TrustFrame's Rules engine is powered by an AI compliance review assistant persona. It's purpose is to help users understand the compliance analysis results for marketing documents. For best results, users should test Rules with sample documents and refine them to ensure accurate and consistent results. Admin users are encouraged to reach out to [email protected] if Rule building assistance might be needed.
To learn more about the inclusion of AI, please see our TrustFrame AI Compliance Assistant FAQ.
Rules Page
The Rules Page includes all of the accounts' existing Rules, including both Active and previously Disabled Rules, divided between the tabbed Views. In this grid view, Admin users can see when a Rule was created, it's current Severity level, the Description provided to the AI compliance review assistant, and more.
Components of a Rule
All components of a Rule are utilized by the AI compliance review assistant when creating a review analysis. Key components of a Rule include:
Risk Severity: users should select a Risk Severity rating for a Rule based on level of importance. Options include Info, Low, Moderate, and Critical in an ascending order. The higher the Severity of a Rule, the higher the Total Severity numerical score will be. (Note: the Risk Severity of "Info" Rules will not prevent a Submitter user from receiving an Auto-Approval status, but will allow Admin users visability for a particular Rule).
Name: the Rule Name will be a key component of the instruction provided to the AI compliance agent. The Rule Name should be a concise directive whenever possible.
Description: the Rule Description is where users may apply some further direction as to what the AI compliance assistant should be analyzing for. The Description section can include some additional qualifying information, such as key regulations to reference and examples of ideal findings. Users should avoid vague terms and should instead define measurable criteria, such as "flag any mention of interest rates without an accompanying APR disclosure".
Agent Instructions (optional): provide specific instructions for how the AI agent should evaluate the rule. Users should include any special considerations, edge cases, or exclusions to watch for, or guidance on severity assessment.
Categories: the Rule Categories may be utilized for reporting purposes and will often feature key reporting categories related to an applicable regulation (such as TILA, RESPA, etc.) or another type of policy election (such as State, Corporate policy, etc.). Rule Categories are highly customizable; users may pull from an existing Categories list or create new Categories at any time.
Tags: the Rule Tags may also be utilized for reporting purposes as they relate to content application (such as custom-created content or advertising pieces meant to be displayed in a particular state(s) like WA). Rule Tags are highly customizable; users may pull from an existing Tag list or create new Tags at any time.
Advice on Building Rules
To create a new Rule, users should select the Add Rule button at the top right of the screen. As these Rules provide instruction to the AI compliance review assistant, some key tips and tricks can assist users with creating Rule prompts for specific desired outcomes:
Ensure the Rule Name is a clear directive: the Rule Name should be a concise directive whenever possible. The more obscure the command, the less likely the desired response will be returned.
Incorrect outputs: as users begin to introduce more and more Documents into the system, there will inevitably be additional instructions and exceptions needed within a Rule to achieve the desired response. Users should be prepared to potentially make many Rule tweaks in the beginning to dial the AI compliance agent in to meet institution standards.
Outlining exceptions to the Rule: a key component of building a Rule should be any exceptions to a Rule. Users may use the phrase "exceptions include" or something similar within the Agent Instructions section for the AI compliance review assistant to reference.
Rules will evolve over time: business deicions, risk tolerance levels, and allowable exception may evolve for an institution over time. Users should be prepared to adjust Rules regularly to meet institution standards as they evolve.
Note: all components of a Rule are utilized by the AI compliance review assistant when creating a review analysis, including the Rule Name, Severity, etc. Users should keep this in mind when creating and modifying Rules.
Rule Example
Below is an example of how a Rule might be built. Note how there is mention of a specific regulation ("TILA"), examples of true findings ("as low as 6.5%"), and exceptions to the Rule ("Mention of closing cost savings does not count as a trigger term"):
Name:
"Improper Use of Trigger Terms"
Description:
The advertisement contains any of the TILA triggering terms, such as: the amount or percentage of any downpayment, the number of payments or period of repayment, the amount of any payment, or the amount of any finance charge WITHOUT ALSO disclosing: (i) The amount or percentage of the downpayment, the terms of repayment, which reflect the repayment obligations over the full term of the loan, including any balloon payment, the “annual percentage rate,” using that term, and, if the rate may be increased after consummation, that fact. For example, the advertisement mentions an interest rate ("as low as 6.5%"), a down payment amount ("only $10,000 down"), or a specific monthly payment ("payments of just $1,800/month"), but fails to include the full, required disclosures.
Agent Instructions:
Exceptions: Mention of closing costs or closing cost savings does not count as a trigger term. Mentions of loan terms in a small text disclosure should not be flagged. Use of the phrase "Interest rates" is not an issue unless a numerical value is also present. Mention of LTV does not count as a trigger term. An exception is the use of terms for a Fannie Mae & FHA product does not constitute a triggering term. Descriptions for loan products, such as "as low as 3% down payment financing available", should not be flagged.
Editing & Disabling Rules
Once a Rule has been created, Admin users may later edit components of the Rule and select the Save button or may elect to Disable the Rule entirely using the toggle at bottom of the pane.
