Enterprise Virtual Agent Platform Conversica Debuts Generative AI Chat Features
Enterprise AI developer Conversica has begun incorporating generative AI into its platform through its new Conversica Chat service. The new feature leverages large language models to engage in more complex, open-ended conversations like ChatGPT, but with a far more business-focused approach and more guardrails to limitinaccurate or inappropriate responses.
“We jumped into [generative AI] pretty early in 2019 but didn’t talk about it earlier because it was so immature and so hard to use, Conversica CEO Jim Kaskade told Voicebot in an interview. “The promise of this technology is that it can mimic a human and maybe even improve on what a human can do. Chat just one [channel], we’re providing [Conversica Chat] for emails, SMS and we even have a prototype for voice.”
Conversica Chat employs a mix of OpenAI’s GPT-3.5 API and LLMs designed in-house to satisfy the need for both a brand-specific conversation and the wider range of topics and responses that broader generative AI chatbots can offer. Kaskade described a theoretical situation where a company like Adidas might deploy Conversica Chat to guide customers through the process of finding and buying shoes, switching from the Adidas-specific LLM when asked about available models, to OpenAI’s API when the conversation turns to fastest runners and their shoe choices, then back to the custom model to complete the transaction. The GPT-3.5 model is fine-tuned as much as is feasible, but the bulk of any brand-specific training goes into the Conversica Chat model so it can represent a company not just in terms of accurate information, but with regard to tone, word choice, and other marketing considerations.
“Use cases are uniquely tailored to customer prospects and services. It’s very revenue-centric,” Kaskade said. “From our perspective only thing limiting [Conversica Chat] ts is making sure the AI doesn’t make mistakes. It still needs filtering on the front end and the back end.”
The combination of models also allows Conversica Chat a firmer grip on the responses to users, making sure they are accurate and that customers can’t inadvertently or deliberately provoke awkward or offensive responses from the AI. Conversica Chat supports more than 100 languages and on every communication channel. That means a discussion with a chatbot can then send an email or text message follow-up with reference to the earlier conversation, like how a human agent would keep up with a potential sale.
There’s been a real rush among customer service and other business software developers to embed generative AI in their systems in different forms. For instance, Yellow.ai recently released its Dynamic Conversation Designer to streamline designing a voice or text-based conversational AI. Conversica has a unique approach and history with this kind of technology, but Yellow is far from the only potential rival in the field. NLX has brought generative AI to its multimodal virtual assistant platform. The growing list also counts business conversational AI provider LivePerson, Gupshup, Yext and its Auto Bot Builder chatbot maker, and the upgraded Kore.ai platform are making generative IA a part of their business model.