Off Script: Into the future with AI-first Customer Service

AI has ushered in a fundamentally new way of doing customer service, and there’s no going back now.

This is what we’re exploring in Off Script, our new series of candid conversations with Intercom leaders about the extraordinary technological shift being driven by artificial intelligence.

Every major technological breakthrough, from the Industrial Revolution to the invention of the internet, has fundamentally altered the way we live, work, and interact with the world around us. We believe AI will do the same.

We’re only a year and a half in, and yet we’ve already seen major transformations in nearly every industry. Customer service is no exception. AI happens to be incredibly good at the tasks customer service agents typically do, such as answering questions at scale, analyzing and retrieving information, disambiguating questions, and troubleshooting.

“It’s not about sprinkling some AI on top of your existing setup – it’s about embracing a whole new system”

We’ve witnessed firsthand just how AI can boost your customer service. ​​Since the launch of our AI agent, Fin, last May, we’ve seen our average resolution rate climb from around 28% to nearly 50%, with some customers experiencing resolution rates of more than 70% of their inbound requests.

This transformation is happening at a faster pace than even we could have predicted, and it’s clear we need a completely different approach to tackle the challenges ahead. It’s not about sprinkling some AI on top of your existing setup – it’s about embracing a whole new system, one that’s cheaper, faster, more effective, and that will help you deliver remarkable customer experiences. We call it AI-first Customer Service.

In the second episode of Off Script, our Chief Product Officer Paul Adams talks about how AI has raised the bar for what great customer service looks like and what it takes for support teams to take advantage of the possibilities.

Here are some key takeaways from the episode:

  • There are three main components in AI-first Customer Service: an AI Agent for customers, an AI Copilot for agents, and an AI Analyst for managers.
  • The AI Copilot can help the agent fix the issue by looking up information, analyzing conversation history, and crafting the best answer for that customer.
  • CS leaders can use AI to tap into insights that can help them improve performance, train new agents with real-time feedback and suggestions, and drive customer satisfaction.
  • The AI is only as good as the data it feeds on. For this approach to work, you need a data and knowledge layer that contains both internal and external data and the right content.
  • By centralizing operations into just one platform, you can benefit from a seamless feedback loop that helps the AI get better, faster.

We publish new Off Script episodes on the second Thursday of every month – you can find them right here or on YouTube

What follows is a lightly edited transcript of the episode.


Off Script: Episode 2
Paul Adams on AI-first Customer Service

Eoghan McCabe: AI is finally, really, actually here, and it’s going to reshape society in ways I doubt we can imagine. It’s clear that some areas of the economy are already primed for epic levels of disruption, and customer service is one of these. AI will deliver the holy grail of customer service. Instant, consistent, excellent support for every single customer, 24/7.

This is not just some abstract future we’re building towards. It’s actually happening today. We have customers who are having over 70% of their inbound support requests answered by our AI agent, Fin. And that’s all while maintaining their historical levels of customer satisfaction. Over the past year, as we’ve built out our AI platform, we’re seeing there’s a fundamentally new way of doing customer service.

We call it AI-First Customer Service. And this is the future. Paul Adams, at Intercom, who’s been building out much of these ideas in this second episode of Off Script, is going to lay this out for us in a way that only Paul can. Enjoy.

The waves of innovation

Paul Adams: Technology only moves in one direction. Once you go through these before/after moments, you never go back. Oftentimes, there is some breakthrough moment early in a technology cycle everyone can connect with. For example, the iPhone came out, and it just changed everyone’s perception of what’s possible. It was just like this before/after moment. With AI, ChatGPT came out, and people could use it. It was easy to use. It’s like using Google. And when you go through these before/after moments, you don’t go backward. People who got iPhones didn’t go back to Blackberries. Blackberry went out of business. Once you’ve gone through that door, you never go back.

Today, we’re here to talk about AI for customer service. But to really understand that, you’ve gotta zoom way out and look at waves of technological change. This is a well-studied thing, and it’s not complicated to understand. Technology develops in these waves. At the start of the wave, it’s early, and not a lot of people are using the thing. Some people are cynical or skeptical of it. Then, it’s like mass adoption, and then the innovation era ends, and a new wave starts. Throughout history, we’ve seen the Industrial Revolution, and things like electricity, the internet, and mobile phones.

