Pioneer 2025 Product Keynote: Fin 3 and Customer Agent

Thank you, good morning and welcome to Pioneer.

Thank you, thank you.

We appreciate you being here, here in New York City.

Those of you who are live on our stream, people who watch later.

And thanks also to our title sponsor, AWS, and all of our sponsors without whom this overly extravagant production would be even more expensive.

So we call this Pioneer.

As I explained last year, because that's how we think of you guys, you're pioneers in your fields, bringing us to this inevitable world where all of our service operations, customer operations are run by agents.

We, of course, think of ourselves as pioneers.

We were first in this space a whole year ahead of everyone else by shipping our service agent, Fin.

And today, Fin is...

the largest by customer count, we think by revenue.

It's number one in our internal benchmarks against our competitors.

It beats our direct competitors in the performance benchmarks that our peers, or rather our prospects run, and it's number one on G2.

Thank you.

So our promise to you is essentially that for as long as you want to continue to push out ahead of your competition and be the first amongst your peers, that we'll be here for you, we'll be here to deliver you the first innovation and the newest technology in this space out ahead of everyone else.

And so today, in October 2025 in this little moment in time, we want to show you what that looks like.

We've got some major product announcements coming extremely shortly, right after me.

And we're going to tell you a little bit about what's coming after that.

So to kick us off, I'd love to welcome to the stage our Chief Product Officer, Paul Adams. Hey thanks Eoghan.

Good morning, everybody.

It's fantastic to be here in New York City.

Thanks for coming and thanks to everyone tuning in online too.

We have a lot to show you today, a lot of new product updates.

And I want to start by kind of reminding us we live in quite a remarkable time.

Like I'm in this game too long and I've seen nothing like AI before.

The technology is just mind blowing.

It's changing all sorts of industries and it's already transformed customer service in unimaginable ways.

We're fast moving from a world where customer service had to worry about head counts and staffing and the never ending battles against volume.

And nowadays we can deliver instant, accurate, reliable, consistent service 24/7 in almost any language.

It's a total change, a total transformation.

Since the beginning of Intercom, we had this mission.

Our mission was to make internet business personal.

And The way we do that, the we try and do that at least, is to give all of you the tools so you can deliver this concierge level service to every single customer, every single time.

That's our goal, that's why we build what we build.

And that service that feels very individual to people, very personal to people, you know who they are, you know what they need, you can deliver that to them, you can do it quickly, efficiently, accurately, reliably, and that way you can grow your business faster.

And what a lot of customer service leaders that we speak to, aspire to is to really help grow their business.

to help you do that, today we have a big announcement, a big product launch.

We're announcing Fin 3.

So a huge step forward.

Fin 3 is our latest big upgrade to Fin.

It's extended our performance lead that Eoghan talked about.

It now does a lot more work by answering these deeper, more complex queries.

And we've also got new channels.

Fin is the most omnichannel agent in the market.

So before I get into all the details of Fin 3, I'm gonna show you demos.

loads and loads of product stuff.

I wanna first have us reflect on how far we've come and how quickly the change has happened.

So if you've got just two and a half years, March 2023, we launched Fin 1, which was then called Fin.

It was built on GPT-4, which was released the same day.

We were very early.

It was the first agent for customer service and it was focused on informational queries.

So you'd hook up your knowledge base, your content, and Fin could do all your frontline support.

That had huge customer impact alone.

Suddenly people were able to think about transformation, think about how their org might work and operate differently, think about new, better ways they could deliver a really, really great customer experience.

And we quickly learned from all of you, you were the pioneers of Fin, many of you, we got a lot of product feedback, thank you for that.

We learned very quickly that this is not a small product, this is actually a big product, and it's a broad product, it's a deep product.

We had to build things like reporting.

better ways to manage your knowledge, better ways to test Fin, new channels for Fin, Fin in every channel that you use.

So we relentlessly shipped product updates from day one.

And we saw this amazing thing happen.

We saw that the resolution rate of Fin, which started out at 23%, over the course of the first year or so, grew to 52%.

And this 52 % is the resolution rate on average across our entire customer base.

This is not a cherry pick number.

This is the average of our thousands of active Fin customers.

And this was one year ago.

One year ago we launched Fin 2, and today's Fin 3.

We launched it Pioneer last year, exactly one year ago.

And the question back then, the frontier of our industry then was, could these AI Agents deliver human quality service?

Like could they really follow your brand voice, follow your guidance, your guidelines, be a good representative for your company?

Could they connect to third party systems, internal systems, use this personalized information?

And the answer was yes, they could.

Fin 2 was a significant upgrade.

Fin 2 acted like your best human reps, and it could perform as well as them on a whole bunch of tasks.

We then realized that if we're gonna have this big product, this transformational customer experience, we needed to design a system where you could manage Fin.

And we built this thing, we call it the Fin Flywheel, we built it a year ago or so, and it's a continuous improvement loop.

We learned that this product is actually really a system, and it starts with training Fin.

Training Fin in your knowledge, your behavior, then testing those changes.

We learned very, very quickly that you can't just like roll out or YOLO out all these changes.

You needed to know these things would actually work the way you intended to.

So we have to build a big testing suite.

You know, deploy Fin on different channels or with different target customers, different segments, things like that.

And the last step then is analyze where you can measure performance with Fin and measure its scale and see what's really happening.

And the beautiful thing about the Fin Flywheel is that the more you use Fin, the better it gets.

That's really important.

The more you invest in these four steps, the better Fin gets.

So since Fin 2, which we launched a year ago, we grew our resolution rate from 52% and it's grown all the way to 66% today.

So we've over 6,000 Fin customers and the average resolution rate across those 6,000 customers is 66%.

I think this chart is astounding.

Like I've seen many a good chart in my career, but this one continues to amaze me.

Look at the line.

The line is just consistent, continuous improvement.

