Fin

I am working on new Fin Tasks where I want to create a branches with conditions that are dependent on the results of a custom action. Can anyone help me understand how to build this? For example if a Custom Action returns [X] then follow this path, and if Custom Action returns [Y] follow a different branch

Hi there,We’ve implemented Fin AI (Chat and Email) on our support desk, and it’s been fantastic so far—it’s doing a great job handling first-line queries! 😊However, we’ve noticed a challenge when a customer submits a vague message. Fin will respond with technically correct information, but it’s often not relevant to what the customer truly needs.I’m wondering if it’s possible to improve this by getting Fin to ask clarifying questions before responding. I’ve been advised to set up a new workflow where specific vague questions prompt Fin to ask follow-ups, but trying to predict and account for all the potential vague queries customers might come up with feels inefficient.Is there a better way to manage this? Any advice or insights you could share would be hugely appreciated!Thanks so much

Hi, The feature of "Fix grammar & spelling" is handy, however, when the selected sentence is incomplete (not ended correctly), it does add the missing word and/or punctuation marks to the end of that sentence (Expected); but, it also prefixes the missing word/punctuation to the top of the text!!!Example: It changes"I can assist you in this"to"matterI can assist you in this matter."

Would love to keep a thread going of what prompts are working for your teams, so we can generate some best practices across the community. 🙌🏻Share away!

Hi!I am a software developer working at Greenly.We are already using Intercom for custom support, for months.We recently saw that Fin will soon be available over an API (https://bfy.intercom.com/ “06 - Fin over API”Can you give more information of when will this be available please?Best regards,Maxime Broussart

We have a large database of typical customer questions and answers. These come from a service desk that is not currently supported by Intercom. What is the best strategy to add this content to Fin Knowledge?There are too many entries to manually input them as snippets.We’re considering adding them as a PDFs, but we’re not sure how Fin will chunk the data—whether it might split Q&As incorrectly or merge unrelated entries.Is there any guidance on how to format the PDF, or an alternative way to upload the data in a Q&A format?

Hello team! This maybe very normal question. Just would like to understand what are the differences between FIN AI and Chatbot please? In terms of how it can help to support customer service team We are looking to improve percentage of conversations which resolved by FIN AI. Also, percentage of improvement for conversations which closed by Chatbot as attached picture. To start with, I would like to get more understandings on how Chatbot can help the team to close conversation different from FIN AI please? Saw that percentage in report between FIN AI and Chatbot are separated. So I assumed that the way they work in the system on conversations are different? Moreover, if you have any suggestions/real practices about Intercom Chatbot improvement would you mind help to suggest. That would be really helpful on my end! Really appreciated :) Thank you in advance! Ploy

Our confirmed resolution rate is between 6-7% usually and our assumed resolution rate is near 60%. Looking at the conversations, I would estimate that 1/3rd at least of the assumed resolutions represent a correct answer and our clients are just not clicking the “Yes that helped” button that would mark it as a confirmed resolution. My experience with bots is similar, I rarely give bots feedback, because I know its not human. Any tips or methods that you have tried that increase clients use of the confirmed resolution? 

Hi,We have been using Intercom for our customer support for several months now and we are exploring AI automation solutions. Initially, we built workflows to handle basic support tickets, but we quickly found this approach to be unsustainable, as it became burdensome to maintain and scale.We also tried Intercom Fin, which we found to be an improvement over workflows. However, it is quite expensive, occasionally provides incorrect answers, and struggles with more complex tickets. With Fin charging $1 per resolution and our support agents resolving tickets at about $0.66 each, we decided to explore alternative solutions.I was tasked with identifying a viable option, so I’ve reviewed nearly all the apps listed under the Automation category in the Intercom App Store.There are approximately 140 apps in the Automation category, and we focused on identifying those that can automatically respond to customer messages. From this list, we found over 20 apps that either offer or claim to offer thi

Our deflection rate was 10-15 points below our average yesterday. It’s not the end of the world, but we’re typically very consistent.Was this a fluke or were there system bugs with Fin?

Hi all, I’m new here and trying to improve the FIN AI resolution rate. However, I would like to understand why the extracted data is always shows 0% or not higher than 2% if FIN AI Resolution rate is filtered with "Team currently assigned" but shows 27% for FIN AI Resolution rate of the company We do have our CS team across countries using Intercom chat. As the extracted data, there are two resolution rates that we’re looking at1.) FIN AI resolution rate of company as a whole2.) FIN AI resolution rate with filtered “Team currently assigned” each countries’ CS team However, the data extracted are always different. For an example, latest company FIN AI resolution rate is 27% but if I filtered the FIN AI resolution rate with “Team currently assigned” and see data of CS team in each countries. The data always show as 0% and never higher than 2% Would anyone have an explanation about this please that why it is different? I’ve read about the definition in Intercom but still unsure if my thou

Hi team,I’m looking into the possibility on FIN collecting information when a customer starts a live chat, and sets the Conversational Attribute to a certain value (‘I want to make a manual booking’).When this attribute has been selected, I’m envisioning that FIN will collect information from the customer and then pass over the compiled list of answers to an agent in the form of a ticket/a conversation, but with little need for the customer to reach out back and forth with the agent. For example:Customer starts chat Customer selects ‘I want to make a manual booking’. FIN answers, but because of Step 2, he has specific instructions to ask for the answers to a list of pre-determined questions (such as, ‘What is the name of the supplier, what dates, what room type, etc...’  FIN then thanks the customer and says that this will be booked for them ASAP, and we will reach out if we need anything else.For step 3, I am struggling to find where we would set up this list of questions that we know

FINs newest feature Guidance is now live in Beta? Are you using it?With such a new feature, that is possible the most powerful and complex, moved to Beta recently I’d love to hear how many of you are using it and how you are all finding it?Lets try share tops tips in the below, so we can all learn from each other.Examples of how you are using it What's working well What's not doing what you expect Anything else that is good to know

Is there a way to train FIN for more scenario based questions. Our customers have different use cases for our product and it would be very helpful if we can train fin to help answer scenario based questions. 