“You can go back as far in history if you want, the pattern is the same every single time. The question is, how big will it be?”

All of these things reshaped society. Prior to the Industrial Revolution, things were manufactured by hand. And suddenly, you could mass produce things, which led to factories. And that changed how people worked, how people lived, and it totally changed the economics of what was possible.

If you go back a hundred years, people were literally riding around on horses. Then the car showed up. People didn’t really think that much about it; they were cynical and skeptical of the car. But it took over, and it led to suburbia. It led to Walmart. It led to McDonald’s. And you can debate whether suburbia and McDonald’s are good or bad for humanity. But it totally changed. If you look at Earth from space, the car totally changed what the planet looks like.

AI is going to do the same. You can go back as far in history as you want. The pattern is the same every single time. The question is, how big will it be? It’s certainly as big as the smartphone. I think it’s certainly as big as the internet. And I think there’s a very good chance it’s as big as the Industrial Revolution.

From mundane to memorable

We speak to customer service leaders every day of the week, and we try and understand them, learn what they really care about, and all sorts of things come up. They care about things like operational efficiency, they care about the happiness and retention of their team, and the thing they care most about, by far, is the customer experience. The experience of their customers.

It’s incredibly hard to deliver a really good customer experience at any kind of scale. If you’ve got thousands of customers, tens of thousands, millions of customers, giving them a great experience consistently every single day is this nirvana, holy grail type state that no one ever achieves. So, this idea that customer service can be instant, accurate, fast, and engaging is a night and day difference from anything we’ve experienced before. And so, there’s an incredible opportunity for companies to change how people think about them through incredible customer service. The interactions people have with customer service are typically when you have a problem, not when you’re delighted with the business. And so, customer service becomes this really critical brand-building opportunity.

Intercom is a technology company. We’re 12 or 13 years old. And so, you would think that a lot of our customers are modern technology companies too. But we have customers who are 100 years old. Banks, utility companies… Some of our most pioneering customers, those who are leaning most into AI, the company is 100 years old.

Customer service can totally change how you think about a company. You go on to a utility company to pay your electricity bill or something like that. It’s not really something people get excited about. They don’t wake up and think about paying their electricity bill. But then they do it, or they have a question, and suddenly they’re interacting with this cutting-edge, next-generation AI, and it changes how they think about the company. Suddenly, they’re thinking, “Wow, this company is incredible. Not only are they at the forefront of this, but they’re also 100 years old. How did they even manage to do that?” It’s a huge opportunity for businesses to change how people think about them.

“These big technology waves take anywhere between 5, 10, 20, 30 years. We’re one year in”

A lot of people say that customer service will be one of the first industries to be impacted by AI. We think that’s true, and it’s mainly because AI happens to be incredibly good at the things customer service agents do. AI can already do things that many knowledge workers can do. It’s really good at doing things like retrieving information, analyzing information, generating insights, answering customer questions, pulling data from the customer record, conversing back and forth, troubleshooting, disambiguating questions… It can do this conversationally, and it can do it incredibly accurately already.

These big technology waves take anywhere between 5, 10, 20, 30 years. We’re one year in. It’s already doing incredible things, and it’s just going to get better. It’s going to be able to do more and more things. It’s going to be able to start taking actions and doing things for customers. Today, it answers questions. It’s removing all the repetitive questions that customer service agents have to answer, which, by the way, is the thing that makes the job not fun. The AI agents are going to do all of these things. It’s going to free up the human customer service agents to do more interesting things.

But AI is not just good at answering the questions customers have, it’s also really good at a lot of the things that support managers have to do. Support managers have to look at the overall system, look at the performance of their team, and think about things like: Are customers getting a good experience? Are they answering questions? Are the questions being answered accurately by people? So, what AI has done is it has given us a new way to do customer service. And that new way is way better.

AI-first Customer Service

Right now, there is a lot of AI news, noise… pick your word. And in a lot of cases, what businesses are thinking is they need to add AI to what they do today. That’s not how we think about this at all.

We think this is a complete mindset change. It’s not just a product change. The new types of products being built in customer service are unlike anything we’ve seen before. It’s not just that, it’s a mindset change. And the mindset change is one where you need to think about AI first. The way to think about this isn’t adapting and tweaking and iterating your existing customer service setup. You should push the old way to one side and embrace a new way of doing it. A completely new mindset. We describe it as AI-first Customer Service.

