AI Agent for Intercom: Supercharge Your Small Business Workflows
Empower your small business by integrating intelligent AI agents directly with Intercom. Automate repetitive tasks, save valuable time, and boost your team's efficiency to get more done with less.
Common Workflows for Small Businesses
- Instant responses in the Intercom chat widget
- Lead qualification during a chat conversation
- Routing conversations to the right team
- Creating support tickets from chats
What Your Small Business Needs: Required Permissions
- - Intercom API Access Token
- - Read/Write access to conversations and contacts
Where Your Small Business Data Comes From
- - Intercom conversation history
- - Your knowledge base
- - CRM data
Quick Setup: Implementation Timeline: 1-3 weeks
How We Set Up Intercom, Step by Step
1. Create a private app in the Intercom Developer Hub
In your Intercom workspace, open Settings, then the Developer Hub, and create a new app scoped to that workspace. A private app is the right choice here: it is visible only to your workspace and does not go through Intercom's public app review. The app is what issues the access token; do not use a personal token belonging to a teammate, because it stops working the day that teammate leaves.
2. Grant the four scopes the agent actually uses
Read conversations, write conversations, read contacts (users and leads), and read admins. Read admins is the one people forget: without it the agent cannot resolve which human teammate to assign a conversation to, so every handoff lands unassigned in the inbox. We do not request billing, article-authoring, or company-management scopes, and if your security review asks, that is why.
3. Point the webhook at your endpoint and subscribe to three topics
conversation.user.created, conversation.user.replied, and conversation.admin.closed. The first two are what wake the agent; the third is what tells it a human took over so it should stop replying. Intercom signs each delivery, so the endpoint verifies the signature before doing anything with the body.
4. Connect the knowledge the agent answers from
Usually your Intercom Help Center articles plus whatever lives outside it, a pricing page, a policy doc, a Notion space. We index that content rather than pasting it into a prompt, so an article edited in Intercom is reflected in the agent's answers without a redeploy.
5. Run it in a test workspace against replayed conversations
Before the agent ever answers a real customer, we replay a sample of your last few hundred conversations through it and read the transcripts with you. This is where the handoff rules get set, and it is the step that most often changes the scope: teams routinely discover a category of question they did not want automated at all.
6. Go live on a narrow slice first
One conversation type, or business hours only, with a human watching the inbox. We widen from there. A messenger agent that goes wide on day one is how a wrong answer reaches a thousand customers before anyone reads a transcript.
Supported Actions in Intercom
- Reply in an open conversation. Posts as the agent, in the thread, with the customer's history in context.
- Add an internal note. Used when the agent is confident enough to draft but not to send, so a human sees the suggestion without the customer seeing anything.
- Assign to a teammate or team. Requires the read-admins scope. We assign to a team rather than a person by default, because person-level routing breaks on holiday.
- Apply and remove tags. How reporting later answers 'what did the agent actually handle', and how you audit it.
- Update contact attributes. Writes qualification data (plan, company size, intent) back onto the contact so sales sees it in the same place they already look.
- Snooze or close a conversation. Only after a resolution the customer confirmed. We do not let the agent close on its own judgement.
- Trigger an outbound workflow. Hands off to an Intercom Series or to your own automation, for anything that needs to happen after the chat ends.
Limitations You Should Know Before You Start
These are the constraints we would rather you hear from us than discover in week three.
Rate limit: 10,000 API calls per minute per app
Intercom also caps a workspace at 25,000 calls per minute across all apps, and the quota is released in 10-second windows rather than all at once each minute, so a burst can be throttled even when the per-minute total looks fine. Our client reads X-RateLimit-Remaining and backs off on X-RateLimit-Reset instead of retrying blind.
The agent cannot see what it was never given
It answers from the sources you connect. If pricing lives only in a spreadsheet on someone's desktop, the agent will not know it and will say so rather than guess.
Attachments and images are not interpreted
A customer who sends a screenshot of an error gets routed to a human. We treat any conversation containing an attachment as a handoff trigger by default.
Voice and phone are outside Intercom
Intercom is the messenger. If you also want calls answered, that is a separate voice agent, and the two share a CRM rather than sharing an inbox.
What Happens When Something Fails
| Failure | What the agent does |
|---|---|
| Intercom API is down or returns 5xx | The agent stops replying rather than retrying into a broken workspace. Conversations queue and are delivered to the human inbox; nothing is silently dropped. |
| Rate limited (429) | Exponential backoff honouring the reset header. Sustained throttling raises an alert to your team, because at that point the agent is no longer responsive and a human needs to know. |
| The agent is not confident in an answer | It says it is checking with a colleague and assigns the conversation, rather than producing a plausible-sounding guess. Confidence thresholds are set with you during the replay step. |
| The customer asks for a human | Immediate handoff, no retention attempt, no 'are you sure'. This is non-negotiable in every deployment we ship. |
| Repeated negative sentiment | Escalates and tags the conversation so it surfaces in review, even if the customer never explicitly asked for a person. |
| A human replies mid-conversation | The conversation.admin.closed and reply events take the agent out of the thread. It does not talk over your team. |
FAQs for Small Business Integration
Can the AI agent see past conversations in Intercom?
Yes, the agent can be given access to past conversation history to provide more context-aware and personalized support.
What exactly do we have to give you access to?
A private app in your own Intercom Developer Hub with four scopes: read and write conversations, read contacts, and read admins. The app lives in your workspace, the token is yours, and you can revoke it at any time without contacting us. We do not ask for billing or article-authoring permissions.
What happens if Intercom rate-limits or goes down mid-conversation?
The agent stops replying rather than retrying into a broken workspace, and conversations fall through to your human inbox. Nothing is dropped silently. On a 429 we back off using Intercom's reset header, and sustained throttling raises an alert to your team, because at that point customers are waiting and someone needs to know.
Can the agent close tickets by itself?
Only after a resolution the customer confirmed, and only on conversation types you approved. Closing on the agent's own judgement is off by default: a closed conversation is one nobody reviews, so a wrong close is the most expensive mistake this integration can make.
Explore More AI for Small Business
Ready?
Put AI to Work for You.
Call now and talk to Aria, our AI strategist, or book a free 30-minute assessment.
Aria picks up instantly · 24/7 · Free assessment · 30-day guarantee