Intercom

Getting your customer support AI Agent live is one thing, but maintaining its performance over time is another. Long-term performance takes more than good training data, and we’ve seen the highest-performing teams focus just as much on the system around the AI Agent as the Agent itself. They build operating models that keep it learning, improving, and aligned with the business as things evolve. We're curious how this is working for different teams. What has helped most with your AI Agent performance? Vote below, or let us know in the comments 👇

AI performance isn’t static, but most teams treat it like a one-time implementation. The most successful organizations design systems that learn. They analyze where the AI Agent struggles, then feed that insight directly into structured improvement. Whether you follow a formal loop (like the Fin Flywheel framework) or something simpler, the goal is the same: Make improvement inevitable. You can find more details on the systems and structures that sustain AI Agent performance in part 4 of our 2026 customer service planning series. Find the link in the comments.

AI Agent performance can stall for multiple reasons. Here are three we see show up frequently: 1. No one owns the system. 2. Iteration is either too slow or too risky. 3. Underinvestment in content strategy. They’re not the only issues, but they’re common, and costly. The good news is, they’re fixable. The best teams are solving them by assigning ownership, making iteration safe, and treating content like infrastructure. We’ve seen these patterns firsthand in teams like Dotdigital, Anthropic, and in our own support org at Intercom. If you’re scaling AI for support and want to build an operating model that sustains AI performance over time, this will help.

Trust is much more than a feature. It’s a foundation. Intercom is a Founding Technical Contributor to AIUC-1 and is now among the first companies certified under the world’s new standard for responsible AI Agents. This certification isn’t compliance for compliance’s sake. It gives every business using Intercom confidence that our AI is safe, resilient, and built to the highest level of accountability. Achieving it required independent, enterprise-grade testing of our AI systems, ensuring safety, reliability, and strong protections against issues like hallucinations and brand risk. It confirms that Intercom’s long-standing commitment to trust fully extends to Fin. And it gives every business using Intercom confidence that our AI won’t make unsafe choices, leak data, or behave unpredictably. Read more about it in the link in the comments.

“The first time you answer a question should be the last.” In part 4 of our 2026 planning series, Intercom’s VP of Customer Support, Declan I. explains why the real lever for long-term AI success isn’t the tech – it’s the system you build around it. This week’s edition covers: • How to stop AI performance from degrading over time • Ways to iterate fast without risking quality standards • What it takes to build a system that learns by default • And why knowledge isn’t content, it’s infrastructure Read Week 4 in the series now ↓

We talk a lot about AI resolution rates. But what actually drives them? From what we’ve seen, it’s rarely just one thing. It’s the interplay between structured knowledge, reliable automation, intentional conversation design, and clear ownership. The teams seeing the strongest results are building the systems around it. They're assigning responsibility, tightening feedback loops, and making the whole thing self-improving. Still, most teams have a center of gravity — the one thing that unlocks momentum and compounds progress. We’re curious where others are seeing the biggest lift. Which of these do you think drives the biggest improvement in AI resolution rate? Vote below, or share what’s made the biggest difference in your team’s performance 👇

This week we launched Office Hours. You showed up. Now they’re here to stay. Earlier this week, 236 Intercom and Fin users from 210+ companies joined our first-ever Community Office Hours and solved problems, live. 85+ questions were asked, and over 80% of attendees rated the session very useful. Why? Because it’s built around you. Q&A-first. Peer-led learning. Practical guidance. Real answers from top experts. Office Hours happen every Monday. You bring the challenge, and we work through it together, with help from Community Experts like Nathan, Milan, Conor, and Julian, and our community manager Diana. This is community, working as intended. Join us, learn from others, and get more from Fin and Intercom. Register for next week’s sessions 👇

AI is a system. Not a feature. Last week, we introduced the four essential roles that make AI actually work in a support org. This week, we’re connecting the dots. Here’s the operating loop that turns those roles into a high-performing AI system: 1. AI ops lead → Identifies performance gaps and drift 2. Knowledge manager → Fixes inaccuracies and missing content 3. Conversation designer → Improves clarity, tone, and flow 4. Automation specialist → Teaches the system to take action Each role makes the next one more effective. Each improvement compounds the system. A loop that learns, adapts, and scales without throwing more humans at the problem. This week’s edition of our 2026 customer service planning series unpacks each role in detail and includes a phased playbook to help you start building your AI-first support team, even if you can’t hire yet 👇

Support teams that have scaled AI fastest have one thing in common: Someone owns the AI’s performance. With ownership, everything changes: the system improves, the outcomes compound, and AI becomes a reliable part of how support work gets done. This week in our 2026 support org planning series, Declan Ivory lays out a three-phase approach to building your AI-first support team (even if you can’t hire yet) starting with just a few hours a week, and scaling from there: 𝗣𝗵𝗮𝘀𝗲 𝟭: 𝗔𝘀𝘀𝗶𝗴𝗻 𝗼𝘄𝗻𝗲𝗿𝘀𝗵𝗶𝗽 Start small. Give existing team members 5-10 hours a week to manage AI knowledge, monitor its performance, and flag improvement opportunities. 𝗣𝗵𝗮𝘀𝗲 𝟮: 𝗙𝗼𝗿𝗺𝗮𝗹𝗶𝘇𝗲 𝗿𝗲𝘀𝗽𝗼𝗻𝘀𝗶𝗯𝗶𝗹𝗶𝘁𝗶𝗲𝘀 As resolution rates grow, this becomes core operational work. Codify the responsibilities early to prevent drift and knowledge debt. 𝗣𝗵𝗮𝘀𝗲 𝟯: 𝗦𝗽𝗲𝗰𝗶𝗮𝗹𝗶𝘇𝗲 𝗮𝗻𝗱 𝗵𝗶𝗿𝗲 Once AI is resolving 50-70% of volume, these become full-time roles. Specialized ownership becomes the infrastructure for scale. You can read the full article on the four foundational AI roles showing up inside support orgs leading the way below.

AI doesn’t fail because the model is bad, it fails because ownership is missing. Once someone owns it, everything changes. In part 3 of our 2026 planning series, Intercom’s VP of Customer Support, Declan I. dives deeper in the four foundational AI roles that are already showing up inside the teams who are scaling AI the fastest. Inside this edition you'll learn what these roles look like in practice: • What they do • How they work together • Why they're crucial to your AI performance • And how you can get started, even without new headcount Week 3 is live now.