TL;DR: In 2025, we tried to build a GenAI app to map customer service phone menus, launch in 4 weeks, and make $1. We failed at most of it. This is the story of what went wrong, what we learned, and why 2026 will be different.

In the early summer of 2025, we embarked on a project with five simple goals:

1. Reverse engineer customer support phone trees using GenAI.
2. Use GenAI as a core component.
3. Wrap the project in 4 weeks.
4. Make $1.
5. Get to know each other as potential co-founders.

We only really succeeded at one of them. We got to know each other. We also have a website with about 10 phone menus partially mapped. But we didn't launch in 4 weeks, and we certainly didn't make a dollar.

This isn't a story of failure, though. It's a story about learning. As we reboot for 2026, we're building in public, and the first step is sharing what we learned the hard way.

What We Learned the Hard Way

Our ambition was high, but our execution was hampered by a mix of technical and non-technical hurdles. For any other hobbyists or builders out there, maybe our struggles can help you sidestep similar issues.

Technical Lessons

• The Steep Learning Curve of Voice AI: We had no experience with voice apps. We underestimated the complexities of getting the Twilio API to reliably dial numbers, listen for responses, and send inputs at the right time. It was a constant battle of syncing our script with the IVR's timing.
• GenAI Isn't Magic: Getting Google's Gemini 2.5 Pro to reliably parse unstructured voice transcripts into a structured phone menu was incredibly difficult. Prompt engineering is an art, and we spent countless hours tweaking prompts to handle the sheer variety of ways an IVR can present options.
• Schema is incredibly important: We started without a clear data structure for our phone trees. This was a mistake. A vague schema meant we were constantly refactoring how we stored and processed the data. We still don't love our schema, but we now understand that it should have been one of our first priorities.

Team & Planning Lessons

• Side Projects Need Structure: As a side project, interest came in waves. When work or family life got busy, the project was the first thing to be deprioritized. Without a clear plan and accountability, a few hours a week at a coffee shop wasn't enough to maintain momentum.

What Went Right

Despite the challenges, it wasn't all for nothing.

• We learned a tremendous amount about prompt engineering, voice APIs, and the challenges of building a real-world AI product.
• Most importantly, we kept the project alive.

2026 is Going to Be Different

This year, we're all in on building in public. We'll be posting regular updates on our progress, sharing our technical learnings, and documenting our journey to finally make that $1.

We're building PhoneSupported for people who are tired of navigating confusing phone menus, and we want you to be a part of it. Have you ever tackled a project like this? What did you learn?

Follow our journey on Twitter/X [@PhoneSupported]!

Team PhoneSupported