How to Build an AI Upsell During Customer Support System

You can turn every live chat or ticket into a revenue opportunity by wiring a generative-AI model into your support platform, letting it suggest the next best product in real time. The result is a fully automated "AI upsell during customer support" flow that surfaces a personalized offer, sends the customer a checkout link, and logs the conversion back into your CRM - all without the agent having to type a single line.

Below is a step-by-step, production-ready guide that uses Intercom or Zendesk for the support channel, OpenAI for the language model, n8n as the glue, and Shopify (or any Shopify-compatible store) as the fulfillment engine. If you already have a CRM (HubSpot, Salesforce, etc.) you'll see where to plug it in.

What you need

Estimated build time: 6-8 hours for a developer familiar with REST APIs and basic workflow tooling.

Step-by-step build

1. Prepare the support platform webhook

Both Intercom and Zendesk can push a payload whenever a conversation is updated.

• Intercom: In the Intercom UI go to Settings → App → Webhooks and create a new webhook. Set the Event to conversation.user.replied and point it at https://your-n8n-host.com/webhook/ai-upsell. Leave the default "Send full payload".

• Zendesk: Navigate to Admin → Extensions → Target URLs → Add target. Choose HTTP Target, give it a name like AI Upsell, paste the same n8n URL, and select POST JSON. Then create a Trigger under Admin → Business Rules → Triggers: when Ticket is Updated AND Assignee is not empty, fire the target.

"A webhook that fires on every reply guarantees you never miss an upsell opportunity."

Make sure the endpoint is reachable from the internet. For local testing you can spin up ngrok http 5678 and copy the generated https URL into the webhook config.

2. Spin up n8n (self-hosted)

docker run -d \
 --name n8n \
 -p 5678:5678 \
 -e N8N_BASIC_AUTH_ACTIVE=true \
 -e N8N_BASIC_AUTH_USER=admin \
 -e N8N_BASIC_AUTH_PASSWORD=changeme123 \
 n8nio/n8n

• The command runs n8n on port 5678 with basic auth. Replace changeme123 with a strong password.
• Open https://your-host:5678 in a browser, log in, and you're ready to build the workflow.

3. Create the AI Upsell workflow

1. Webhook node - set HTTP Method to POST, Path to /webhook/ai-upsell. This is the entry point that receives the conversation payload.

2. Set node (Extract relevant data) - map the incoming JSON to a clean object:

   • customerId → {{ $json["user"]["id"] }} (Intercom) or {{ $json["ticket"]["requester_id"] }} (Zendesk)
   • latestMessage → {{ $json["conversation"]["conversation_message"]["body"] }} (Intercom) or {{ $json["ticket"]["description"] }} (Zendesk)
   • email → {{ $json["user"]["email"] }} or {{ $json["ticket"]["requester"]["email"] }}

3. HTTP Request node (OpenAI Completion) - configure as follows:
   

{
 "url": "https://api.openai.com/v1/chat/completions",
 "method": "POST",
 "headers": {
 "Content-Type": "application/json",
 "Authorization": "Bearer {{ $env.OPENAI_API_KEY }}"
 },
 "body": {
 "model": "gpt-4o-mini",
 "messages": [
 {
 "role": "system",
 "content": "You are a sales assistant for an e-commerce store. Suggest one relevant product upgrade based on the customer's last message. Keep the tone friendly and concise (max 50 words). Return only JSON: {\"product_id\":\"...\",\"reason\":\"...\"}."
 },
 {
 "role": "user",
 "content": "{{ $json[\"latestMessage\"] }}"
 }
 ],
 "temperature": 0.3,
 "max_tokens": 150
 }
}

• What this does: Calls OpenAI's chat endpoint with a system prompt that forces a short JSON response containing the recommended product_id and a brief reason.

1. Function node (Parse OpenAI response) - add a simple JavaScript snippet:

items[0].json = JSON.parse($json["choices"][0]["message"]["content"]);
return items;

• This extracts the JSON produced by the LLM into item.json.product_id and item.json.reason.

1. HTTP Request node (Shopify Checkout Link) - create a checkout URL for the suggested product:

{
 "url": "https://{{ $env.SHOPIFY_STORE }}/api/2023-10/checkouts.json",
 "method": "POST",
 "authentication": "basicAuth",
 "user": "{{ $env.SHOPIFY_API_KEY }}",
 "password": "{{ $env.SHOPIFY_PASSWORD }}",
 "body": {
 "checkout": {
 "line_items": [
 {
 "variant_id": "{{ $json[\"product_id\"] }}",
 "quantity": 1
 }
 ],
 "email": "{{ $json[\"email\"] }}"
 }
 }
}

• What this does: Calls Shopify's checkout API and returns a checkout_url that the customer can click to complete the upsell purchase.

