AI Knowledge: Generative AI bot information source

If you use generative AI, the "AI knowledge" section of the automation section allows you to create sources of information that your respondents, bots and advisers, can use to handle conversations with your customers autonomously.

Principle

The information contained in your AI knowledge can be used for different purposes and come from different sources.


At present, you can configure 2 types of AI knowledge :

  • AI Knowledge of the "product catalog" type via API
  • "FAQ" type AI knowledge
NB : It is possible to use two types of AI knowledge at the same time, both for a standalone bot and for assistance on the console. However, they must be two distinct types of knowledge. You cannot, for example, use two types of FAQ AI knowledge.

 

Before configuring your information sources in the Automation section, make sure you have your product catalog and/or FAQ up to date.

 

The different types of AI Knowledge

AI Knowledge of the "product catalog" type via API

It is recommended that you integrate a "product catalog" type AI knowledge via API if you have a catalog with a large number of references for which the data is updated frequently (more than 100,000 items).

 

To synchronize your product catalog using our public API, please consult the dedicated article on our developer platform.

 

AI Knowledge of the "product catalog" type in a .csv file

In this paragraph, you'll find all the specifications for a "Product catalog" file.

We recommend that you integrate an AI Knowledge "product catalog" type in a .CSV file if :

  • your catalog contains fewer than 100,000 items, is no larger than 500 MB and available in less than 1 minute (timeout delay)
  • you have the URL of your .CSV file and a daily update suits your needs.

This catalog enables your respondent, bot, or advisor to refer to it to identify the products your visitors are looking for.

 

You can view your .csv file directly from iAdvize Administration, or choose to download it for safekeeping. To do this, go to the "Automation" > "AI Knowledge" tab and click on the "Browse" or "Download" button.

 

To ensure optimal operation, the information provided must be present, correct, and in a supported format. The following table shows the prerequisites for configuring your Knowledge and how to format the information on your product sheets so that it is supported.

 

NB : The various fields in the list below are essential for a product catalog. They are not mandatory, but they are indexed by the system. They should be filled in as far as possible, as they can be used for search options.

 

Data Required Description

Supported values

id

Required

Unique identifier representing a single product.

Once an id is assigned to a product and imported, the id may not be used for a different product in the same Knowledge Source.

Note: This information is required.

Exemple :

JEAN-12345-1

JEAN-12345-2

item_group_id Required for variants management

The item group id is an identifier used to group similar products.

It can consist of alphanumeric characters and is assigned to a parent-level product. This means that a product can be expressed as a collection of child products, and all of them can share the same item group id.

It's important to note that several products can have the same value for item group id since it can represent a collection of products (variants).

Exemple :

JEAN-12345

title Required for the Product Discovery use case

The item group id is an identifier used to group similar products.

It can consist of alphanumeric characters and is assigned to a parent-level product. This means that a product can be expressed as a collection of child products, and all of them can share the same item group id.

It's important to note that several products can have the same value for item group id since it can represent a collection of products (variants).

 
description  

La description du produit fournit des informations supplémentaires au-delà de son nom pour aider à répondre aux questions des visiteurs.

Le texte de la description sera automatiquement nettoyé des balises HTML afin d'assurer une meilleure compréhension par l'IA. (Note : non disponible pour les balises JSON et XML)

 
link  

He title is the product’s name, typically displayed on the product’s page. It will be used to display the product's name during conversations or research.

The title should include variant information like color, size, or others.

 
availability Required for the Product Discovery use case Information about accurate product availability that matches the product page and checkout. The supported values are :
  • in_stock
  • out_of_stock
  • preorder
  • backorder
availability_date  

The date a preordered product becomes available for delivery.

YYYY-MM-DD is the only date format supported.
price Required for the Product Discovery use case

The product's original price. It should match the standard price displayed on your product page.

It should include the currency in which the price is expressed.

Must consist of a figure for the value, and the ISO code of the currency.

 

Note : If the currency is not provided within the price, the default currency of the project will be used.

sale_price  

The product's original price. It should match the standard price displayed on your product page.

It should include the currency in which the price is expressed.

Must consist of a figure for the value, and the ISO code of the currency.

 

Note : If the currency is not provided within the price, the default currency of the project will be used.

product_types  

Product category defined for the product.

It should include ​​the full category.

Example: Home > Men > Clothes > Shirts

Example: Home > Men > Clothing > Trousers
gender   The gender your product is intended for.

The supported values are: 

  • male 
  • female 
  • unisex
brand   The brand name of the product, the one known by consumers to ensure the best user experience.  
color  

The product’s color.

May include multiple colors, list the primary color first.

Note: You can use item_group_id to regroup multiple color variants of the same product.

 
size  

The product’s size.

Note: You can use item_group_id to regroup multiple-size variants of the same product.

 
material  

The product's fabric or material.

It can be indicated by a primary material followed by secondary materials (using a separator like a coma, slash, or others).

Note: You can use item_group_id to regroup multiple material variants of the same product.

 
condition  

The condition of the product for sale.

Note: You can use item_group_id to regroup multiple condition variants of the same product.

The supported values are: 

  • new
  • refurbished
  • used
image_link Required for the Product Discovery use case Link to the main image of your product (URL).  
additional_image_link   Link to a secondary image of your product (URL).  
dimension_detail  

Dimension of your product

 ex : 1220x323x23

product_line_detail  

Customizable data field, using name/value description, for additional product info.