And this chart, if you look at the resolution from early 20s, now it's 66% on average.

By the way, a fifth of Fin customers get over 80%.

So this is like the average.

This chart is incredible, and I don't think any other company has a chart like this.

This chart is 1% resolution rate increases pretty much every single month without fail.

It's an incredible thing to look at.

It's because of all these product improvements.

It's got these improving resolution rates across over 6,000 customers, like I said, but also over many, many millions of resolutions.

We're looking at here consistent improvement across millions and millions of real customer conversations happening in the real world.

I do want to take a moment to say thank you to all of you.

I said earlier, you give us a lot of feedback.

I want you to keep doing that because without your partnership and your hard truths, reality bats last, as they say, we can't make this better.

So our product and engineering team are here in the building, plenty of them.

Find them, hunt them down, give them all your feature requests.

They'll, I hope, promise you, over promise and over deliver maybe.

So where are we now?

When we dig into the...

resolution rate increases and try and study what happens across our very diverse customer base with big companies, small companies across all different industries.

em It's not just the Fin Flywheel that's driving this change.

fact, there's another, aside from like the Fin Flywheel, it's an important concept.

There's another really important thing to understand about building AI Agent products.

And that is that there's multiple layers to the product.

The app layer, the AI layer, and the model layer.

There's three layers and it's important to understand these things.

The app layer is what you're familiar with.

It's the Fin Flywheel.

It's the features that you use every day, train, test, deploy, analyze.

It's the user experience within your customer's guests.

So Fin over WhatsApp.

Fin and all the different channels use email.

All the channels Fin works in.

That's the app layer.

You're most familiar with it.

You use it every day if you use Fin.

The layer below that is the AI layer.

This is the RAG system of Fin.

This is a system, it's quite complicated, as you can see.

This is a chart from our AI group.

And we didn't try to make it look prettier.

This is, you know, and this is like a gross simplification of what's going on.

But this is like a really, really complex system.

We've it at the start of Fin, the RAG system.

We were, think, one of the first kind of in-production, out-scale RAG systems in the world.

And the RAG system now, we think, is the best for customer service in the world.

That's what kind of our data benchmarking tells us.

And in this system, in this kind of layer, a lot of things are happening.

Fin is understanding customer context, by pulling different types of content.

trying to use a sophisticated search system to pull back the right kind of content that might answer the question, applying your guidance, your policies, all the things you've told Fin to do and not do, and then avoiding hallucinations.

And we improved this, kind of relentlessly improve it.

There's a big team, very talented AI group working on this, by hundreds of A-B tests.

We just like big experimental mindset where we're running these A-B tests constantly at all different parts of the system to try and work out how can we optimize it, how can we make it better.

But the AI layer isn't the only thing.

We also have a model layer.

The model layer is what sits underneath the AI layer.

This is the underlying LLM layer.

So all of you are familiar with GPT and Claude and different models like those.

The RAG system relies on the model.

Now the thing about models is GPT and Claude, know, built by amazing companies, and Anthropic are here today, like an amazing company and Claude is an incredible product.

These are general models.

So these models are like, you know, on kind of general information, trained on the internet.

They're not specialized for customer service.

They're not trained on customer service data.

And we all know, but living this world every day, there's a lot of detailed nuance and a lot of complexity in customer service.

And so we started to ask ourselves, you know, is there a better way to do this?

What if we build custom models?

What if we trained models on all of our millions of customer service historical conversations?

This huge data set we have.

So we started this as an experiment and just two or three weeks ago, We had an event in San Francisco.

We announced the first five for custom models.

They are purpose-built for customer service.

They outperform general models, which is an amazing thing for us to see.

And it just gives us conviction that this model layer is something that you have to invest in deeply.

If you want to build the highest performing product, you must have a great app layer, a great rag system with the AI layer, and you must have custom models.

That's how we together, these things, are how we deliver this amazing Fin performance.

And the reason we're doing this, we invest a lot of our resources into the AI layer and the model layer.

And lots of you don't see it necessarily.

And it's hard to understand.

I certainly do not understand the details of the layer or the model layer.

We're doing it.

We're making these huge investment because we believe the highest performing product will win.

We just think the highest performing product will win.

At the end of the day, there's many, many people out there trying to build these AI Agents for customer service.

The highest performing one will win because it'll give the best customer experience.

And every time we speak to customer service leaders, no matter how pioneering and ask them like, what is the thing, the one thing you care about the most, it's the customer experience.

It's the experience our customers get.

And so you're going to want to prioritize that.

You're to want to have the highest performing product.

So that's why we invest in it.

Today, 66%, like I said, over 6,000 customers.

But we want to get you to 100%.

What if we had a world where was 100%, Fin could resolve 100% of your queries?

So that's what we've been going hard on over the last kind of 12 months since Fin 2 came out.

And one thing we were looking a lot at is, you know, kind of stating the obvious, not every resolution is equal, right?

So a 30 second FAQ and a billing dispute 30 minutes over the phone, we've all had them as a consumer.

These are not the same thing, right?

They're not the same thing.

The amount of work that's delivered to try and resolve these types of queries is not the same amount of work.

It's a huge, huge difference.

And so this is the current frontier of AI Agents.

The frontier is how do you get Fin to do more work for you, taking all these queries that are really complex, really time consuming, really expensive.

How can Fin do more and more and more of those?

And that's what we've been digging into.

And I said, it's a frontier.

The second way that you can increase the amount of work Fin does for you, and that's to deploy Fin on new channels.

So lots of our customers have Fin on all different channels.

I'll show you later.

Fin works on messaging, email, voice, social channels, like loads of different channels.

So you can increase the work Fin does by obviously deploying Fin on new channels.

And these two things have been the focus of Fin 3.

How do we solve these complex queries and how do we get Fin working really well on all the channels that you use for customer service?

So we have significant upgrades across the Fin Flywheel with Fin 3.

