Introducing Fin 4: A major step toward a fully agentic customer experience

At Pioneer 2026, our Chief Product Officer Paul Adams announced Fin 4, a huge step toward our mission of delivering perfect customer experiences to every single customer, every single time.

Fin 4 brings major upgrades to every part of the Fin system – a more capable Customer Agent, a more powerful Operator, and new ways for Fin to work with your team.

Here’s what’s new across the system:

1. Customer Agent: We believe every business will have one Customer Agent. Ours just got a big upgrade. Fin now drives customer success and, with Skills, figures out how to solve problems without you defining every step.
2. Operator: Fully agentic customer experiences need an Agent to run complex operations. Routines and Connections give Operator powerful new ways to do that work for you, proactively and autonomously.
3. Copilot and Human in the Loop (HITL): As Fin resolves more, human teams take on a supervisory role. A second-generation Copilot and Human in the Loop give them entirely new ways to work together.
4. Fin Apex: The next generation of Fin’s AI models is here. Apex 2.0 and Apex Flash 2.0 push performance further across resolution rate, speed, and accuracy. Apex for Procedures extends Fin’s specialized model approach to complex, multi-step queries.

That Customer Agent will fully understand the business goals and the customer context, and will resolve all their problems excellently. It will manage 90%+ of inbound conversations, and will also act proactively to achieve business goals and help customers who might be stuck.

It will be incredibly powerful, and will need management and oversight.

Rather than people trying to manage the Customer Agent by going from screen to screen and menu to menu, businesses will have an Operations Agent to do that work. That’s why we shipped Operator, the Operations Agent for Fin, in May of this year.

Human teams will manage the Operations Agent, set the strategy, and design the customer experience, and the Operations Agent will manage and optimize the Customer Agent.

The helpdesk solutions we have been using for the past decade, based on cases or tickets, are no longer fit for purpose. You would never design that solution if you started today.

Support teams need a completely reimagined helpdesk that is built for AI.

We strongly believe that this setup is the future of our industry, and at Pioneer, Paul shared exciting updates across all of these areas.

Fin now onboards, activates, and
re-engages your customers

Fin has already expanded beyond Service into Sales and Ecommerce. Now, Fin proactively drives customer success, helping customers onboard, achieve their first win, and use more of a product, more deeply.

There are three important updates to how it works:

• Fin is now proactive. It starts the conversation and interaction with the customer itself.
• Fin is now outcome-driven, based on what you want to achieve for your business and customers, over any time horizon.
• Fin also adapts to each customer, handling whatever comes up in conversation.

Fin will proactively engage customers, deciding when to help, what to say, and where to reach each customer. Fin can follow up on a stalled conversation or nudge a customer through an unfinished step, without your team manually reviewing accounts and coordinating individual outreach.

For example, Fin can:

• Introduce customers to features they haven’t used, explain why they’re useful, and help them start using them.
• Create an onboarding plan for each new customer and help them complete it.
• Reach out to customers whose usage has dropped and help them get started again.
• Identify signs that a customer may churn and reach out to help address the problem.
• Identify growth signals and surface expansion opportunities.

Fin will drive your customers toward any number of outcomes, over any days, weeks, months.

Fin now has Skills to resolve customer problems without a defined path to follow

Our vision that every business will have a single Customer Agent that handles all customer communication means that Fin has to handle the widest variety of customer problems – from those that are more procedural and follow rules and policies, to those that don’t. These include things like troubleshooting or situations that need investigation.

Last year, we launched Procedures, which let Fin resolve complex, multi-step queries like refunds, account changes, and subscription issues. What makes Procedures powerful is the control they give you over exactly how some of your most critical processes get done. But not all customer problems follow a defined path to resolution, which means that support teams themselves often have to handle complex troubleshooting where investigation is needed.

Today, we’re launching Skills to address this. They give more agency to Fin to solve ambiguous customer problems.

With Skills, you can teach Fin how to resolve a complex customer query that doesn’t follow a defined process. You can give Fin a job to do, and it will use different Skills to figure out what to do to complete the job. The more Skills you give Fin, the more capable it gets.

