Book Calls For October Is Open Book Now

Back to blog

AI

Fin Memory Explained: What Intercom's AI Agent Remembers About Returning Customers

Intercom's new Fin Memory feature lets Fin AI recall details from past conversations with the same customer. Here's what it actually stores, how long it keeps it, and what to check before you leave it on default settings.

N

Written by Nevil Paul

Published on October 14, 2026 · 5 min read

AI
Fin Memory Explained: What Intercom's AI Agent Remembers About Returning Customers

Intercom quietly rolled out one of the bigger behavioral changes to Fin AI this year: it now remembers people. If a customer talks to Fin today, closes the conversation, and comes back next week with a related question, Fin can pick up where things left off instead of starting the troubleshooting from zero. For businesses running Fin as their front-line support agent, this changes how a returning customer experience actually feels, and it introduces a few things you should check in your own setup before you assume it's working the way you think.

What Fin Memory actually is

Fin Memory is a new capability for Fin for Service, currently in open beta. Before this, Fin treated every conversation as its own isolated event. It had access to help center content, your macros, and whatever the current conversation contained, but nothing about what that same customer told it last month. Memory changes that by giving Fin a persistent internal record tied to each contact, built from their past conversations, that it can draw on the next time they show up.

This is not the same thing as the conversation history a customer can scroll back through in the Messenger. That transcript view has always existed. Memory is a separate, internal layer that only Fin uses to inform how it responds. Customers can't see it or edit it, and your support team can't open it as a record in the inbox either.

What it actually stores, and for how long

Intercom breaks Fin Memory into three categories, and each one has its own retention window.

The first is recent messages, which is essentially a running log of what was said between the customer and Fin in past conversations. The second is summaries, short recaps of what happened in previous conversations rather than the full transcript. Both of these are kept for 45 days and then deleted automatically.

The third category is facts: specific personal details a customer explicitly told Fin at some point, like a plan tier they mentioned, a preference, or a detail about their setup. Facts stick around longer, for 180 days, before they're cleared out.

None of this is something you configure manually. There's no dashboard where you edit what Fin remembers about a given contact, and no documented way to purge an individual record early. It expires on its own schedule or it doesn't exist yet.

Where it works, and where it quietly doesn't

A few scope limits matter a lot here, and they're easy to miss if you're assuming memory applies everywhere Fin operates.

Memory is specific to Fin for Service. If you're also using Fin for Sales or Fin for Ecommerce, memory doesn't cross over between them. A fact a customer shared in a support conversation won't surface in a sales conversation with the same person, even though it's technically the same contact in your workspace.

It's also limited to the Messenger and email channels right now. And during this open beta period, it only works for identified users, meaning contacts you actually have on record. Anonymous visitors and leads who haven't been identified don't get remembered between visits, which matters if a meaningful chunk of your traffic is anonymous browsing before signup.

It's on by default, which is the part worth sitting with

Fin Memory ships turned on for every Fin for Service workspace. You don't opt in. If you want it off, it's a toggle under Settings, then Workspace, then Security.

That default matters because memory means Fin is now retaining and reusing personal information customers share in conversation, across sessions, without them being told explicitly each time. Intercom's own documentation flags this directly, noting that workspaces using the feature may need to update their privacy notice and tell customers their information is being retained this way. If you haven't looked at your privacy policy since this rolled out, that's worth doing before a customer asks you directly how long their data sticks around.

There is a way to audit when memory actually gets used. Inside a conversation's event log in the inbox, you can look for a "Fin referenced memory" entry, which tells you Fin pulled something from a past interaction to shape its answer in this one. It's not a full view of what it recalled, but it's the closest thing to visibility your team currently has.

What this means for how you set up Fin

If you're running Fin for a client or for your own business, this is worth testing deliberately rather than assuming it just works in the background. Have a returning customer conversation with a real account, reference something specific from an earlier chat, and see whether Fin actually picks it up. Because this is still in open beta and scoped to identified users on Messenger and email, it's easy to test it in a scenario where it won't fire and wrongly conclude it's broken.

It's also worth deciding, deliberately, whether the default-on setting is right for your business. A company handling sensitive account details might want this off until they've updated their privacy language and confirmed what their compliance obligations actually require. A company where repeat conversations are common, like ongoing account management or multi-step onboarding, might find this one of the more useful things Fin has picked up recently, since it removes a lot of the repetitive "let me explain my issue again" friction that usually pushes customers to ask for a human.

Either way, this isn't a feature to leave on autopilot just because Intercom shipped it that way. It touches customer data retention and it changes what Fin is capable of referencing mid-conversation, and both of those are worth a few minutes of deliberate configuration rather than silent default behavior.

If you want help auditing how your Fin AI agent is actually set up, including things like this that ship on by default and quietly change behavior, that's the kind of tuning work I do for businesses running Intercom. Feel free to reach out if you want a second set of eyes on your setup.

Cover photo by Steve A Johnson on Unsplash.

Free checklist

The Fin AI launch checklist

The 12 checks I run before switching Fin on for a client: content, guidance, handoff, testing, and what to watch in the first month. Sent straight to your inbox.

One email, no spam. I only follow up if you reply.

Need help with Intercom? Book a strategy call to get personalized recommendations.

Book a Call

Leave a comment

Share your thoughts or questions below.

Latest articles

Browse every article