Marc Erra´s interview: Oct8ne+ChatGPT & Decantalo´s case study
Before going into details, can you explain what ChatGPT
Chat GPT is something most of us are familiar with; it’s an AI model that has impressed us mainly with its language comprehension and its ability to provide responses based on customer questions.
What does Chat GPT bring in terms of “improvements” to Oct8ne?
In our case, we use it to add a layer of artificial intelligence to what would be Oct8ne’s decision tree-based robot. So, what we do is harness that comprehension ability to try to make the robot more human-like. We use this AI to allow ChatGPT, through a training process we can provide with prompts, to offer real-time responses to customers on a website or platforms like WhatsApp, Facebook, and Instagram. We then link it to all the transactional aspects that ChatGPT can’t handle on its own, but we can. Thus, using this artificial intelligence, we can check order statuses, provide recommendations through the coviewer, handle returns – all those transactional actions that any business and website linked with AI would have.
Tell us how the Decántalo chatbot works and what benefits the integration with GPT has provided.
Decántalo had been working with us for some time, both with human agents providing answers to customers and through a decision tree-based chatbot that could automatically respond to customer queries. What Decántalo sought in this case, being an online store that always tries to stay at the forefront of technology to boost sales, was to humanize the robot. They wanted to retain all those integrations and functionalities the robot currently offers, like identifying if the user is logged in to offer different options, loyalty points, order status, and catalog product recommendations. This last point is crucial because, at this moment, chatGPT doesn’t know what products are on their webpage. However, through its comprehension ability, it can discern what the customer is looking for. With these instructions, we can search for products on the website and provide real-time recommendations, turning the chatbot into a sales-oriented one. If the bot isn’t able to solve the customer’s problem or the response isn’t what they expected, the conversation can be handed over to a human agent to address the queries.
How was this bot configured specifically for Decántalo?
Technically, since Decántalo already had a bot with us, we added the Chat GPT artificial intelligence layer to their existing robot. The process involved defining the project, followed by defining the prompt, which is a crucial part to ensure the robot stays on script. It’s important to note that ChatGPT has information we haven’t provided and might provide inaccurate or made-up information. We want to avoid that, so the prompt definition is crucial. We decided what we wanted the robot to be able to respond to using AI and what could be answered with more transactional information. After that, it was about a two-week process followed by testing and launch.
Why would you recommend integrating the Oct8ne chatbot with Open AI technology?
I wouldn’t recommend it universally; it depends on the case. Not everyone needs an AI-powered chatbot or a conversational one. It depends on what a business is looking for, and based on that, choosing the option that’s most useful and fitting for the business. Why add Open AI? To humanize the conversation. Open AI turns the conversation from a rigid flow to a more natural one.
Do you think Open AI’s technology is applicable to websites in different sectors and can add value to each of them?
Absolutely. I think the sector isn’t the main concern. What we leverage from Open AI is its language comprehension and conversation abilities, so it’s less about what you sell or the type of business you have and more about what you want to achieve. The configuration is what matters to make Open AI help us achieve optimal results.
How do you see the chat market evolving with the advancement of this new technology?
I believe it’s evident that the evolution will focus on adding more artificial intelligence to make the robot’s interactions much more human-like. About 4 to 5 years ago, we started working with AI models involving language comprehension, but the understanding was limited. As a result, the outcomes were not as expected. So, we turned to decision trees, which were more result-oriented. With the advent of Open AI and others in recent months, the barrier of language comprehension has been surpassed. It’s now so good that it can even consider previous questions. Therefore, the evolution will undoubtedly lead to more artificial intelligence incorporation to make solutions increasingly human and provide a smoother experience for end-users.