We've got improved training with Procedures.

I'm gonna show you all of these things.

We've got improved testing with Simulations.

We've improved channels.

We've got voice upgrades.

We also have now Fin working over Slack and Fin now works in Discord.

And we've made improvements to Insights.

So first up, I'm gonna show you Procedures.

Procedures enable you to set Fin up to resolve your complex queries.

These are really hard, time-consuming queries.

These are the things, they're business-critical workflows.

They have many different steps.

They need to use business logic, the reading and writing to third-party systems. You're getting input and approval from different teams. Things need to of move around sometimes.

There's a high cost of failure.

It's high risk.

And so the way you do this historically is you train your human team to do it.

You train your human team to use their experience and apply their judgment to resolve these really complicated things.

And customers often in a kind of high state of emotion when they're calling or emailing or messaging about these things.

So that's what you do.

You train your human team.

You can now train Fin to do these things too.

With Procedures, you can train Fin to do exactly the things that your customers have your human team do.

So it's really cool.

With Procedures, Fin applies advanced reasoning.

It's an agentic experience.

I'll show you that in a second.

And the idea is that Fin can reason and think to get your customers to a happy resolution, happy for you and obviously happy for them.

So there's four parts to Procedures, four things you need to know about it.

One, it's got natural language instructions.

Two, it's got deterministic controls.

Three, it's got this fully agentic behavior.

And four, we've built an AI assistant to help you make sure you set it up the way that you want.

So let's take a look at it.

This is the product here.

I want you to imagine that you're a subscription-based business.

Many, many, many people here have subscription-based businesses.

And in subscription-based businesses, you often have SOPs, standard operating Procedures, SOPs for all different types of things that you want applied in a certain way.

Things like pausing subscriptions, canceling, getting refunds, are people eligible for refunds, all those kinds of things.

So you do a Procedure that is very simple.

You just start with natural language with a description of when Fin should use the Procedure.

Just telling Fin when to do it.

You can see it here, you describe it in natural language.

You then outline different steps.

So determining the customer's subscription status is a step.

Clarifying the type of change the customer wants with the customer is a step.

Finding out the refund reason if they're asking for a refund and so on.

So you can build in all the different steps that you need.

for whatever the Procedure is that you're designing.

It's really easy to do.

You just do it in natural language.

Or if you have SOP docs already, you just copy and paste the SOP docs, put them in, and it will start to work already.

Sometimes though, you want Fin to follow rules exactly.

You don't want Fin to reason and think and generate the reply.

You want it to be very, very specific about what it does.

So in this kind of powerful, elegant editor we've built, we've built in, along with the natural language, deterministic control.

So you can actually pick when Fin does very specific things.

We think this is the best of both worlds.

The way this works today in a lot of cases is like workflows.

These big, complicated, long workflow tools chaining together different kind of paths and so on.

Workflows are great, they're great for lots and lots of things, but they get complicated and they get really complicated for complex queries.

We think this is just a better way to do it.

So you can build a determinism in three ways.

We have data connectors, branching logic, and code.

So I'm gonna show you the three.

This is the first one.

This is adding a data connector.

So you can kinda see it here in the screen.

They're adding a different data connector.

In this case, you're fetching a customer's subscription details through an API.

So asking Fin, hey, fetch the customer's subscription details.

But you can connect any data imaginable to this.

This is through an API.

You can connect to whatever systems, whatever different tools you use.

The second is inserting branching conditions.

These are if/then/else statements.

And again, you can kinda see it here.

You can add in the branching condition within the Procedure.

So if a customer wants to pause their subscription, you can start a pause flow.

If they wanna cancel, tell Fin to start the cancel flow.

And then Fin will go and follow those very specific rules if these things come up through the conversation.

The third area is that you can go right under the hood and write code directly.

So this is like full control of what Fin does and how Fin does it.

So things like checking a customer, whether a customer is eligible for a refund.

You need to like really hard code that and make sure it's absolutely connecting to the right system, looking up the right data, applying the right attribute and so on.

So that's how this works.

You can go right into the code and you have this really powerful, generative, natural language editor with these deterministic controls built in.

So here's the thing about Procedures.

We write Procedures in this linear fashion, know, like perfect, nice, happy path, step by step, you know.

If only that was how people thought.

Customers are not like that.

Right, the real world is messy.

It's really messy.

It's, kind of hard to predict.

And so customers interrupt.

Customers ask new questions.

Customers change their mind halfway through.

All these things happen.

And so we've built Fin to handle these Procedures agentically.

That means, like I said, Fin reasons and thinks about what to do.

It doesn't follow the steps in this linear sequence, like you would cut it back going through a workflow.

It reasons at each step and tries to think about what the right thing to do is.

Should it invoke these flows that you've kind of hard coded in?

Or what should it do?

What's the right thing to do?

So let me show you how this works.

On the right here is WhatsApp.

And this is a customer talking to you in WhatsApp.

We've used WhatsApp to remind everyone that Fin works equally well over all these channels, email, WhatsApp, et cetera.

And it's a customer requesting to stop their subscription.

So pretty simple.

On the right is WhatsApp.

And on the left is the Procedure we've built.

So the yellow highlight is what Fin is doing at each point.

So you can see here.

and the customer is talking on the right, and then the yellow highlight is what Fin's actually doing.

So straight away in this example, Fin recognizes that it's a cancellation request.

Okay, so kind of knows what's happening.

But Fin doesn't follow the Procedure in a linear way.

It reasons and thinks through the whole thing and says, actually hang on, the first thing I need to do is get subscription details.

So you can see here, that's what it's jumped right down to get the subscription details.

Fin then clarifies if the customer wants to pause or cancel.

and explaining each one, you can see it there on the right.

And it's now jumped down to a different part where it's using this clarify if the user wants to pause, cancel, these kind of instructions within the Procedure.

