You can discover applicable regulatory obligations using either the chat interface or by building an agent. For one-time or ad hoc applicability analyses, we suggest using chat; however, if you are frequently analyzing new products or features, a simple agent can automate some portions of this workflow for you.

​
Getting Started

To get started, navigate to the Chat screen and start a new chat.

​
Inputs

You’ll want to start by sharing key information about your organization and your products or services, along with what any direction on what type of regulations or obligations you want to search for. When you give context about your organization, be sure to include key drivers of regulations (if you’re not sure, it’s okay to over-share). For example, you’ll want to share things like:
  • Jurisdiction(s) where you operate
  • Licenses or Charters you’re operating under
  • Prudential Regulators
  • Assets Under Management (AUM)
  • Products and Services you offer
Remember, it’s important to be as specific as possible in your prompt, and it’s also good practice to keep your questions fairly targeted (for example, asking about disclosure-related obligations will help you get a better response than asking about all regulatory obligations in a single prompt).

​
Text Prompts

The simplest approach is to provide a text-based prompt, like follows:
My organization is a Texas-chartered bank with FRB oversight. We offer consumer and business deposit accounts, and our assets under management are between $500 million and $1 billion. I need a list of of all of our regulatory obligations related to periodic disclosures. Please create a table containing all of our obligations in this area. The table should contain these columns:: Jurisdiction, Regulator, Regulation, Section, Summary of the Obligation.
The main drawback here is that you have to create your prompt on the fly, which will limit how easily you can share information.

​
Document Prompts

You can also add documents to your prompt. This let’s you combine a text-based prompt with references to more detailed information that will help the models retrieve more accurate rules.
My organization is a Texas-chartered bank with FRB oversight - please see the attached document that provides detailed information on the organization, including our assets under management (AUM) and descriptions of each of our products and services. I’m trying to create a list of of all of our regulatory obligations related to Periodic Disclosures, and I’d like you to share them in a table that matches the format of the spreadsheet I’ve attached here.
The specific guidance about output format from the table, and detailed context, will help the models to create a more precise analysis.

​
Crosswise References

You can also use references to different documents inside of your Crosswise account in your prompt. This is particularly useful for re-usability. For example:
Please use the information you have about my @Organization and my @Products to create a list of of all of my regulatory obligations related to Periodic Disclosures. I’d like you to share them in a table that matches the format of the spreadsheet I’ve attached here.

​
Outputs

Crosswise supports a number of different output formats. For your applicability analysis, we recommend either generating a spreadsheet or linking to Crosswise Rules. Generating a spreadsheet is great if you just want to find and download your rules, while linking to Crosswise Rules is best if you want to re-use your applicability analysis in Crosswise for use cases like Change Management, Gap Analysis, and Policy Writing.

​
Generate and Download a Spreadsheet

Crosswise will automatically create a spreadsheet for you when you ask the model to create a spreadsheet or a table (you can also use any similar meaning words like “csv” or “tabular data” and the model can figure out what you’re referring to). You’ll be able to see the table being created on the right side of your window, and when the model is done you’ll get a message notifying you that spreadsheet generation is complete. Once the model has completed generation, you can download your spreadsheet from the “Download” button above your spreadsheet. For re-usability, Crosswise Rules will provide dramatically more value than a spreadsheet. To link to Rules instead of creating a spreadsheet, you can ask the model to find and link to @Rules instead of creating a table. Here’s an example of how you can adjust your prompt:
My organization is a Texas-chartered bank with FRB oversight. We offer consumer and business deposit accounts, and our assets under management are between $500 million and $1 billion. I want you to find and all of the @Rules relating to Periodic Disclosures that apply to my organization, and link them to my account. Rules should be linked to both my organization and any @Products that triggered the Rule.

​
Advanced

​
Confidence Scoring

Ask the model to rate its confidence on a scale, and you can get even better precision and ability to review the output of your applicability analysis. For example:
When you produce the table of disclosures requirements, I want you evaluate the applicability of each requirements on a scale of 1 to 5, where: 1 = Very Poor Alignment, meaning the requirement is definitely not applicable 2 = Poor Alignment, meaning the requirement is unlikely to be applicable 3 = Moderate Alignment, meaning that you’re uncertain about whether the requirement is applicable or not 4 = Good Alignment, meaning the requirement is most likely applicable 5 = Very Good Alignment, meaning the requirement is definitely applicable
Then you can have the model do things like including the confidence score in its output, and even filter its output by score.
Please include the confidence score as an additional column the table. When you provide the final applicability analysis, only include requirements with a confidence score of 4 or 5.
Having the model go through these additional steps will also force it to “think” more deeply, and produce a better output while reducing hallucinations.

​
Model Explainability

You can also ask the model to explain its reasoning and include that in its output. For example:
When you come up with a confidence score, please explain your reasoning for the score in 3-5 sentences and include that in an additional column in your output.