In an earlier insight, we explored how chatbots can use the OpenAI‘s Generative Pre-trained Transformer 4 (GPT-4) model. ChatGPT is one of many products that uses this framework to showcase the technology’s prowess with some interesting results. This new Have Your Say entry lets you share your thoughts about chat agents and whether GPT-4 is perfect for them.
Introduction – Chatbot
A chatbot is an interactive computing program that uses a mix of artificial intelligence, natural language processing, and data to learn customer experiences and automate responses to them. Many retailers use them today as Level 1 customer service team response agents to manage simple requests like transaction inquiries, payments, address changes, etc. The main objective is to enable customers to self-manage their retailer needs with a platform that sounds or feels more natural with a conversation interface.
For those who are not as familiar with the use cases for conversational interfaces, here are some examples:
– Finding restaurant suggestions in a neighborhood
– Inquiring about the latest mortgage rates and providing simple calculations on payments
– Booking an appointment
– Getting updated flight information and completing the check-in process
For this chat technology to be effective, it has to go beyond automating questions with standard responses. Customer words and actions along with personal and transactional data should be taken into consideration in how the chatbot should respond.
For example, Bill is a new customer who is searching for last month’s bill (neutral tone) and uses a chatbot for self-service. After receiving the bill, he questions the content of the bill (possibly angry tone) and needs a trained human agent to respond.
How will ChatGPT influence chatbot innovation? Image by: Andrew Neel at Pexels
Current technologies being deployed by retailers are not able to differentiate between the two tones. GPT-4 and other multimodal models have the capabilities and processing power to generate and process more inputs in providing the best options possible. In Bill’s case, the tool would be trained to identify words like “this is wrong”, “incorrect amount”, and “need to talk to agent” as potential queues for possible de-escalation.
This would reduce the friction between the retailer and the customer. After all, the technology is meant to facilitate a seamless interaction and not create more issues.
We want to get your thoughts!
1. Have you used this technology before?
2. How do you like it?
3. Do you think advanced models like GPT-4 will produce better bots in the future?
Feel free to sound off in the comments below on the technology.