Welcome to this fascinating podcast episode where we have the pleasure of diving into the world of AI with Eugene Mandel, the Head of AI at Loris.ai. Mandel, a natural language understanding expert and serial entrepreneur, offers intriguing insights into how AI is reshaping customer conversations in the business world. He shares his journey from a software engineer to an AI pioneer and discusses the future of AI in customer communication.
Here are the top five key points from this episode:
🚀 Eugene Mandel's career transition from a senior software engineer to co-founding a social communications (VOIP) startup and eventually AI in customer conversations.
💡 The unique mindset required for success in a startup, especially while managing a probabilistic product like AI.
🤖 AI's transformative impact on customer support - how AI can be used to recognize conversational markers, create scorecards for customer conversations, and provide context for better customer interaction understanding.
🔮 Eugene's vision of AI's future potential - creating guided summaries, improving accuracy in conversation representation, and enabling non-tech experts to create models for customer conversation recognition.
🔬 An in-depth look at managing experiments and transitioning them into products, with insights into the potential of AI replacing human agents in customer support.
Get ready for an enlightening exploration of AI's influence on business and customer communication!
(0:00:11) - AI Implementation & Startup Experiences (14 Minutes)
Eugene Mandel, the head of AI at Loris.ai, is a passionate natural language understanding expert who has been a founder of multiple companies in the space. We explore Eugene's journey from being a senior software engineer to working with machine learning for HR departments, and his co-founding of the telephony over API company Jaxter. We also discuss the challenges of engineering problems, and the trials and successes of transitioning experiments into products.
(0:13:47) - Startups, AI, and Unusual Mindset (6 Minutes)
Eugene Mandel, the Head of AI at Loris.ai, shares his experience in building startups and how he has come to work with human language and data science. He talks about the unique mindset it takes to be successful in a startup, and how he has seen the development of AI in language rapidly evolve in recent years. Eugene also discusses what makes product managing a probabilistic product different from a deterministic one, and how building trust with users and understanding what is good or bad can make or break a product. Finally, he talks about his current role at Loris AI and how they are using AI to help conversations between two individuals and in call centers.
(0:20:02) - AI in Business Conversations and Support (11 Minutes)
Eugene Mandel's experience building startups and his decision to move into the field of human language and data science are explored. AI is being used to help customer support departments assess the quality of their support, create a culture of continuous improvement, and train their agents better. Eugene's analogy of a conversation in a business context being like a dance, where different moves, sequences, and how to respond to particular questions can be defined, is discussed. Additionally, conversational markers and how AI can be used to recognize them and surface them into metadata to help CX departments improve their conversations, are discussed.
(0:31:26) - Managing Experiments and Transitioning to Products (14 Minutes)
Eugene Mandel, the Head of AI at Loris AI, approaches customer conversations, product management and engineering management with a timeline, distinguishing between delivery and experimentation. He explains how to manage experiments in order to turn them into products. Eugene also shares his insights on how AI can be used to improve customer support, and how AI might eventually replace the need for real live human agents.
(0:45:07) - From Experiment to Product (10 Minutes)
Eugene Mandel, the Head of AI at Loris AI, discusses the challenge of creating scorecards for customer conversations that can be evaluated automatically. He talks about prompt crafting and experimenting with AI models to extract goals and problems from customer messages. The importance of giving AI models the right context and definitions to better understand customer conversations is discussed. Finally, the implications of AI-assisted content creation and what it means for ownership are considered.
(0:54:59) - AI Models in CX (11 Minutes)
We explore how to create custom scorecards for evaluating customer conversations and the implications of allowing clients to inject prompts into these conversations. We discuss the importance of providing context that the clients cannot change, breaking the prompt into small atomic parts, and following the Unix philosophy of small and composable tasks. We also explore the real-world implications of these conversations, the importance of putting guardrails in place, and how user feedback can be used to reinforce learning. Finally, we discuss the possibility of building a machine that non-tech subject matter experts could use to create models to recognize customer conversations.
(1:05:53) - Guided Summary and Predictions for AI (7 Minutes)
Guided summaries can be created by asking specific questions and constraining the output to provide contextual representation and actionable insights. Chat GPT can be used to experiment with different summaries and AI can be used to create summaries that are more accurate and representative of the conversation. Scorecards can be created to evaluate customer conversations and the implications of allowing clients to inject prompts into these conversations.
(1:12:54) - Improving the Logic of LLMs (1 Minutes)
We discuss the potential for natural language understanding (NLU) to improve logic and reasoning in the next year, the possibilities this could open up, and how conversations with customers and product management can be improved. We also explore how scorecards can be created to evaluate customer conversations and how structured summaries can provide contextual representation and actionable insights.