When a customer shows up and has a question, AI answers first. You’ve got to set up the system so that AI answers first. Why? Because it’s typically better. It can answer the question more often than not. It can do it 24/7. It can do it in any language. It’s instant. It can do things that the best human agent team in the world wouldn’t be able to do.

“The AI can look at all conversation history and customer satisfaction across all those conversations. So, for any individual question, it can pick out the best answer”

But not just that. When the AI agent can’t answer the customer’s question, and when it does end up with the human team, which will happen, the team also must be AI-first. And what that means is they have an AI Copilot with them in the inbox. So when they’re answering questions or trying to work through tickets, the AI Copilot is there all the time helping them. If you don’t have an AI Copilot, your job’s harder. You’re looking up information and manually crawling through stuff. Whereas, if you give the AI Copilot access to all of your internal knowledge, all the external knowledge, and all the data, the AI Copilot is really good at picking apart all the different pieces. It can do it at a scale that a human just can’t. It can analyze at rates that are just far faster across a much broader set of information and suggest things to the human agent. It’s not going rogue answering customer questions. It’s saying to the human agent, “Hey, I’m here to help. I’m your Copilot.”

Something that the AI Copilot can do that’s incredible is it can look at all conversation history and customer satisfaction across all those conversations. So, for any individual question, it can pick out the best answer. Not just a right answer, but the very best answer.

When customers get these types of answers, their satisfaction increases. They’re happier. You can even look at things like retention. They retain better, grow more, and use the product more. For experienced agents, that may not be that big of a deal. They are good at their job and typically know the best answers from experience. But think about a new agent who’s just joined the team. The ramp time with new folks can be long. Staff turnover can be high, and a lot of managers spend a lot of time just trying to train new folks. ​​They don’t need to anymore. They can train the Copilot, and the Copilot will train the new team member.

The first part of this is that customers have an AI agent answering questions for them instantly, 24/7, in any language. The second part is that support teams have a Copilot helping them 24/7, suggesting the very best answers to questions to help them give great customer experiences. And the managers, leaders, and people who work in support ops are the third part of this. They also have AI, a buddy who’s able to answer all their questions. Any manner of questions around performance, operational performance, team performance, accuracy of answers, or customer happiness.

Today, if a manager or a support leader wants to understand how good their customer experience is, they rely on things like CSAT. They rely on things like asking customers at the end of a conversation, “Hey, did we answer your question?” Very few customers actually fill in those surveys. They don’t really answer them en masse. So, anywhere between 10% and 20% is what you typically get. This means that 80% or 90% of your customer experience isn’t measured at all.

AI can actually measure 100%. The AI can actually read and parse every single customer conversation. Even when customers don’t reply back and say if they are happy or not happy, AI can figure that out. It has all of the conversation history. It can figure out not just whether the question was answered accurately, but whether people were happy. It can pull out tone and sentiment. It’s really good at this kind of stuff. And so, for the first time ever, they’ll be able to measure 100% of it.

One seamless platform

If you want to do AI-first customer service, you must have the three components. You must have the AI agent for customers, you must have the AI Copilot for your human support team, and you must have the AI analyst and AI insights for leaders and managers. You must have all three components to properly do AI-first customer service.

These three parts sit on this data and knowledge layer. And this data and knowledge layer is what the AI learns from. And so, that data and knowledge layer needs to be really, really good. You have the data and customer records, and it can connect into all sorts of different types of places, whether it’s your CRM, other types of data sources you use to power customer information, or internal systems you’ve built. It can connect to all of these data systems.

And then, on the knowledge side, you have the content. It could be externally-facing content that’s approved and accurate, like a help center article, for example, but it also can connect to internal knowledge.

“The Copilot-powered answer delivered via the human agent teaches the AI agent. The system just gets better and better”

The data and content power the knowledge. And this layer sits below the agent, the Copilot, and the analyst. So, for example, the AI agent, when it can’t answer, hands over to the human team with the AI Copilot. Then, they can answer the question, and that teaches the AI agent what to answer next time. The Copilot-powered answer delivered via the human agent teaches the AI agent. The system just gets better and better. There’s this beautiful loop. And the manager has the AI analyst and the AI insights to help the system. It’s one seamless system.

