Task 1: Importing and exploring a sample decision service

You import and explore a sample decision service. The decision service uses a large language model (LLM) to extract data from customer emails, such as product names and item quantities, and computes a price.

Step 1: Importing a sample decision service

  1. Sign in to Decision Intelligence using your instance credentials.
  2. Create a decision automation.
  3. Click New decision +.
  4. Click Discovery tutorials and select Assisted pricing. Then, click Import.
  5. Open the Assisted pricing decision service.

Step 2: Exploring the Assisted pricing decision service

  1. Open the Pricing Data data model in the Data tab to review the defined types.
  2. Navigate back to the decision service name in the breadcrumbs and open the Email analysis generative AI model. It uses the meta-llama/llama-3-3-70b-instruct LLM to analyze the content of customer emails and extract data such as product names, item quantities, and company names.

    For more details about prompt fields, see Creating prompts.

  3. Navigate back to the decision service name in the breadcrumbs and open the Part number decision model:
    1. Click the Part number node.
    2. Open the Logic tab in the right panel and click Part number table.

      This decision table defines the part number of a product based on its name. If the product is not in the decision table, the output-default-setting default rule is applied.

    3. Go to the Run tab and run the decision model with the CorrectProductName test data set.

      The output is "OF-028".

    4. Switch to the MisspelledProductName test data set and click Run.

      The output is "N/A".

  4. Navigate back to the decision service name in the breadcrumbs and open the [Main] Pricing decision model:
    1. Click the Products node.
    2. Open the Logic tab in the right panel and click Get the list of products.

      This rule uses pattern matching functions to extract product names and quantities from the generated text. It then builds the product objects and returns a list of products.

    3. Go to the Run tab and run the decision model with the From a well written email test data set. Then, run the decision model with other test data sets.

      Depending on the email used as input, some products may display a price of 0 if their part number cannot be identified. When the product name does not exactly match the spelling used in the Part number decision table, the part number is set to N/A and a price of 0 is returned.

      This limitation can be avoided by replacing the decision table with a generative AI model to identify the correct part number for each product.