In this episode of Agony Agent, Ben McCulloch welcomes conversation design experts Ilana Meir and Peter Isaacs to tackle pressing issues faced by conversation AI designers.
Ilana, a freelance conversation designer with extensive experience in AR, VR, and healthcare, and Peter, a senior conversation design advocate at Voiceflow, share insights on improving bot interactions, overcoming organisational challenges, and leveraging large language models (LLMs).
Key topics include how to navigate archaic systems when designing seamless user experiences, managing sentiment analysis in conversation design, and balancing the demands of impulsive sales teams with technical developers. They also discuss the impact of context switching for designers managing multiple brands and the challenges of fine-tuning LLMs for scalable AI solutions.
This episode is a must-listen for anyone in the conversational AI industry, offering practical advice on how to combine technical and creative expertise to build better, more human-like interactions. Whether you’re refining chatbots or introducing generative AI, you’ll find valuable strategies to implement right away.
In this episode
00:00 Intro
04:58 Creating beautiful interactions with archaic solutions
11:05 Learn sentiment analysis and do emotionally sensitive conversation design
17:19 Educating user to speak so the NLU understands
21:17 Using LLMs for intent recognition
22:54 Meaningful success metrics
27:49 Advocating for a holistic view between competing stakeholders
32:38 Balancing Gen AI obsession with its abilities
37:30 How to build a chatbot that matches ChatGPT
39:19 Preparation before building
40:47 Think about frameworks where conversations will happen
41:56 Finetuning LLMs
44:33 Making people see value of CAI and experts
46:17 Is chatgpt a good way to show what CAI can do?
47:47 CXD for SaaS services with LLM solutions
51:52 How to advocate for your role
54:19 Context switching multiple brands in a services organisation as a manager
58:24 How to go beyond information provided by client when training an assistant