Hi! I'm currently in the process of setting up Fin for email. As we'd like to offer personalized (Human) Customer Support to our Users, I want to set up Fin for email for Leads. However, 56% of our emails sent by Leads are spam/marketing messages, and I don't want to be charged for all these emails.In related topics (How to avoid Fin trying to offer solutions to marketing emails and other spam?) I see that other users tag these emails and exclude them from Fin's audience. However, I'm not sure how to identify these emails as such so I can set up my workflows and give them a tag. Does anyone have some advice? 🙏 Thanks in advance!

Can Fin read data from my excel sheet or Google sheet and answer user question? I tried to add the google sheet as webpage for Fin content, but it shows error.

We’re trying out Fin for email conversations. One issue that I imagine is not unique to us is that we get quite a lot of marketing mails to our support@ and hello@ emails. And now when we do, Fin treats that as a customer needing some help. I’m not sure this is a huge issue, other than the fact that Intercom will charge us for each such case since the Fin response will go unanswered and eventually he’ll close the case and considered it “assumed resolved” 💸Has anyone figured out a good way to manage this kind of inbound “spam” that ends up with Fin?

We sometimes receive auto-generated structured emails to our support inbox.These emails represent a request to doa. specific action for a cusotmer, not a gerneral qurestion.How can we instruct Fin to ignore specific emails which can be identified from fixed templated body team, and will also include a fixed text string in every subject line?

Hello. In the AI & Automation > Fin AI Agent > Setup > Content the limit for knowledge sources is 10. Aside from manually copying and pasting each link, is there a way to combine multiple knowledge sources with the same domain name into one? 

Hello! I’m starting to use Fin AI to answer more queries. At the moment, it’s only involved in some kinds of conversations and we’ll expand it to other topics gradually.My question is, does Fin process the buttons the user selected when looking for an answer? In my example below, would Fin process that the first message was “Help with my subscription”, and therefore look for the answer in subscription-related knowledge bases? Example of workflow Thank you!

Hello! I’m working on integrating Fin AI to our queries handling.When conversations are handled by a teammate, we use a custom attribute “Subject Line” to ‘tag’ the query and report on volumes each week. Subject Lines would look like this:[Account] Login issue, [Account] Change Email Address[Billing] Refund Request[General] Query / Premium, [Feedback] Contentetc etc If I want Fin to handle tickets, how can I get it to tag the conversation based on its content? either via our Subject Line attribute or via tagsAt the moment, we have a workflow where users select what they want to talk about with buttons (“Question about my account”, “Help with my subscription”, “Share feedback”, “Something else”). Subscription queries go through Fin, the rests goes to a teammate. I’d like to open “Something else” to Fin AI too but I’d need to be able to tag the conversations properly to create usable data. What’s the best practices here? What have you found the most helpful?

I’ve been testing Fin and I haven’t been satisfied with the responses just yet to turn it on for all users. I understand it’s based on our articles and related knowledge base, but I’ve found Fin to provide wrong or sometimes unhelpful information. Here are some examples:It frequently tells users to contact support to fix the issue (Isn’t this what people think they’re doing?) It get company branding incorrect. For example I work for a company called “G.J. Gardner” and Fin often defaults to “GJ Gardner”, missing the periods in the name.  Users often don’t provide enough info such as the record name in the software they want us to troubleshoot (i.e. a customer name or project number). How can I train Fin to inquire about this information? Fin defaults to US English. While this isn’t always bad, we have some industry specific terms that differ between contries (i.e “variation” in Australia and “change order” in the United States). How can I train Fin to use the correct terminology when re

1- I want away to add context to fin, more of general guidance of how to best reply; stuff like:Your answer has to be no longer than x lines. Avoid using the following terms. Avoid adding resources to the end of the msg.2- Also, my fin is taking to long to answer, how can I fix it? 

I have a CSV list of 5000+ frequently asked sales tax questions and their corresponding answers. I would like intercom to ingest these questions surface the answer to customers. I would also like to be able to update these questions quickly over time as the answer changes. What is the easiest to ingest these into intercom? Should I use something like Confluence instead of CSV? Any advice? 

When creating public articles, you have four components that can help get them discovered:Title Description Body CollectionAI Agent will ingest this content and attempt to provide a best-fit answer based on its knowledge.However, if you want multiple queries to lead to the same article, it seems there isn’t a direct way to accomplish it other than including wording throughout the title, description, and body such that alternative queries may pick up on that language. ExampleSuppose we have an article:Title: How do I add a payment method? Description: Learn how to associate a card with your account.That’s straightforward enough and satisfies a couple ways that queries might go (“add a payment method,” “associate a card with your account”).But, what if I want these other queries to lead to the same article:How can I pay? Where do I go to make sure you can charge me? I received a notice for non-payment. What do I need to do? I’m seeing an error for collecting an invoice balance. How do I