Finetuning GPT to develop a Copilot that can Start and Close Online Sales

Automating Online Sales with a New Breed of Copilots: the next generation of GenAI Copilots moves from passively answering customer questions to actively executing online sales. These new Copilots are proactive, they can start and drive an interaction with a potential customer; and are context-aware, they know the different steps in the sales process, where they are in the process and how to move to the next step. You can test the Bitext Sales Copilot at, in English and Spanish. You can read about the Bitext Sales Copilot at

We continue to see initiatives to bring LLMs to leverage GenAI power for business users, particularly for GPT. Volkswagen, Mercedes Benz, Toyota, Ford and BMW are good recent examples. Customer Support emerged initially as one of the use cases with the highest potential, probably following the pre-chatGPT chatbot trend.

More recently, closing Online Sales is becoming an enticing new challenge for automation. While a Customer Support Copilot passively waits for a user question; an Online Sales Copilot needs to drive and close a transaction with different steps like selling a ticket, making a reservation or onboarding a client. To complete a full transaction a Sales Copilot needs two main abilities: proactivity and awareness of context:

  • PROACTIVITY to start the conversation and drive the user through the transaction. The Sales Copilot needs to request the different types of information needed, taking the initiative and not just passively waiting like regular assistants. For the case of plane ticket sales, the Copilot will start the conversation asking where and when the user wants to fly.
  • CONTEXT AWARENESS to be able to guide the user through the different steps till the transaction is complete. The Sales Copilot will structure the conversation in different stages. For the case of ticket sales, the Copilot will request flight information and personal data, will offer ticket options, etc. Also, the Copilot can make a difference between items that can be ignored, like the Frequent Flyer program or seat preferences; and items that are essential like flight dates.
In short, the Sales Copilot needs to overcome the limitations of old web/app forms and of passive chatbots that expect a question from users. Forms, and chatbots sometimes, are the most frequently used tool for completing transactions online today, like onboarding potential clients in banking, selling plane tickets, booking hotels or restaurants, etc. Forms are uncomfortable, rigid and passive tools that make clients abandon the purchase process, the most typical case being in online shopping, like “abandoned cart”.

At Bitext we’ve released a new GPT-based Copilot that drives transactional processes, focused on executing web-based sales. It leverages all the power of GPT while avoiding the typical issues for enterprise use: hallucination, PII conflicts and bias. To share a taste of how it works, we’ve released a demo for selling flight tickets. Some of the capacities that you can experience in the Bitext Sales Copilot:

  • UNSTRUCTURED INPUT – Users can provide information as they like, with no need to adjust to the requests of the Copilot. The user provides her/his flight information provided with no structure:
  • BACKTRACKING – Users can change opinion at any point in the conversation and modify information already provided
  • SUPPORTING FAQs – The Copilot will answer user queries that are relevant to complete the transaction. The Copilot will interrupt the flow of the process to answer the question and then recover control of the flow
  • NON-RELEVANT QUERIES – The Copilot can identify queries that are not related to the transaction, gently avoid them, and keep focus on the transaction flow

You can test the Bitext Sales Copilot at, in English and Spanish. You can read about the Bitext Sales Copilot at

Sharing is caring!