Generating Opportunity Solution Trees with AI: How Vistaly Rebuilt Its Product Around Interview Synthesis, Evals, and Re

Vistily, founded by Matt O'Connell, CP Daily, and Steve, emerged from a shared recognition of gaps in product development processes - specifically, the lack of structured methods for identifying and validating meaningful problems. Drawing on their experience across agencies, startups, and large enterprises, the team initially experimented with root cause analysis techniques like the 5 Whys before adopting Teresa Torres' Opportunity Solution Trees (OST). This visual, structured framework aligns teams around customer needs and business outcomes, forming the foundation of Vistily's software. Their collaboration with Torres was rooted in genuine engagement, as Matt spent two years actively participating in her Continuous Discovery Habits community before any formal partnership.

Vistily developed a digital platform to enhance the OST methodology, addressing limitations of general whiteboarding tools like Miro and FigJam, which become unwieldy when managing complex, data-rich trees. The software maintains a clean visual structure while storing detailed information - such as customer insights, assumptions, and requirements - behind each node. Early versions allowed manual input, but Vistily V2 introduced AI-driven automation, where uploading customer interviews generates an initial Opportunity Solution Tree. The system emphasizes human-in-the-loop validation, treating AI outputs as drafts to be refined collaboratively. Challenges include ensuring data quality, managing cognitive load, and maintaining trust in AI-generated content, especially when poor interview inputs lead to inaccurate trees. The team continues to refine workflows, focusing on usability, compliance, and meaningful integration of AI without over-reliance on automation.

• What if you built your MVP around AI-generated Opportunity Solution Trees from day one?

  • Move: Start your next solo product by uploading 3 real or synthetic customer interviews into an AI pipeline that auto-generates a draft OST (root, opportunities, solutions), then let users edit it directly on a clean canvas.
  • Why Now?: LLMs like Sonnet 4.6 can now extract structured insights at near-human quality, and developers can build end-to-end workflows in weeks instead of years - Vistily rebuilt V2 in just 2.5 months.
  • Expected Upside: You skip the manual whiteboard phase entirely, delivering immediate value to overwhelmed solopreneurs who lack time for synthesis but need strategic clarity fast.

• What if you treated every user edit as feedback to improve your AI's tree-building logic?

  • Move: Instrument your app to track when users reframe nodes, merge branches, or delete AI-generated opportunities; use this data to train lightweight fine-tuned models or adjust prompts in your repair loop.
  • Why Now?: Users won't give explicit feedback, but their actions reveal what "good" looks like - Vistily found that tracking moves and edits was more reliable than thumbs-up/down ratings.
  • Expected Upside: Your AI gets smarter with each user, creating a defensible feedback flywheel where usage improves output quality, which increases trust and retention.

• What if you enforced strict interview quality before allowing AI processing?

  • Move: Build a pre-upload validator that scores incoming interviews on criteria like question neutrality, customer focus, and depth, blocking low-scoring files unless overridden.
  • Why Now?: Garbage-in-garbage-out remains the top failure mode for AI discovery tools - many "interviews" are actually sales demos or stakeholder chats that corrupt OST integrity.
  • Expected Upside: Higher-quality inputs lead to trustworthy trees, reducing support burden and increasing perceived accuracy, while nudging users toward better discovery habits without extra coaching.

• Join and actively participate in communities related to your domain before pitching or seeking partnerships - build trust and demonstrate value through consistent, non-transactional engagement.
• Focus your product on solving a specific, high-friction problem (e.g., cumbersome manual workflows in digital whiteboards) by combining visual simplicity with backend data depth for better usability.
• Rebuild rather than patch your product when adopting transformative technologies like AI - use the shift to modernize architecture, clarify vision, and align with new user expectations.
• Implement automated error detection and repair loops (e.g., validating tree structure rules like "no more than N children per node") to maintain quality in AI-generated outputs.
• Design AI workflows that prioritize delivering a complete draft output quickly (e.g., generating an Opportunity Solution Tree from three interviews) rather than forcing step-by-step validation, matching real user behavior.