In May, we launched Operator, an Agent for your customer operations. Since then, we’ve seen more than 1,600 customers transform their customer experience work with it.

Today, Operator is more proactive, connected, autonomous, and collaborative. It works across tools like Slack and GitHub, investigating performance, identifying what needs to improve, and carrying out the work while you stay in full control.

Here are Operator’s new capabilities:

• Routines: Operator can complete ongoing work on a set schedule or in response to a specific event. Ask it to send a team performance report every Monday or alert you to an incident with the diagnosis and a proposed fix included.
• Connections: Operator uses information from tools like Slack and GitHub and acts on what it finds. It can review product announcements in Slack and draft the changes they require in Fin, or submit bug reports in Github.
• Memory: Operator remembers preferences you've shared, like your writing style or how you structure reports, and carries that context forward across conversations.
• Artifacts: Share Operator's work, like a drafted Procedure or an analysis of customer conversations, with the people who need to review or act on it. Artifacts make it easy for teammates to review Operator's work and collaborate on what happens next.

Together, these let Operator take on more of the complex work involved in running and improving your support operation.

At Teachable, for example, the team used it to orchestrate a payment migration by itself.

In a world where 90% and more of customer questions are being resolved by a Customer Agent, the role of the human support team evolves. And they need next-generation helpdesk capabilities designed for that world.

We’ve made updates to the Intercom helpdesk with Human in the Loop and Copilot.

Human in the Loop

When a conversation needs a review, piece of information, or an approval, Fin will loop in a human to approve, edit, or take over. This changes the role of a support rep from answering customer questions to supervising Fin – which is an important signal for the future.

Copilot

We’ve rethought, redesigned, and rebuilt Copilot as an Agent that proactively prepares the work when a teammate needs to reply directly to the customer. The second generation of Copilot is now more autonomous and proactive. Every time a rep opens a conversation, Copilot has already searched all the connected tools, summarized the customer issue, drafted the best answer, and recommended the actions to take.

Both Copilot and Human in the Loop work natively inside Intercom, but will also integrate with other helpdesks.

These are the first set of changes on our path to giving you a next-gen helpdesk, designed for a world of 90% Agent resolution.

None of what we’ve announced today matters if Fin is not delivering the highest performance in the market.

Earlier this year we announced Apex, a custom trained family of models that outperform every commercially available frontier model on real-world support tasks. Today, we’re announcing significant updates to our Apex family of models.

This includes:

• Apex 2.0, our most capable model for customer service yet
• Apex Flash 2.0
• An entirely new model: Apex for Procedures.

These updates unlock more capable answer generation, faster, more natural voice conversations, and stronger reasoning through complex, multi-step queries.

All of these upgrades outperform the models they replace on core metrics like resolutions, speed, and accuracy.

Apex 2.0

Apex 1.0 already outperformed the leading frontier models on resolution rate, hallucinations, and speed. Apex 2.0 improves on Apex 1.0 in three ways:

1. It delivers higher performance across the core metrics that tell us Fin actually solved the customer’s problem.
2. More users confirm that Fin answered their question – what’s known in the industry as a hard resolution. That’s a really hard metric to improve.
3. It’s more grounded in each company’s content, which means 16.5% fewer hallucinations. Fin gets answers right even more often.

Apex Flash 2.0

Earlier this year, we launched Apex Flash – a lightweight version of Apex built for latency-sensitive channels like Voice.

It works with a real-time speech layer that manages the audio and the pacing of the call, while Apex Flash decides what Fin should say. That's what lets Fin keep up with the caller and give accurate answers in the moment.

Today, Fin Voice moves to Apex Flash 2.0, built on NVIDIA Nemotron 3 Nano, and it's our fastest model yet.

It answers faster, follows guidance more closely, and keeps its answers short, which is what callers need on the phone.

Apex for Procedures

Finally, Apex for Procedures is a new model trained to reason through complex, multi-step queries.

It outperforms Sonnet 5, the best-performing frontier model we tested for this work, on all the metrics that matter most for complex queries. This includes 15% more hard resolutions and 37% fewer major hallucinations.