February 2022 Summaries

When building conversational experiences in Voiceflow, maintaining organization by leveraging topics and reusable components (flows) can help developers work smarter, not harder. Topics are organized around global and local intents, allowing for a more intuitive system of organization that keeps the conversation canvas clean. Components, or flows, are used to organize workflows triggered after an intent is activated, enabling easy reuse across projects. By utilizing these features, developers can create more efficient and scalable conversational experiences.

We're working closely with customers to bring simple yet inspiring features to the Voiceflow platform that leverage our understanding of voice conversations, focusing on a highly active segment who's passionate about the platform and willing to become beta users. We're using rapid prototyping to develop capabilities quickly on both UX and technical sides, and have built tools outside our core application to decouple from the application team and get results faster. Our ML CLI allows developers and data scientists to create isolated end-to-end Machine Learning environments quickly, while MLEnvs provide a set of terraform templates for generating models and environments in 15 minutes. We're building a Real Time Platform with an event-driven architecture using sockets, and separate model development and deployment to focus on building a V1 of the platform. We're leveraging open source tooling and contributing to it, hiring for three core technical skills: data science in our core competency areas, data engineering, and cloud/platform engineering. Our goal is to work closely with customers, build a world-class ML platform that can scale to millions of users, and contribute to the ML community through articles, talks, and open source projects.

The key to structuring a conversation design involves considering statistical approaches such as Bayesian reasoning, which provides actionable results faster and focuses on reaching statistical significance without requiring in-depth knowledge of statistics. To start, analyze conversational transcripts, historical data, and usage patterns to identify areas for improvement and formulate a hypothesis tied to a specific metric, such as NLU accuracy or human handover percentage. Next, create a variation that directly relates to the hypothesis and tests only one aspect at a time to ensure accurate results. Analyze statistical significance using an A/B test calculator and deploy changes to refine the conversation design. By following this structured approach, designers can iteratively optimize their conversational experiences to drive optimal user experience.