How Viable Built Their AI Moat: A Framework for Defensive Data Strategy

Most AI startups begin with a simple assumption: collect enough data, and you’ll build an insurmountable lead. But in a recent episode of Category Visionaries, Viable CEO Daniel Erickson revealed why this thinking falls short – and how his team built a more sophisticated approach to AI defensibility.

The Training Data Paradox

“With any sort of AI startup, I believe that your biggest moat is always going to be your data moat, specifically around training data,” Daniel explains. But there’s a catch: you need high-quality training data before you have enough customers to generate it naturally.

This challenge is particularly acute in enterprise AI, where Viable operates. The company discovered that “80% of data that is collected by companies today is unstructured text” – including everything from survey responses to support tickets. But having access to this data isn’t enough. The real moat comes from knowing how to turn it into effective training data.

The Two-Pronged Strategy

Viable’s solution combines two distinct approaches to building their training data advantage:

1. Embedded Feedback Loops Rather than just collecting data, Viable embedded quality signals directly into their product. As Daniel explains, “When you thumbs something up, it actually tells us that was a good answer, and so we can use that as training data going forward.”

This creates a virtuous cycle: every customer interaction not only solves their immediate need but also improves the system’s capabilities. The key insight is that not all user interactions are equally valuable for training – you need mechanisms to identify the high-quality examples.

1. Systematic Data Generation For new features or capabilities where they don’t yet have user feedback, Viable developed what Daniel calls “amazing systems to pump out new training data sets very quickly.”

This capability allows them to expand into new areas without waiting for organic data collection. It’s particularly crucial for enterprise AI companies, where customers expect comprehensive capabilities from day one.

Building the Analysis Engine

What makes Viable’s approach unique is their focus on building the analyst, not just the analysis tools. “What’s different about us is that we are actually building analyst,” Daniel notes. “We’re not building tools for analysts. Our end user is the business user who would normally be going to analyst to get this information.”

This distinction shapes how they think about data collection and model training. Instead of optimizing for technical metrics, they focus on replicating the insight generation process of skilled analysts.

The Enterprise Advantage

The enterprise focus actually strengthens Viable’s moat. When working with large companies, they noticed that teams were struggling because they could only report on “the most recent thing or the loudest thing” in their customer feedback. This insight led them to build systems that could process feedback across multiple channels and time periods.

Each enterprise customer provides not just more data, but more diverse use cases and edge cases that help improve their models. This creates a compounding advantage – each new enterprise customer makes the system more valuable for everyone else.

Future-Proofing the Moat

Looking ahead, Viable is evolving toward what Daniel calls “generative analysis,” which involves “allowing our customers to guide analysis of the analysis that we’re doing for them.” This next phase focuses on “tying company goals back to customer feedback and helping you use customer feedback to achieve those company goals.”

This evolution demonstrates a crucial principle about AI moats: they need to grow in sophistication, not just size. The goal isn’t just to accumulate more data, but to build systems that can generate increasingly valuable insights from that data.

Building Your Own AI Moat

For founders building AI companies today, Viable’s approach offers several key lessons:

1. Focus on training data quality over quantity
2. Build feedback mechanisms directly into your product
3. Develop systematic approaches to generate training data for new capabilities
4. Think about your moat in terms of insight generation, not just data collection
5. Use enterprise complexity to your advantage

The real insight isn’t just that data creates defensibility – it’s understanding how to turn data into a continuously expanding competitive advantage. As Viable demonstrates, building an AI moat requires thinking beyond simple data collection to create systems that generate increasingly valuable training data through every customer interaction.