Intercom AI Chatbot Integration
Intercom has Fin, which is capable and worth trying first. This is for teams whose answers depend on data Fin cannot reach, or who want a bot they own rather than a per-resolution bill.
Discuss your projectWhen a custom bot alongside Intercom makes sense
Fin answers well from your help centre and public content, and if that covers most of your volume it is the sensible choice — no build, no maintenance, and it improves without you doing anything. We would check it against your actual ticket mix before proposing anything.
The pattern that leads teams to build is answers requiring lookups into their own systems: this customer's order, their account state, their entitlements, why their last job failed. Fin can be extended, and there are limits to how far, particularly when the answer requires combining several internal sources or applying logic specific to your business.
The second reason is commercial. Per-resolution pricing is straightforward and predictable, and at high volume it can exceed what a build plus model usage would cost. That is an arithmetic exercise rather than an argument, and we would do it with you honestly — including the cases where it says to stay with Fin.
Either way the conversation lives in Intercom. Your agents do not change tools, and the project is reversible.
How we integrate
Answers written into the conversation
Via Intercom's API as a bot participant, so the thread reads naturally and the whole exchange is in one place.
Handoff with context
Assigned to the right team with a summary note, the reason for escalation, and whatever the bot looked up already attached. The agent does not start from nothing.
Custom attributes set
On the contact and the conversation, so your existing Intercom workflows, views and reporting act on the bot's output exactly as they would on an agent's.
Canvas Kit apps
Showing the agent something useful in the inbox sidebar — account state, similar past conversations, the reasoning behind a routing decision.
Respecting your office hours
Different behaviour inside and outside them, with honest out-of-hours messaging rather than implying someone is about to reply.
Reversible
Switch it off and Intercom behaves as before, with your conversation history intact and nothing stranded elsewhere.
Practical notes
- Webhooks on conversation created and replied, so the bot responds within seconds rather than polling.
- Idempotent handling, because webhooks arrive more than once and a duplicated reply in a customer conversation is visible and embarrassing.
- Rate limits respected, which matters most during a backfill over historical conversations used as the training signal.
- Signature verification on inbound webhooks, so the endpoint cannot be driven by anyone who discovers the URL.
- Clear attribution in the conversation so customers know they are talking to an automated assistant, which is both good practice and increasingly a legal requirement.
- A mapping agreed up front of which system sets which attribute, so Fin and a custom bot are not both writing the same field.
Common questions
Related in AI Chatbot Development
Tell us what you need
Describe the problem and we will come back with scope, timeline, and a rate. No obligation to proceed.
Get in touch