Questions about our new AI chatbot, Fin? Here's everything you need to know

It's been just over two weeks since we launched Fin, our new AI chatbot, globally.

The idea of ​​an AI chatbot answering your customers' questions is incredibly exciting, but it's also new, and maybe a little scary. So when we held our recent webinar to discuss the application of ChatGPT in customer service and introduce Fin to others, we weren't surprised to receive many thoughtful and important questions during our chat session. questions answers.

Find the webinar recording here

This area is brand new and evolving rapidly, so naturally there are still a lot of unknowns we are working on. What we do know is that Fin is an incredibly powerful tool, even in its current initial state. To tell you a bit more about where we are and what you can expect from Fin, we wanted to share some of the most important questions we received during the webinar - and the answers we offered.

> Main questions of the webinar Privacy, Security and Compliance

Q: How can companies with a certain degree of sensitivity be sure that Fin will not give an answer outside the bounds of the Help Center content, which could pose a security concern?

A: This is an essential part of the offer. This is something you can try out yourself when we have a demo version available for your help center. For now, here are the safeguards we have in place to ensure your AI chatbot provides the most accurate answers:

It only provides answers based on your Help Center articles. The AI ​​chatbot tries to disambiguate a question when it does not understand the end user's question. When the AI ​​chatbot doesn't know the answer, it will say it didn't find an answer, and then it can hand over to a human.

"OpenAI stores some of the data, however, they don't use it to train their models"

Q: How is the privacy of a customer's conversation with the chatbot protected? The client's input goes into the OpenAI system. What privacy protections are in place?

A: We always negotiate privacy terms. From now on, here's the scoop: in order to provide answers based on your articles, our AI chatbot will process your entire help center through OpenAI at regular intervals during the beta. During each conversation with Fin, we will send customer questions and comments verbatim to OpenAI. If your customer sends personally identifiable information (PII) as part of their message, those PII may be sent to OpenAI during processing. OpenAI stores some of the data, however, they don't use it to train their models.

Q: I heard that this feature will not be available in the EU. Is it true? Could you provide more information on this?

A: Fin will be available worldwide upon launch. However, there is no regional data hosting, so if you need EU data hosting, using Fin will not be possible. OpenAI does not currently offer hosting in the EU, so any personal information sent to them as part of their integration with Intercom services must be handled in the United States.

Q: Regarding the GDPR, what personal data is sent to OpenAI, and does OpenAI store any data in its system?

A: We will send customer questions and comments verbatim to OpenAI during every conversation with Fin. If your customer includes PII in their message, those PII could potentially be sent to OpenAI during processing. OpenAI stores some of the data but does not use it to train their models.

End Abilities

Q: Is Fin multilingual?

A: Fin is multilingual enough already. We have work to do to make Fin and the whole UX properly multilingual. This is a priority for us.

"Fin respects Intercom's article audience targeting and knows who the user is, so it can match the two things to only give relevant articles to the relevant user"

Q: How does Fin handle product changes i.e. answers change and change as products change and change?

A: Near real time! As soon as you update your Help Center content, Fin will adapt. There are no preset answers: they are always based on your live content.

Q: Can Fin collect data about his chats? Can it notify me if something unexpected happens?