So the customer asks for a refund and Fin checks and confirms their eligibility.

So here you can see Fin is determining the refund reason and then it's going down to a different part of the Procedure and running code.

to check the refund eligibility.

Is this customer actually eligible for a refund?

So it's really powerful.

You can kind of see how it works here.

And then here in the bottom yellow highlight, asking the user if they want to cancel or pause instead, which is exactly what Fin is doing over on the right.

So you can see here, Fin is gonna jump into the pause flow instead.

And all through this kind of experience, Fin is acting agentically.

It's reasoning, it's thinking, it's trying to work out, okay, what's the right next best step at any point through this Procedure?

So you can kind of see, I hope, with this product, it's a really powerful way to have natural language instructions get started really easily or build in this really powerful, hard-coded configuration.

For your customers, you can see it on WhatsApp, there's a lot of stuff happening here, a lot of powerful, you know, hard stuff.

And over here, your customer's experience is something easy, simple, nice, feels natural, it's just a really, really good, simple customer experience.

So this is kind of the cutting edge of what AI Agents in service can do today, resolving these really complex queries, blending this natural language generative AI with these deterministic rules.

It's still like the very, very cutting edge.

And it's a new scale, honestly, for all of us.

We're still, all companies building AI Agents and all different types of AI products.

This world of determinism and rules and this world of generative AI and probabilistic technology, blending these together is still kind of like the very cutting edge of software design.

And to help people do this, we're all learning this kind of new scale, this new world.

So to help you all do it, we've built an AI assistant so you can do it faster.

The AI assistant basically drafts new Procedures for you.

It's really, really simple.

You just write a little overview of the Procedure you want.

You can optionally attach docs, maybe SOPs, SOP docs that you've got.

And then what it will do is it'll pull context from your knowledge, your content, your historical customer conversations, and it'll create a first draft of the Procedure.

So we've been working closely with customers to do this, partnering with them, them telling us how to do it, us trying to build that and seeing if it works well for them.

And some customers are seeing really, really great success with it.

So I want to show you real example.

We have a video of Natalie from Nuuly.

Natalie's here, by the way.

So you can find her later and ask her questions about it.

But let's see what Natalie had to say about it.

I'm Natalie Hurst and I'm the Senior Director of Customer Success at Nuuly, which is the clothing rental subscription company within the Urban Outfitters family of brands.

We know that pausing and canceling subscriptions is going to be part of the customer's journey with us.

Because there's billing involved, of course, there's going to be a heightened sense of urgency and emotion in these types of conversations.

Really before Fin, every single one of these tickets went to an agent.

Someone still had to check the account.

look up the return status, what is the account status, and then determine what actions to take.

They're also incredibly high effort and high context work.

has been taking these actions now for two months, which is really exciting, and the results are super clear.

We've seen a 10% increase in Fin resolution rate, which equates to about 20,000 conversations on a monthly basis.

We see a 30% increase in Fin CSAT scores, which is...

really mind blowing and we've seen a 5% drop in conversations handled by associates.

We've also got many other customers, Nuuly aren't obviously the only one building the kind of cutting edge with Fin.

So we've got some quotes here.

George from Clay, they're using Fin to handle refund requests.

What used to be a slow manual process is now faster and more reliable.

I think this quote from Lee is really, really great.

Lee from MoneySuperMarket.

He says, moving from monstrous workflows.

that were hard to maintain to simple natural language instructions.

Loredo from Jukebox doing complex queries like delivery times, cancellations, it's like loads of great use cases here that Fin can now do.

And people think they're talking to a person because it's so natural like the WhatsApp example.

So Procedures really work.

This is kind of, it's the cutting edge but they really work.

You can get them to do really complex things and Fin will do them really well.

You can replace these complicated workflows with much easier to manage things.

Fin will act agentically and we'll have an AI assistant to help you.

Okay, so next I wanna show you Simulations, which is the second big thing we're announcing today.

Simulations is kind of the sister product to Procedures in lots of ways.

It's a brand new automated testing capability.

So we learned from talking to lots of you that you need to test changes.

can't just, like I said earlier, put them out at scale and kind of hope they'll work like you hoped, or like you intended.

So with Procedures, we've added significant power to Fin.

And this power also adds uncertainty.

So.

how well will Fin follow the Procedure?

If it jumps around, how will it jump around?

How well will it reason?

And then if I change the Procedure later, businesses change, what will happen?

So we've built Simulations to help you with this.

A simulation is a fully simulated customer conversation from start to finish.

We've also got AI assistant to help you with this too, and a library for all your Simulations.

So let me show you, you kind of have to see this to kind of really understand what the Simulation does.

So this is the same Procedure we looked at earlier, managing customer subscription requests.

We've put the Procedure into test mode.

So over here you can see in the right hand column, we now can run a Simulation on this Procedure.

The first thing you do is choose what user, what customer you want to test as, or what customer segment you want to test as.

You can look at different customers and have the Simulation run in different ways.

You can write in that user's first question.

That's the very, start of the conversation.

So for example, they wanted to pause their subscription.

And then third, any additional context, you can kind of add it in there as well.

You then lastly add success criteria.

The Simulation needs to know if it passed or failed.

And so you add in the success criteria.

For example, in this case, Fin confirms the correct subscription.

And then you simply hit run, and it'll run the Simulation.

So watch how this happens.

Watch how the Simulation runs.

So now we're simulating a customer conversation based on that criteria running through this Procedure to see if Fin is gonna do it.

Like what does Fin do?

Like let's check it out.

Let's run the Simulation and see.

Our sophisticated AI engine that I talked about earlier is gonna run this Simulation.

And you can see, is Fin reasoning?

When is it reasoning?

Is it following specific steps?

Did it get into deterministic stuff or not?

You can kinda see the whole thing.

If the test fails, like it's done in this case, you can go back up through the Simulation conversation and debug.

Like where did it go wrong?

