- AI is fundamentally transforming customer service, moving from a supplementary role to becoming the core future of businesses like Intercom.
- Intercom's product "Fin" automates repetitive customer support queries, significantly improving efficiency for both businesses and end-users.
- Fin leverages advanced language models, specifically Claude, chosen for its accuracy, trustworthiness, and strong multilingual capabilities which unlock global reach for AI products.
How Intercom is redefining customer support with Claude
- AI is poised to completely change customer service by automating repetitive tasks that humans find boring and concentration-draining.
- Intercom developed "Fin," an AI product designed to quickly and accurately answer frequent end-user support questions.
- Fin operates by looking up relevant information within a business's existing customer support knowledge base to generate replies.
- The primary goal of AI in customer support is to enhance efficiency and save time for both the business and its end-users.
- Intercom employs rigorous A/B testing with millions of user interactions in production to evaluate and select the best-performing AI models for their products.
- Modern language models offer an underappreciated benefit in their robust translation capabilities, which is crucial for achieving global reach (Fin supports over 45 languages using Claude).
- Adopting AI solutions like Fin can lead to significant business success, exemplified by achieving eight-figure revenue in its first year.
AI — Artificial Intelligence; the simulation of human intelligence processes by machines.
AI models — Computational algorithms trained on data to perform specific AI tasks, like natural language understanding or generation.
customer support knowledge base — A centralized repository of information, articles, and FAQs that helps customers find answers and support agents resolve issues.
end-user — The ultimate consumer or client for whom a product or service is designed.
A/B test — A method of comparing two versions of a single variable to determine which performs better in achieving a specific goal.
production — In software development, refers to the live, operational environment where software is used by real users.
language models — AI models specifically designed to understand, generate, and process human language.
translation layer — A component or process responsible for converting content from one language to another, often within a larger system.
This is one of the defining technological revolutions of our lifetime. My name is Fergal Reed. I'm VP of AI here at Intercom, which is in useful, sunny Dublin today. At Intercom, we build software for customer service teams to talk to their customers. We've been building AI products in customer support for six or seven years, but there were always part of our business, whereas now AI is the future of our business. Customer support customer service is going to be completely changed by AI, and that's just become the most important thing for our business. There are tasks that humans are good at, and tasks that require a high level of empathy, maybe a high level of judgment. And then there's other tasks that AI models like Claude are better at. If you have a human customer support representative, and they are answering, literally, the same question for the 15th time today, it gets boring. They start to lose concentration. So, yeah, we've built this product called Fin. And what Fin does is it answers those repeated end-user customer support questions and does it quickly and accurately and without anyone needing to wait around. Fin will look up all the information in the business's customer support knowledge base. I don't know, give them a reply. Where do we try to save time for the business? We're trying to make them more efficient, but we're also trying to save time for the end-user. On trial, I could really say that I had to try and build a model. You know, it has a lot of the same values we care about. They want to make it trustworthy, they want to make it accurate, and we care deeply about that as well. And when they came out with Claude's son at 3.5, we started to use it, and we were really like, wow, this is a very impressive model. When we get a model that looks impressive, we go and we tend to test it in production. And we did a massive AB test with literally millions of end-user interactions. And Claude was the winner of that test. That's why we chose it to be perfect. An underappreciated benefit of reading modern language models like Claude is the extent to which they can handle translation between many different languages. And that really unlocks global reach of AI products. So Fin, for example, works in over 45 languages from Chinese to German to Japanese. And we really rely on Claude to kind of handle the translation layer there. Fin is being an amazing, successful product for us. So we hitched eight figures in revenue in our first year. And we're really delighted with Fin's progress. Absolutely, and Robert, come help to attend. Yeah.
TL;DR
- AI is fundamentally transforming customer service, moving from a supplementary role to becoming the core future of businesses like Intercom.
- Intercom's product "Fin" automates repetitive customer support queries, significantly improving efficiency for both businesses and end-users.
- Fin leverages advanced language models, specifically Claude, chosen for its accuracy, trustworthiness, and strong multilingual capabilities which unlock global reach for AI products.
Takeaways
- AI is poised to completely change customer service by automating repetitive tasks that humans find boring and concentration-draining.
- Intercom developed "Fin," an AI product designed to quickly and accurately answer frequent end-user support questions.
- Fin operates by looking up relevant information within a business's existing customer support knowledge base to generate replies.
- The primary goal of AI in customer support is to enhance efficiency and save time for both the business and its end-users.
- Intercom employs rigorous A/B testing with millions of user interactions in production to evaluate and select the best-performing AI models for their products.
- Modern language models offer an underappreciated benefit in their robust translation capabilities, which is crucial for achieving global reach (Fin supports over 45 languages using Claude).
- Adopting AI solutions like Fin can lead to significant business success, exemplified by achieving eight-figure revenue in its first year.
Vocabulary
AI — Artificial Intelligence; the simulation of human intelligence processes by machines.
AI models — Computational algorithms trained on data to perform specific AI tasks, like natural language understanding or generation.
customer support knowledge base — A centralized repository of information, articles, and FAQs that helps customers find answers and support agents resolve issues.
end-user — The ultimate consumer or client for whom a product or service is designed.
A/B test — A method of comparing two versions of a single variable to determine which performs better in achieving a specific goal.
production — In software development, refers to the live, operational environment where software is used by real users.
language models — AI models specifically designed to understand, generate, and process human language.
translation layer — A component or process responsible for converting content from one language to another, often within a larger system.
Transcript
This is one of the defining technological revolutions of our lifetime. My name is Fergal Reed. I'm VP of AI here at Intercom, which is in useful, sunny Dublin today. At Intercom, we build software for customer service teams to talk to their customers. We've been building AI products in customer support for six or seven years, but there were always part of our business, whereas now AI is the future of our business. Customer support customer service is going to be completely changed by AI, and that's just become the most important thing for our business. There are tasks that humans are good at, and tasks that require a high level of empathy, maybe a high level of judgment. And then there's other tasks that AI models like Claude are better at. If you have a human customer support representative, and they are answering, literally, the same question for the 15th time today, it gets boring. They start to lose concentration. So, yeah, we've built this product called Fin. And what Fin does is it answers those repeated end-user customer support questions and does it quickly and accurately and without anyone needing to wait around. Fin will look up all the information in the business's customer support knowledge base. I don't know, give them a reply. Where do we try to save time for the business? We're trying to make them more efficient, but we're also trying to save time for the end-user. On trial, I could really say that I had to try and build a model. You know, it has a lot of the same values we care about. They want to make it trustworthy, they want to make it accurate, and we care deeply about that as well. And when they came out with Claude's son at 3.5, we started to use it, and we were really like, wow, this is a very impressive model. When we get a model that looks impressive, we go and we tend to test it in production. And we did a massive AB test with literally millions of end-user interactions. And Claude was the winner of that test. That's why we chose it to be perfect. An underappreciated benefit of reading modern language models like Claude is the extent to which they can handle translation between many different languages. And that really unlocks global reach of AI products. So Fin, for example, works in over 45 languages from Chinese to German to Japanese. And we really rely on Claude to kind of handle the translation layer there. Fin is being an amazing, successful product for us. So we hitched eight figures in revenue in our first year. And we're really delighted with Fin's progress. Absolutely, and Robert, come help to attend. Yeah.