Overview
AI Large Language Models (LLMs) are inherently probabilistic, making them less suitable for scenarios requiring fast and consistently 100% accurate results. However, integrating a Dataverse Agent with the North52 business rules engine offers a powerful solution—enabling the use of AI capabilities while maintaining strict accuracy and deterministic outcomes.
Here's why deterministic rules are essential for your Dataverse AI Agent:
- Compliance: Ensure regulatory requirements are always met
- Data Quality: Maintain consistent data validation and formatting
- Business Logic: Apply your organization's specific rules without exception
- Reliability: Remove uncertainty from critical decision points
This article shows you how to get deterministic rule-based outcomes in your Dataverse Agent.
Rules for Credit Card Eligibility
In our simplified example we have rules in a North52 decision table Formula which determines credit card eligibility for 3 types of card Platinum, Gold and Silver. The thresholds are specified across three decision sheets, one for each card type. These decision tables are shown in the Decision Suite Solution section below.
| Rule # | Description |
|---|---|
| 1 | Customer is eligible for a Platinum card if they meet the Annual Income, Total Assets and Credit Score thresholds as defined in the Platinum Rules Decision Sheet. They will be allocated a credit limit of 10% of their annual income. |
| 2 | Customer is eligible for a Gold card if they meet the Annual Income, Total Assets and Credit Score thresholds as defined in the Gold Rules Decision Sheet. They will be allocated a credit limit of 10% of their annual income. |
| 3 | Customer is eligible for a Silver card if they meet the Annual Income, Total Assets and Credit Score thresholds as defined in the Silver Rules Decision Sheet. They will be allocated a credit limit of 10% of their annual income. |
| 4 | If no thresholds are meet the eligibility result is false. |
North52 Decision Suite Solution
The North52 Decision Suite solution works like this:
- A Copilot Agent is set up with a Tool to call a Custom Action
- A Formula is set up on the Custom Action which:
- Evaluates eligibility based on the values provided by the Agent
- Returns a result, and if eligible a card type and credit limit value
- Copilot then uses this information to return a summary to the Agent
Set up Global Action, Formula and Copilot Agent
The first thing to create is the Global Action.
Global Action
Create a Global Unbound action with the following parameters:
- Inputs
- AnnualIncome / Integer / Required
- TotalAssets / Integer / Required
- CreditScore / Integer / Required
- Outputs
- Result / Boolean / Required
- CreditLimit / Integer / Optional
- CardType / String / Optional
Next, we create the Formula.
Formula
- Open the North52 App
- Navigate to Business Process Activities > Formulas
- Create a new formula, setting the following values in the Formula Guide:
- Source Entity set to N52 Command
- Set Formula Type to Action
- Select the Decision Table editor
- Change the Name of the formula to Action - Credit Card Eligibility
- Set the Action Event to the action you created above, in our case n52demo_CreditCardEligibility
- Click Save
- Build out the Decision Tables below using the Action parameters from the Source tab
- Select the Source tab
- Expand Processes > Actions
- Expand the none node (these are the global unbound actions)
- Expand the n52demo_CreditCardEligibility action
- For the Condition columns, select the cell in row 2 and then click on the Input parameter from the Source tab
- For the Action columns, select the cell in row 2 and then click on the Output parameter from the Source tab
- Enter in the threshold criteria as shown in the Decision Table sheets below:
Decision Table - Platinum
Decision Table - Gold
Decision Table - Silver
Finally, we create the Copilot Agent.
Copilot Agent
To utilize the rules logic within your Agent, you need to add a Tool.
- From within your Agent, select Tools
- Choose Add a Tool
- Select Microsoft Dataverse
- Select Perform an unbound action in selected environment
- Select a Connection
- Click Add and configure
- Complete the Details section:
- Provide a name and description which allows the agent to easily identify it
- Complete the Inputs section:
- Select the Environment and your Custom Action (created above)
- Label the input parameters appropriately, and ensure that under the advanced section the Should prompt user option is selected (this makes sure the agent asks for this input if not provided)
- Complete the Completion section
- In this example we chose the Write the response with generative AI option
Testing
Once the Tool has been configured, click the Test button to start a session to test your agent.
Experiment with both sentences and one word responses.
Sentences
Single word responses