Why did it not pass?

You can then edit that, change it, change Procedure, and get Fin working the way you want, get the Procedure working the way you want.

This is again a new skill to master.

So we built an AI assistant into Simulations.

The AI assistant is over there on the left-hand side and you can kinda just chat with it.

It will suggest things to you.

So in this case, the Simulation failed.

The AI assistant suggested what to change.

You can kinda just one click accept the change, rerun the Simulation, see how it works instead, see if it passes now.

In this case, we did that.

The new change worked and uh it passed.

So that's kind how they work.

You kind of run these Simulations and there's an assistant to help you.

The assistant though can do a lot more than just kind of suggest changes to the Procedure.

It can suggest new Simulations.

So it can say, hey, actually you're running a Simulation for Paul's subscription.

I think you need Simulations for other types of changes to subscriptions too.

And so for a complex scenario like a subscription business and a subscription Procedure, you're gonna want lots of different Simulations for different types of things.

So we have a library, all these Simulations.

kind of sit in the Simulation library.

And the reason this matters is because you can kind of run these Simulations, you can edit them, check them, test them, improve them, but businesses change.

Like products change, policies change, teams change, businesses change so fast, and you need to know that if the business changes and the product changes, like do these tests still pass?

Or what's Fin doing?

Is it using out of date stuff or connecting to the wrong system now?

So what you can do in the library, is just hit run all.

On the top right you can see run all.

You can just hit run all.

It'll run through all the Simulations and see if they pass or if they don't and tell you what to update the Procedures and the AI assistant will help you with that too.

So it's really cool.

It's really powerful.

That's Simulations.

It's kind of like a sister to Procedures.

Now with both of these things you can get Fin doing way, way, way more complicated stuff.

Okay next we've Channels.

We're kind of onto the deploy section of the flywheel and we've got some really cool updates with Channels.

because we want you to deploy Fin on all the channels that you have.

It works really well on them and we think that more and more people should try Fin on all these new channels and they'll see like really, really cool results.

We already have all these channels available for Fin, so not everyone realizes this.

Fin works equally well over all these channels, email, WhatsApp.

We launched Voice last March, Facebook Messenger, like over API.

Fin works really well.

Today we're adding Slack, we're adding Discord.

Fin works in Slack exactly the same way.

are exactly the way you'd expect it to work in Slack.

It replies, stay threaded, emoji reactions are shown.

If your teammates reply, their names and avatars appear.

So it works just like you'd expect.

And Fin now works in Discord.

Discord is an increasingly popular channel.

A lot of customers for different types of businesses are there.

And people want Fin to work in Discord too, like go where the customers are.

So Fin works exactly like you'd expect in Discord also.

The next big thing is Voice.

So we've launched a ton of upgrades to Voice.

We first launched Fin over Voice in March.

can kind of see there's lots of customers using Fin over Voice already.

You can kind of see a quote from Rizwan.

Customers often tell us they're surprised it's not a real person.

It's really, really good.

But if you actually think about the default phone experience today, it's not AI Agents.

AI Agents over voice is still a very new thing.

People are at the kind of, cutting edge, the frontier in deploying Fin over Voice.

The default phone experience for most people is still like waiting on hold.

pressing buttons, stuck in voice jail, you you just can't get out, talking to some kind of dumb bot, eventually you're shouting down that you wanna talk to a person, okay?

Like this is just the norm.

And it shouldn't be like that.

That's a terrible customer experience.

It's not good for business.

And so now with AI and with voice and Fin working over Voice, we can deliver a radically different experience, a way better experience.

So we've added tons of configurability to Voice.

We've added better guidance, more customization.

better testing and deployment, transcripts and summaries, and it's easy to set up, it works with all major telephony providers.

So Fin Voice is really good.

The challenge of Fin Voice is I can't show you it because there's no screen.

So what I'm gonna do instead is I'm gonna call Fin.

We're gonna really call Fin.

Again, this worked perfectly in rehearsal.

So there's a few things with this.

One is other companies will do Hollywood demos of voice and they reduce the latency and they'll make it sound amazing.

They speed these things up.

We're not gonna do that.

This is the very real experience of Fin over the phone.

I'm call a phone number and I'm talk to Fin to solve a problem I have.

I need all you to stay really quiet because Fin is talking to me and if you start talking, Fin will think you're talking to Fin too.

and everything will go sideways.

So let me show you how this works.

So here we have a very, very simple, a phone, and I'm gonna call Fin, and let's see what happens.

Hi, this is Fin, Halo TV's AI assistant.

How can I help you today?

Hi, my subscription order renewed and I didn't want that.

I want to cancel my subscription.

Let me start by pulling up your subscription details.

Is the phone number you're calling from the one linked to your account?

Yeah, this is the right number.

Thank you for confirming.

Let me look up your subscription details.

I see you have two active subscriptions.

Oh, Hang on.

Do I have two active subscriptions?

Yes, it shows you have two active subscriptions, Sports Extra and Movies Plus.

Which one would you like to cancel?

Ah, yeah, my wife had a Movies Plus, just remembered.

We want to keep that.

We'll cancel the sports subscription.

Got it.

You have two options.

You can pause your subscription for three months.

You won't be charged during this period and your subscription will automatically restart on the 9th of January or you can cancel completely.

Since you're reaching out within a week of your renewal, there's no cancellation fee either way.

Which would you prefer?

I didn't realize I could pause.

That's actually better.

I might use the sports thing again.

So I'll pause it.

Perfect.

Your subscription is now paused for three months.

You won't be charged during this period and it'll automatically reactivate on the 9th of January.

I've also sent you an email confirmation with all the details, including a link in case you'd like to reactivate early.

Is there anything else I can help with today?

The email thing is very useful actually.

Other than that, I'm good.

Thank you.

You're very welcome.

Thanks for calling and have a great day.

Goodbye.

All right, so Fin over Voice works.

