Skip to main content

Cross-sell and up-sell assistant

A chat agent that builds cross-sell and up-sell relations for you. Give it a product SKU or name, it finds products that go with it, proposes them in chat and, once you confirm, adds them to the product as relations.

Runs: Embedded in Propeller · Trigger: Chat · Difficulty: Advanced

Built on n8n's AI agent

This workflow uses the AI Agent node in n8n. You give it a goal and a set of tools, then it decides for itself which tools to call and in what order to reach a result. That flexibility has a cost. Two runs of the same request can come out differently. The result also depends on which chat model you connect.

What you need​

  • An n8n account. See Set up n8n.
  • A Propeller API key.
  • An OpenAI account for the chat model.
  • Owner access on the Configuration role to create the agent.

How it works​

You open the chat agent and type a product SKU or name. The agent then:

  1. Finds the product. It looks up an exact SKU directly. When you give a name it searches the catalogue instead. If several products match it asks which one you mean.
  2. Finds candidates two ways: products in the same category (a reliable source of alternatives and options) and a keyword search for complementary items in other categories. It leaves out the product itself and relations it already has.
  3. Proposes a short list, each with a suggested relation type (ACCESSORIES, ALTERNATIVES, OPTIONS, PARTS or RELATED) and asks you to confirm. With a single suggestion it simply asks yes or no.
  4. Adds the relations you confirm, then reports the result. If a relation cannot be added, for example because it already exists, it says so instead of reporting success.

It remembers the conversation, so the confirm step knows what it proposed. It writes only after you confirm. Any relation it adds can be removed again in the PIM.

The agent uses four tools against the Propeller API: one to look up a product by SKU, one to search the catalogue by keyword, one to list the products in a category and one to create a relation.

Install it​

  1. In n8n, import the downloaded JSON as a new workflow.
  2. Connect your OpenAI account on the OpenAI Chat Model node.
  3. Create a header auth credential with your Propeller API key and select it on all four tools: get_product, search_products, list_category_products and add_crossupsell.
  4. On the Webhook node, set a basic auth credential (a username and password you choose) and copy the production URL.
  5. Activate the workflow.
  6. Go to Admin > Agents and create an agent with Trigger Chat and Interaction Mode Multi Turn. Under Types, pick BACKOFFICE_PRODUCTS so it appears on the product catalogue. Paste the webhook URL and the same basic auth values. See AI Agents.
  7. Open the agent, type an SKU or product name, then confirm to test.
Filling in your API key

In n8n, create a credential of type Header Auth and fill in its two fields exactly like this:

FieldWhat to fill in
NameapiKey
ValueYour Propeller API key

Name is not a title you pick yourself. It has to say exactly apiKey, otherwise the workflow cannot sign in. See Set up n8n for the full steps.

Placement matters

An agent type is the page where the agent appears. Stick to the types listed in the install steps: the workflow depends on the data that page sends along, so placing the agent on other pages may not work.

Make it yours​

  • How it proposes: the system prompt on the AI Agent node controls how the agent searches for candidates and how it picks a relation type. Steer it toward accessories, spare parts or alternatives to match your catalogue.
  • The relation types: the agent chooses from ACCESSORIES, ALTERNATIVES, OPTIONS, PARTS and RELATED. These are the Propeller cross-sell and up-sell types.
  • The model: pick a different model in the OpenAI Chat Model node. A stronger model follows the multi-step tool flow more reliably, a smaller one answers faster. If you see the agent misbehave, for example add the wrong relation, try a stronger model first.
The agent decides when to write

The agent adds relations by calling a tool. The system prompt tells it to wait for your confirmation first, but an AI model can still act too eagerly, so treat the confirm step as a review, not a guarantee. Any relation it adds can be removed in the PIM.

Use it outside n8n​

Copy the JSON into an AI assistant and follow Adjust a workflow with an AI assistant.

Good to know

The workflows and agents in the Agent Hub are free to use and adapt. They are examples, not supported product features, so test them in a staging environment and check the results before using them in production.