Skip to main content
If your account supports more than one product, you usually want IrisAgent to answer a ticket using only the knowledge that belongs to that product. A billing question should pull from billing articles, a shipping question from shipping articles, and so on. Product segmentation lets you do exactly that. It works in two halves that need to agree with each other:
  1. Article Product Classifier tags each knowledge base article with a product.
  2. Case Product Classifier tags each incoming ticket with a product.
When a ticket comes in, IrisAgent matches the ticket’s product to articles carrying the same product, so only the relevant knowledge is considered. If you only have a single product, you can skip this entirely.
Both classifiers live in the portal under Settings in the Ingestion section. They are optional. With no configuration, every ticket can retrieve from your full knowledge base.

​
How matching works

A ticket and an article are connected when they resolve to the same product value. The product value is just a label you choose (for example, billing or shipping). Use the same label in both classifiers so the two sides line up.
  • The Article Product Classifier decides an article’s product from its URL.
  • The Case Product Classifier decides a ticket’s product from its Zendesk brand ID, tags, keywords, or text.
Get the labels consistent across both and the right articles surface for the right tickets automatically.

​
Configure the Article Product Classifier

This maps knowledge base articles to products using patterns matched against each article’s URL.
  1. Go to Settings in the portal and open the Ingestion section.
  2. Click Configure on the Article Product Classifier card.
  3. Click Add Group and enter a Product Category (the product label, for example billing).
  4. Add one or more URL patterns (regular expressions) that match the article URLs for that product. For example, ^https://help\.example\.com/billing/.*.
  5. Repeat for each product, then click Save.
When an article URL matches a group’s pattern, that article is tagged with the group’s product.

​
Configure the Case Product Classifier

This maps incoming tickets to products. You can use any combination of the available signals:
  1. Zendesk Brand ID exactly matches the ticket’s source brand. This option is available only when Zendesk is the connected ticket source.
  2. Tags that were applied to the ticket at creation time (for example, a billing tag).
  3. Keywords found in the ticket subject or description. Keyword matching is whole word and case insensitive, so bill matches Bill but not billing.
  4. Regex patterns matched against the ticket subject and description.
To set it up:
  1. Go to Settings in the portal and open the Ingestion section.
  2. Click Configure on the Case Product Classifier card.
  3. Click Add Group and enter a Product ID (use the same label you used in the Article Product Classifier, for example billing).
  4. Add any combination of the signals that identify that product. For Zendesk, you can enter the numeric Zendesk Brand ID shown in Zendesk for that brand. Otherwise, add Tags, Keywords, or Regex Patterns.
  5. Repeat for each product, then click Save.
Each Zendesk Brand ID must contain only digits and can be assigned to only one product group. IrisAgent trims spaces from the beginning and end when you save the configuration.
New mappings apply only to tickets ingested after you save the configuration. Existing tickets are not reclassified.

​
Match precedence

When a ticket could match more than one product, IrisAgent decides by signal type in this order:
  1. For Zendesk tickets, an exact Zendesk Brand ID match wins first.
  2. If no brand ID matches, a regex match wins.
  3. If no regex matches, a keyword match wins.
  4. If none of the other signals match, a tag match wins.
Within regex, keyword, and tag signals, groups are evaluated in the order you arranged them, so the first matching group wins the tie. Duplicate Zendesk Brand IDs are rejected when you save the configuration.
The Case Product Classifier is config first. Once a group is configured, it fully owns how that account’s tickets are assigned to products.

​
Segment the chatbot by product

You can scope a chatbot to a single product so it only answers from that product’s knowledge. This is set on the chatbot configuration rather than on a ticket.
  1. Go to AI Products → Chatbot, click Configure and deploy, and select the chatbot configuration you want to scope.
  2. In the Product ID section, enter the product label you want this chatbot to use (for example, billing).
  3. Click Update Configuration.
The chatbot now retrieves only from articles that the Article Product Classifier tagged with that same product label.
Only set a Product ID on the chatbot if you have a matching product configured in the Article Product Classifier. The label must match exactly (a typo like billing versus Billing breaks the link). If the chatbot’s Product ID does not match any classified articles, every chat returns no answer because no knowledge is in scope. Leave the field empty to let the chatbot retrieve from your full knowledge base.

​
Example

A two-product account selling billing and shipping might configure: Article Product Classifier
  • Product Category billing, URL pattern ^https://help\.example\.com/billing/.*
  • Product Category shipping, URL pattern ^https://help\.example\.com/shipping/.*
Case Product Classifier
  • Product ID billing: Zendesk Brand ID 123456789, keyword invoice, regex (?i)\b(invoice|payment)\b
  • Product ID shipping: tag shipping, keyword ship
A Zendesk ticket from brand 123456789 matches billing before IrisAgent evaluates its text or tags. A ticket from another brand reading “my payment failed” matches the billing regex. In both cases, IrisAgent answers using only the billing articles.
Keep the product labels identical across the two classifiers. A typo (for example billing versus Billing) breaks the link and the ticket will not find its articles.