You can see what it's doing.

We think it's really good.

I'm really proud of the team.

Voice, like we built Fin early.

We're early to market.

First kind of in messaging and then email.

Voice is way harder.

It's actually way harder.

You build a kind of different AI system in the backend for it.

It's much harder.

So we're really proud.

But I wanna just show what's happening there.

Remember like voice jail, 20 minute hold times.

I canceled that subscription, or actually better for the business, paused it in what, like 90 seconds, just by calling a number.

It's just such a better experience.

And look what Fin did.

It confirmed my identity using the phone number.

It pulled up accurate account information, turned out two subscriptions.

I interrupted it, and I handled the interruption well.

It asked me to clarify which subscription I wanted to cancel.

It offered that I could pause it instead of canceling, which is part of the Procedure instructions.

And it sent me an email with a link through.

to kind of easily reactivate the account.

There's some latency in there.

You can see like it's a little bit of kind of a delay at times and we'll work on improving that.

What's actually happening is Fin is looking up systems. You it's actually doing things.

And so this is the kind of current state of the art for voice.

This is the real experience.

Things will get better, but I think it's really good already.

I think that is like way better than wait and hold and easily good enough to put into production for loads of your customers to answer these kind of queries over the phone.

So I'd love you to try it.

Try Fin Voice.

Yeah, I think you'd be pleasantly surprised at how well it works.

So, Fin works really well over voice.

We've added Slack, we've added Discord.

Fin is now the most omnichannel agent in the market by far.

Okay, the last thing is you need to understand how well Fin's working at scale and how you can improve performance.

That's the fourth step in the Fin Flywheel, which is analyze.

Last May, we introduced a new Insights product, and the Insights product had three components.

There was CX Score.

which is this new metric, an AI-powered metric, which gives you full coverage over all of your customer conversations.

CSAT is a really limited metric.

It's useful, it's really limited, it's the metric we love to hate.

You know, we brought CX Score in, we think it's just a much better way to measure full customer experience.

We brought in Topics Explorer, again, it's kind of like AI-powered organization of your customer conversations into topics and subtopics, you can kind of see.

where the conversations are happening and where they're not and what's bigger, what's smaller.

We brought in AI-powered Suggestions.

This is like when you go around the full flywheel and when you invest in all the other steps and you Fin running in production, will suggest, use AI to suggest changes to you, like missing content or things that could be better, things that aren't quite right.

And they're one click accept.

You accept the suggestions and then Fin will do it the right way the next time and increase your resolution rate.

So we also, we got loads of positive feedback to these new Insights features, all AI powered, quite groundbreaking in lots of ways.

We also got loads of feature requests too, of course.

So that's what we've been doing.

With Insights, we've been just trying to power through the most common feature requests.

So I hope there's a few crowd favorites in here.

One is we've introduced CX Score reasons.

So now you can see the reason for the CX Score.

Was it product feedback?

Was it a policy issue?

Maybe just the answer quality wasn't great.

So you can dig right in.

These attributes, these are like recorded as attributes, they're now built into our reporting system.

So you can filter and segment by all these different reasons.

And CX Scores are kind of fully battle hardened.

It's like out in the wild, operating at scale.

And we think it's really ready to replace CSAT as the primary metric you use to measure customer experience.

The second thing we did is we've new Topics trends report.

So this automatically highlights the most important weekly changes, like things are spiking or.

the resolution rate dropped for some reason or this emerging customer thing, you now no need to manually dig for these patterns.

We will surface the patterns really early to you in this trends report and you can act early.

So you can actually get ahead of these emerging issues and fix them before they ever come a major customer problem.

So just more proactive support, it's really good.

You can have curation code over topics.

This is the most common feature request we got.

People want to rename topics or merge topics together, move topics around, create new topics and see that kind of fill.

So we've given you the full control, I think heard someone say yes in the audience.

We're giving you full control over how these topics are named and so on.

And then lastly, we've got AI-powered suggestions.

They just do a lot more now.

So AI-powered suggestions will spot duplications in your content, contradictions in the content surface those to you.

It will learn a lot more from the suggestions that you reject and then make better suggestions in the future.

And if you use Fin over, or use Fin with Zendesk or with Salesforce, Fin by the way works with all major other customer service platforms, integrates really well with them.

Lots of people are using Fin, starting to use Fin with Zendesk, with Salesforce.

Now the one-click suggestions will apply in those places too.

So can make your Zendesk docs better by accepting Fin powered suggestions.

We honestly don't care, wherever the content is, wherever your customers are getting, we want it to be as good as possible, so we will happily.

update Zendesk docs to do that.

And the last thing is we have suggestions for today's content and knowledge.

We now have suggestions for data that might help, for actions, for guidance.

So we're just widening the surface area there.

So I want everyone here to think about this flywheel.

I think it's really important.

It's like a way of thinking about doing better customer service, delivering way better experiences.

I think it's also your job now.

You know, customer experience leaders job has changed dramatically.

Again, going back to the start, headcount issues, volume issues, a lot of stress, people quitting because the job was too repetitive.

We're in a different world now.

We're in this new world, this kind of Fin Flywheel world, AI Agent world, and this now really is your job.

One last thing I wanted to cover is when we think about how we design Fin and think about how we partner with all of you to build Fin, we believe that The customer experience is one of the most strategic things in a company.

It's one of the most strategic things a company can do is deliver great customer experiences.

lot of companies compete on this idea.

They're the best at customer experience.

And we think you should be in total control of that.

You should own it yourself.

You should be able to manage it yourself.

And that's why we build things to be self-serviceable with Fin.

We want you to be able to do it yourself.

This, by the way, is in stark contrast to other companies who...

you have to contact them to get feature requests built.

That's like slow and it creates a deep dependency.

It's kind of consulting where at times a deep dependency on them to do the things that you need to do and you want to do.

And it's just really slow.

