherwise go unnoticed. The discussion touches on whether to disclose that the tool is AI-powered. For their high-value B2B clients, they currently do not explicitly state it, as the AI handles the majority of automated tasks, while humans remain involved for bespoke processes and higher-tier client engagement. This hybrid approach satisfies clients who desire instant access to data via agents but also value human interaction. They adapt their engagement methods to client preferences, whether it's Slack, email, or dedicated calls. The episode concludes with an invitation to Saster Annual, where attendees can learn to build their own AI agents with hands-on support from no-code platform providers. They also promise a follow-up session after the event to share analytics and insights from their continued use and development of these AI tools. Jason challenges the audience to identify broken processes in their own organizations, particularly in onboarding, retention, and customer education, and to consider building custom AI solutions to achieve 100% coverage and efficiency.