How SaaStr Built an AI VP of Customer Success for $200/Month
The magical thing that Amelia said, it's so magical, is that unlike off-the-shelf software, when a customer asks you, 'Hey, could you do this?' Instead of saying no, or asking your CS person if it's on the roadmap for the next 24 months, if it's not that complicated, we go into Replit and just say, 'Build it now.' We can ship it that day and we're not engineers. To say that is magical is an understatement.
I put a total $200 a month cap on all token usage across these apps. We couldn't afford more, but I just wanted to not worry about it. That's the allowance. This is for SaaStr.ai, the pitch deck uploader, valuation calculator, all the apps we've built for all times, 10K, Qubi, and we haven't hit the $200 a month cap on AI use yet.
We have literally fought with, giving people who didn't want to do this either, internally, outside agencies. You probably fight some of your CSMs and external teams on like how much coverage they've got. Again, maybe for smaller, starting up, you've got one CSM for 50 accounts, that's never viable, right? Unless you do it with an agent. Now I do think that's viable with person plus agent.
- Building an AI agent for customer success, even if it starts simple, can dramatically reduce human hours and improve customer engagement over time.
- The initial version of their AI VP of Customer Success (Qubi) was a basic project management tool, but through iterative development and user feedback, it evolved into a fully agentic system.
- A significant benefit of using an AI agent like Qubi is the ability to provide highly personalized and timely communication to customers at scale, which was impossible with manual processes.
- The cost of running AI agents for various applications, including customer success, can be surprisingly low, often staying below a few hundred dollars per month for token usage.
- When building AI apps, it's crucial to deploy them to a small group of users first to identify bugs and gather feedback, as real-world usage reveals issues not caught during internal testing.
- For security, sensitive customer data should not be stored directly within the AI agent's native knowledge base but should be integrated from more secure systems like Salesforce.
- The 'magical thing' about building your own AI tools is the ability to quickly implement customer requests and automate tasks that off-the-shelf software cannot, leading to significant competitive advantages.
- Even with AI automation, human oversight and maintenance are critical; AI agents require daily checks and adjustments to ensure continued functionality and address regressions.
- 70% decrease in human hours (Achieved in Q1 2025 compared to Q1 2024 for managing customer success tasks, translating to thousands of dollars saved monthly.)
- 10x increase in customer engagement (Observed in how users interact with their AI agent compared to previous manual processes.)
- $200/month cap on token usage (The maximum budget set for all AI token usage across multiple internal AI apps, which they haven't hit yet.)
- $100,000 average deal size (SaaStr's average deal size, influencing their hybrid approach to human and AI customer engagement.)
RevBots.ai View:
- AI Sprinkler stage teams can learn from SaaStr's approach to building custom AI tools.
- Starting small and iterating is key: even basic AI tools can evolve into powerful systems.
- AI Sprinkler stage teams should prioritize security: integrate sensitive data from CRMs, not AI knowledge bases.
- Human oversight is non-negotiable: AI agents need daily maintenance, even in the ARM stage.
Join The RevBots ARMy
The insider daily for Autonomous Revenue Masters.