And we don't think it should be that way.

It's way harder to design Fin.

It's way harder to design these features so that they're easy for you to use.

You can do it all yourself.

It makes things way harder for us.

We think it's the right way to do it.

We think you should have total control over all of these things.

Anyone in your team can learn this stuff.

That's why we AI assistants.

That's why we let you get deep into the code.

We want you to control this yourself, manage it yourself, get into the experimental mindset yourself, change things, see if they work, try again.

It's really important for us.

And so that's basically it.

It makes me happy to share all this with you.

I'm really proud of our team.

I love that I'm talking to the pioneers of our industry, the people doing this, all of you at the cutting edge.

experimenting, exploring, wanting to like build a better world.

So the recap here is we announced Fin 3 today.

It's the best AI Agent in the market in our opinion, but by the benchmarks and the kind of other objective ways in which we measure, it got a lot better.

We've got big investments in our AI layer, in our model layer.

We think those things are really important for everyone to understand.

We're getting this like crazy, crazy chart, just increasing resolution rates.

Again, not cherry pick numbers on average across our entire customer base.

We announced Procedures to help you do complex queries.

We announced Simulations so you can kind of test all these complicated changes, make sure it works the way you want it to.

We've got Fin over Voice works really well.

Slack, Discord, and as you just saw there, big updates to the Insights product.

So a ton of new stuff.

Some of this is available in the product already and we're rolling out the other things as fast as possible.

So all this means, like what does this all mean?

It means Fin can do way more work for you.

We have seen so much benefit in our industry from freeing people up to do higher value things.

And this is just gonna let you do more of that.

If Fin is doing more work for you, it means you can take the team and have them do different types of things.

Customer success or more time with VIP customers or just more time working through other things with people.

Like it's just, again, kind of a next evolution, next step in how much Fin can do and how much more you can do and the experience you can deliver improving all the time.

We've got Fin Labs here.

Fin Labs is a really cool thing.

You can demo all this stuff.

You can go up, I don't know where Fin Labs is, up or down, somewhere in this building.

You can go check out Fin Labs.

Our product and engineering team are here.

They'll talk to you about all the features.

They'll show them to you.

You can check out your own data and different things and insights and things like that.

And that's it for me.

I'm really excited about Fin 3, as you can tell.

I hope you're excited about it too.

Please try all this stuff.

Give us all that feedback.

You can learn a lot more at fin.ai.

We've updated the site this morning.

All this stuff is on the site.

And that's it.

Thank you so much for your attention.

So we're investing big in Fin 3.

We're investing big in customer service.

But that's not all.

And so I'd like to welcome back Eoghan, our Co-Founder and CEO, to tell you what's next.

Just so you know, I held my breath for the entire voice demo.

There was high fives behind the scenes.

So I'm proud of where we've come in just two short years where at the start we set ourselves the goal, the aspirational goal to create a service agent that could do essentially all service.

I feel like we have with Fin 3 essentially achieved that.

It can handle now the majority of your service queries, including the very sticky complex ones on all the channels that your customers are on now, of course, including Voice very much and Slack. And at a rate of efficiency and effectiveness unlike that, which you'll find with anyone else, particularly now that it runs on our proprietary models.

But of course, we're not going to stop there.

Two years is a long time in AI, short time in the development of any category, there's more to come.

There's more of our vision to play out.

And so I'm gonna start to tell you what that's gonna look like.

And I'll do so by showing you first where our customers are taking Fin.

These guys have been teaching us a lot and we're gonna follow them.

As you know, Fin offers a lot of flexibility and functionality through things like Guidance and Tasks.

And people are using this functionality to do things beyond service.

One great customer of ours, Lightspeed, is using it end-to-end across the customer journey.

Another great customer of ours, Anthropic, are seeing incredible success, of course, with service.

But they're now planning to use it to create a more seamless customer experience that traverses the life cycle of the customer beyond service too.

And WHOOP, a company you likely know, the wearables company, have done something really, really interesting.

We have a short video that I hope works that'll show you exactly that.

My name is Emily Shirley and I'm a business manager on the growth product team at WHOOP.

As we were looking toward one of the biggest product launches in WHOOP history, we were expecting to have 20 times the number of sales chats reach our agents.

We needed something that we could set up quickly and we needed something that we could count on.

Leaving Fin on overnight for the very first time.

I can remember waking up and rushing to my computer and there it was, having conversations and making sales all on its own.

Fin is now able to resolve 84% of the conversations it's involved with.

We've had a 130% increase in sales attributed to the Inside Sales Team since implementing Fin.

It's become an integral part of our sales team, and we really only just scratched the surface.

So 84%, thank you.

84 % of their sales conversations is kind of crazy.

And it's just a little example of what people are going to be able to do with this.

The appetite in the market is very, very real.

So we've landed on two ideas that are important to share as I explain our thinking here as we've been working with these people.

The first is a huge trap that we need to avoid.

know, all these teams, sales, success, marketing as well as service, they're all looking at these various agents that are showing up at the market.

They're all wanting to adopt agents too, but if you play this out a little bit, I worry about the customer experience when you've got multiple agents in the same place.

I worry that what you create is a discombobulated mess of competing agents that do a lot of damage to the customer experience.

These different agents are all of course going to have their different priorities and goals, different contexts and knowledge.

Maybe they're on different channels.

Maybe they don't have the same information.

I'm just unconvinced that this world is actually workable.

I think it's dead on arrival.

Some people will try it, but I don't believe that this represents the future.

But the flip side of this risk is an incredible opportunity.

It's an opportunity to finally bring a level of seamlessness that customers have long expected and deserve.

In the conventional world that we're all super familiar with, know, no single team, no single person is just practically able to be outstanding at the sales job and the lead qualification job onboarding, support, upsell, etc. They're all different specialities.

And so understandably, they get broken into different chunks and the customer gets passed around.