1. IF node (Eligibility filter) - prevent upsell in low-value or support-only tickets. For example, only proceed if the latest message contains keywords like "upgrade", "more", "extra". Use a simple expression:

{{ $json["latestMessage"].toLowerCase().includes("upgrade") || $json["latestMessage"].toLowerCase().includes("extra") }}

1. Send Message back to Support Platform

• For Intercom, add an HTTP Request node targeting https://api.intercom.io/messages with the Authorization: Bearer <INTERCOM_TOKEN> header and a body:

 {
 "message_type": "inapp",
 "body": "Hey! Based on what you said, I think you'll love our {{ $json[\"reason\"] }}. 👉 {{ $json[\"checkout_url\"] }}",
 "from": { "type": "admin", "id": "<ADMIN_ID>" },
 "to": { "type": "user", "id": "{{ $json[\"customerId\"] }}" }
 }

• For Zendesk, use the Create Ticket endpoint https://yourdomain.zendesk.com/api/v2/tickets.json with a body that adds a public comment containing the upsell line.

1. CRM Update (optional) - add a HTTP Request node pointing at your CRM's "Create Deal" endpoint. Include customerId, product_id, and a flag upsell_generated: true.

2. Save & Activate - turn the workflow on. Test by sending a fake message through Intercom/Zendesk that contains "I want something bigger". You should see the AI suggest a product, n8n hit Shopify, and the support UI reply with a checkout link.

4. Secure your secrets

Create an .env file in the n8n Docker container (or set environment variables in your host) with:

OPENAI_API_KEY=sk-...
INTERCOM_TOKEN=...
SHOPIFY_STORE=yourstore.myshopify.com
SHOPIFY_API_KEY=...
SHOPIFY_PASSWORD=...

Restart the container so n8n picks up the new vars:

docker restart n8n

5. Monitoring and logging

• In n8n, enable Execution Logging (under Settings → Execution). Set the retention to 30 days.
• For production, forward logs to a log aggregation service (e.g., Papertrail) by adding a docker run -e LOGGING_PROVIDER=papertrail ... flag.

Where this breaks

"The weakest link is always the webhook latency; a 5-second delay can make the suggestion feel out-of-sync."

For a deeper technical reference, see n8n's documentation.

FAQ

How does the AI know which product to suggest?

The system prompt tells the model to "pick the most relevant product from our catalog". You can make it smarter by passing a few-shot example list of product IDs and descriptions in the prompt, or by enriching the prompt with the customer's purchase history fetched from your CRM before the OpenAI call.

Can I use a different e-commerce platform than Shopify?

Yes. Replace the Shopify checkout node with the equivalent API call for WooCommerce, BigCommerce, or a custom cart service. The only requirement is that the endpoint returns a URL that you can embed in the support reply.

What if the customer replies after the upsell suggestion?

Treat the conversation as a new event. The webhook fires again, the AI receives the latest message, and the eligibility filter decides whether to propose another upsell. You can also add a flag in the CRM (last_upsell_timestamp) to enforce a minimum gap (e.g., 48 hours) between suggestions.

Is it safe to expose the checkout link in a support chat?

Shopify's checkout URLs are single-use and tied to the email you pass in the payload, so they can't be hijacked easily. Nevertheless, enable HTTPS everywhere and never expose your API keys in the front-end. All secret handling stays inside n8n.

Do I need to train a custom model?

Not for most small-to-medium stores. The gpt-4o-mini model with a well-crafted system prompt yields accurate suggestions for a catalog of up to a few thousand SKUs. If you have >10 k products, consider adding a lookup step that selects the top-10 candidates by tag before sending the list to the LLM.

Where can I find more ready-made ideas?

Check out the AI automations you can sell page for a catalog of plug-and-play workflows, or grab the free guide at https://getaab.com/free to get template n8n JSON files you can import instantly.

By wiring together a support platform, OpenAI, n8n, and Shopify you get a scalable ai upsell during customer support engine that works on every ticket, respects the agent's workflow, and captures the extra revenue in your CRM. The pieces are all real-world products with documented APIs, so you can replicate the flow today, iterate on the prompts, and watch your average order value climb. Happy building.

Related reading

• how to build cold outreach ai agents with n8n and Perplexity API
• How to automate customer support with AI: Build a RAG-powered chatbot that knows when to escalate
• 24/7 AI Support Agent (RAG)