You can add several product_detail info:

  • “detail_name1”: “value”
  • “detail_name2”: “value”
  • “detail_name3”: “value”

 

Internally stored as objects with fields “attributeName” and “attributeValue”.

 

Ex: { “attributeName”: “city”, “attributeValue”: “Nantes” )

 

 

Example of a file to download

 

NB : All these fields are declared as plain text and are not limited in size. You can therefore add as much information as you like to your product catalogue. The more data you add, the better the AI will perform, since it relies on the information provided in the product catalog to answer visitors' questions. Consequently, if the data is not specified in the product catalog, it cannot be used.
In order to avoid duplication, a mechanism is in place to delete duplicates when a Product Catalog knowledge is downloaded.

 

AI Knowledge of the "FAQ" type in a .csv file

You can transform your existing FAQ into an AI knowledge, by integrating it into a very simple .csv file with 2 columns: one for the questions and one for the answers.

To ensure that the bot focuses on the right content, it needs to be well guided.
Here are a few best practices to improve its performance:

  • Structure
    • use "question" and "answer" column headers exactly as in the template file
    • don't use capital letters in headers
    • don't add columns to the file
    • one question = 1 subject and a precise, concise answer OR an explicit, descriptive title (the interrogative form is not mandatory)
    • favour more entries with synthetic answers rather than fewer entries with longer answers
      make sure that data is separated by a comma and not a semicolon, or any other separator (note: in its French version, Microsoft Excel uses semicolons when exporting .csv files).
  • Content
    • CSV files support special characters (such as accents, for example) only if the source file format complies with UTF-8 encoding standards.
    • CSV file does not support emojis
    • bot focus is optimized with responses between 1000 and 1500 characters
    • the most exhaustive content possible, fragmented into separate subjects, to limit hallucinations
    • opt for a continuous improvement approach to enrich the content of your FAQ: read conversations, take into account feedback from your agents, analyze the categorization of conversations.
    • the number of lines in the file is unlimited
    • the use of synonyms can facilitate bot detection and retrieval (e.g.: my mobile/phone is down)

 

AI Knowledge "FAQ" by PDF

You can now provide your AI Knowledge with information directly from PDF files. This method enables you to provide rich, detailed content to improve your bot's performance.

Structure and content

  • Import of all types of PDF.
  • Maximum size of 500 megabytes per PDF file.
  • Automatic conversion of information into FAQ (Questions/Answers).

 

Special features

  • Language adapts automatically to the project.
  • Existing files cannot be edited or modified. For any update, the old file must be deleted and the new one added.
  • Need to wait for "ready" status before using imported information, which may take a few minutes.

 

AI Knowledge de type "FAQ" par webscraping

You can now transform your existing FAQ into AI Knowledge by integrating information directly from web pages using the webscrapping method. This technique enables you to efficiently extract and structure content from your website to enrich your knowledge base.

 

Structure and content

  • Retrieve content from a web page by entering a URL.
  • URLs must be supplied one by one, but you can add as many as you need.
  • Extract information from each URL and transform it into FAQs (Questions & Answers).

Special features

  • No restrictions on web page formats.
  • The language automatically adapts to that of the website.
  • Page content is updated manually. If content changes, you can add and/or remove URLs, and knowledge will adjust accordingly.
  • Adding a URL will only provide the information visible on the dedicated page. You need to manually enter all the URLs of a website's pages to ensure that your bot has all the information it needs to function properly.
  • You need to wait for the "ready" status before using the imported information, which may take a few minutes.

 

 

NB : Other methods of retrieving your product information can be studied to meet your specific needs: Google Merchant Center-type XML file, calling up your own product API to synchronize data, etc. These customized projects require a support package. Contact your usual iAdvize contact for more information.

 

 

The "Explore" button is accessible for each source of information in the AI Knowledge section of iAdvize administration. It brings together all the information you have included in your FAQ or product catalogue. To help you find your way through this flow of information, the search function allows you to quickly find information using a keyword, which will be highlighted in yellow. The "infinite scroll" function lets you scroll through the information in each source without reloading the page.

 

Configuration

To add new sources of information, so that your respondents can be more precise and autonomous with your customers, go to the "Automation" tab and then "AI Knowledge".

Then click on "Create your first knowledge source".

Then choose the type of source you want to add:

  • Product catalog knowledge,
  • FAQ-type knowledge.

Discover in the videos below how to create information sources for your respondents and configure them directly from iAdvize Administration.

 

AI Knowledge "product catalog" by .csv file

 

AI Knowledge "FAQ" type by .csv file

 

Once your file is saved, remember to update your bot scenario:
In the "Automation" tab, click on "Edit Bot" for the bot that should use the new knowledge.
In the Generative AI card of this bot, remove the knowledge you no longer want by clicking on the cross next to its name. Then click on "+ Add" and choose the new knowledge.
Save the scenario (top right). A more or less extended loading time can be expected depending on the size of your file.

 

Tip: We recommend using a specific naming convention for your files to easily find different versions of the FAQs used. For example, "EN_FAQ_01_11_23" for your English version of FAQ from January 11, 2023.

 

NB : We apply a 3-year retention period to all .csv files linked to knowledge sources uploaded to the iAdvize platform.
We apply this retention period to ensure that the bots present on your sites continue to function over the long term.