So, you have to have all three, and they have to have this beautiful feedback loop that makes it better and better. This is only possible if it’s built in the same system. If you have a legacy system, acquire an AI company, and try and integrate those two things, that data and knowledge layer sitting in the middle is a mess. It’s a mess of different data formatting and different data types and even different types of platforms these things have been built on. You need everything on one platform.

What’s amazing about AI systems is that they teach themselves to get better. If you have everything in one system where the data is clean and clear, the AI can get better faster. So, AI-first is the mindset people need to adapt.

Towards emotional intelligence

How good is AI at communicating with humans? You need to look at a scale. At one end of the scale is transactional communication. I just need information back and forth. And over here is emotion. I’m angry, or I’m sad, or I want to celebrate. Human communication spans the whole thing. We don’t go around in a highly emotional state all the time. Actually, a lot of the time, humans are very transactional. AI is incredibly good at transactional communication. Even if you think about customer service, most customer service queries are quite transactional. People are just looking for information. That’s why AI is so good at it.

Over here is emotion. And if you tie this to customer service as an example, sometimes people are angry, you know? They feel like a company they love has let them down. Or they really needed something, they bought it from this company, and it didn’t work out. Sometimes, people just need to vent. Sometimes, you just need an ear. And AI can’t be that yet. Now, could it be in the future? Technologically, I think it’ll get close. And then, culturally, will people accept that? Who knows? We’ll have to wait and see.

“People are looking at the movie ‘Her’ as being prescient … It’s emotional communication”

AI is certainly moving along this spectrum very fast, and it’s getting pretty close up here. There are lots of companies building in this space of emotionally tuned-in AI, and I think it’s possible. A lot of people are looking at the movie Her as being prescient. It’s emotional communication. So, who knows? This is highly speculative. But AI is definitely moving along this axis, and a lot of human communication happens down here.

When people hear that AI is going to do more and more work, that AI is going to take over a lot of the knowledge work we do today, it can be scary, but it’s actually incredibly empowering. It frees up all of us to have more time. And with more time, we can do more things. We can be more creative with our time. We can deliver those empathetic situations with higher degrees of empathy. We’re not rushed. AI is going to take away a burden that’s kind of been put on us by recent technologies and give us more time. I think it’s going to be an incredibly positive thing, and people are already starting to see that in different ways.

A new era for Customer Service

Here’s what I think about the timeline. We’ve tried to predict how fast this would happen internally before, and it’s happening faster than we thought, and we thought we were being ambitious and optimistic and so on.

If you look at Fin, our AI agent, we launched it a year ago. And if you look at the average resolution rate, we started out in the 20s when we launched it, and then we started adding things to it. We added things like multilingual Fin, the ability for it to disambiguate questions and go back and forth, the ability for human agents to help it get better and teach it things when it couldn’t answer, and we’ve improved our AI engine underneath it all. So, we take the large language model, tune it, refine it, change it, and do a lot of work at that level. And just in the last year, it went from 20s and 30s, and we’re now in the 40s, heading to 50%.

Suddenly, for a support team using our Fin AI agent, half of their questions are now being answered accurately by AI. And customer satisfaction hasn’t gone down. Actually, at times, it’s gone up. And this is just the start. These are transformational results, and we’re like a year in.

“I think it’s inevitable that all businesses are going to go AI-first, and that all customer service teams will be AI-first. It’s just a matter of time”

And so, where are we going next? I don’t know, but I can say with a lot of certainty that customer service is going to radically change. I think in three or four years, we will look at the customer service of today, and certainly of a year ago, and we’ll think it’s incredibly old-fashioned and archaic. I think it’s inevitable that all businesses are going to go AI-first, and that all customer service teams will be AI-first. It’s just a matter of time.

What matters most is that customers of these businesses are getting great experiences. That’s what all this boils down to. I would highly encourage customer service leaders everywhere to get into this world of AI-first and start trying it. I think you’ll find when you try it that the results are far better than you ever could have expected, and the technology is getting better faster than you ever could have hoped for.

Off Script Episode 7 Brian Eno and Stephen Fry