The result are the super familiar standard lowest common denominator customer experiences.

These proliferate 99.99 % of brands, the very best brands.

The agents of today are gonna change this.

They're clearly capable, as you've seen, of many different use cases across the entire customer lifecycle.

And that's gonna open up the door to previously unimaginable types of customer experiences that I think will really delight your customers once you start to deploy them.

This is exactly what our most pioneering customers have started to do, but they're really just scratching the surface.

And so our intention is to take this far further for them.

Fin will be not just the world's best service agent, but the world's best Customer Agent.

So.

Obviously, it's going to be crucial that we continue to obsess about service.

It's core to what Fin does.

It's going to be the bulk of the work that is needed.

There's a long road ahead of us.

As you saw with Fin 3, it is our big focus.

But it's also going to need to get great at holding the hand of the customer for the very first time they consider your product when they purchase it.

will need to get great at being that personal concierge and deliver the type of speed and sophistication and perfection that you're seeing in service across the life cycle.

And so this is our new goal and certainly our most audacious and ambitious goal since the start of Fin.

From here, you're going to see Fin.

move from its origins as a task-based agent to truly an agentic agent.

You see what Procedures that Paul mentioned.

There's an agentic nature to the way in which it works.

But now you're going to see the AI used not just to tailor responses to queries, but to make decisions about how it interacts with the customer and how it wants to bring the customer through your business.

To make this a reality, we have to add some significant new capabilities.

First, we'll add a bunch of different roles.

Service will be one.

We're going to add goals.

These will be prioritized.

And Fin will be able to determine which goals it wants to pursue and when.

We'll have memory so we can learn about the customer over the life cycle from the very first time it meets the customer through to 10 years later.

It will have deep knowledge about your business.

Everything from the plan that the customer is on, the history, every policy process, every product.

It'll be able to act on completely a complete view of your customer.

And it will need to be able to interoperate with.

all of the different systems that matter, not just to service, but to sales, marketing, success, and so on.

This will become the frontier of what we'll all call Customer Agents, and this is what we're building to lead that category.

So we're already heads down building this.

We have been for a little bit, and I want to start to show you what that looks like.

We have a couple of cool demos of two roles in particular, and these are...

things you can actually experience at the Fin Labs booths.

Nothing here is a mock-up or a demo for this purpose.

This is real technology that is pretty mature at this stage.

So first we're going to look at a shopping assistant.

This example here is an outdoor gear site.

It's a fake site.

This is just a demo conversation with real technology.

The customer asks what kind of jacket they should take on a trip to Iceland.

Now over on the right you can see Fin thinking about what to do, applying reasoning and logic.

It's important to note that nothing in the database has been tagged with Iceland.

Fin has never been trained on this question.

But it was able to use its reasoning abilities to figure out the right inventory to search for.

It shows a nice carousel of some products.

And it even offers to add the product to the customer's shopping cart.

Now, already, this is an experience that likely was never going to happen.

There was never going to be a human ready to go in the moment that the customer wanted, able to answer so quickly and offer unique product options.

So anyone in this business or even outside this business can imagine the conversion rates have already increased, but this is not where this stops.

In fact, in this case, Fin, like any good salesperson, sees an opportunity to take a step further based on the goals that it has.

It uses its reasoning and offers an upsell.

This is what we mean when we will talk about blending the different use cases of sales, some marketing, service.

Now let's look at a different example.

So this next example is on our own site, fin.ai, and this is an SDR. This is Fin selling Fin.

This is a very much, very much a real conversation that happened.

So it starts with some basic product questions and Fin answering.

But as you can see in the right, there's a bunch of attributes that Fin knows it will be useful to collect as part of the conversation.

So the conversation flows.

Fin starts to contextually fill in the blanks and sketch out a more complete profile.

Now, I'm going to skip ahead a little bit.

And in the next moment, you'll see it quickly start to ask a lot more questions at speed.

Let's see, can we do that?

Here we are, it's speeding up.

So before long, Fin has started to fill out all of the questions it has, gather all the information that it needed.

But what's really important to note here is that it's not, you know, stepping through the sequential list of predetermined questions in the robotic fashion that the previous generation of bots did.

It is using its skill to naturally find the right moments to ask the questions it knows are important.

The human involved in this very real conversation clearly found that effective and it just demonstrates the opportunity that this generative AI offers.

Of course, beyond just lead qualification and asking questions, it can then use its judgment to figure out, it pass to the sales team and package up a summary for that team.

Super impressive stuff and yes, Fin is really good at selling Fin.

Okay, this is gonna start to roll out over the next number of months.

We've got a couple months to go to kind of finish it, but we very much want your feedback as we do that.

Now, we don't usually do this.

It's atypical for us to share what we're working on in this fashion.

very product first rather than marketing first, probably to a fault, despite our glitzy marketing.

We really only talk about things that are baked and done, but there's something different about this time.

Here's the two reasons that we're starting to open up in this fashion.

First, we want you to be ready for this change.

The Customer Agent category is coming.

We're going to build it.

Other people will build it.

It's going to happen, and we don't want it to happen to you.

We want you service leaders to be the expert agent leaders in your business so you can assist your peers when it comes to them wanting to deploy agents elsewhere.

And then relatedly and extremely selfishly, when they do talk to you about agents, we want you to say, Fin is an excellent Customer Agent that can help you.

So of course, it's important I reassure you that we are going to continue to work very hard in service.

It will be the bulk, the majority of our resources, our R&D resources that we spend on Fin.

But the inevitable and exciting future that is broader than that, the Customer Agent, will become an increasing focus.

It's an amazing opportunity to build the single seamless unified Customer Agent experience.

that I think we kind of all knew was coming.

So I would ask you to lean in, start to play an experiment, deploy these in your own organizations, help us along the way.

And yeah, I'm excited to see what you do with that.

Thank you very much.

Talk soon.