Creating A Model

To create a new model, click the Create Risk Model button and name your model. This name is currently uneditable later.

When creating a new model, there are three high level items to customize:

  • Risk Categories
  • Risk Inputs
  • Risk Weights

As you work on your model, make sure to save so you don't lose any progress. While there is a warning message if you try to leave the page before saving your changes, it is recommended to save your model after any major changes so you don't lose any progress. To save your model, hit the 'Save' button in the bottom right.

📘

When a new model is created, Unit21 will have out-of-the-box categories, inputs and weights, which can be customized to your needs. Unit21 can also provide a HIFCA & HIDTA matchlist if you need it, ask your Customer Success Manager or email [email protected] to get this enabled.


Risk Categories

There are seven types of ‘Risk Categories’. Risk Categories can be enabled or disabled if not relevant by adding or removing the desired categories. Below is list of all the Risk Categories.

  1. Address Risk - Is the account in a known risky location (e.g. on the HIFCA or HIDTA list)?

  2. New Member Risk - How tenured is the account?

  3. Customer Information Risk - Is there anything risky about the account type or age (e.g. Elder Abuse)?

  4. Suspicious Activity Risk - Are there any SARs associated with the account?

  5. Alert & Case Risk - Are there any risky alerts and cases associated with the account?

  6. Custom Data Risk - Are there any risky occupations on industries associated with the account? Is there anything unique to my business that is important to add to a risk score?

    1. There can be many ‘Custom Data Risk’ categories to better represent separate data points like Industry, Occupation, and Source of Funds.

  7. Rule Risk - Are there a high number or volume of transactions? Are clients sending money to risky counterparties?

    1. There can be many ‘Rule Risk’ categories to represent different types of transitional risk, like ACH, Wire, and Cash, and non-transactional data, like associated login events and device information.

    2. The ‘Rule Risk’ category works with both live models and shadow models so a rule does not have to generate alerts to be incorporated into the risk score.


When creating a new model, we will by default add the first five categories not Custom Data Risk or Rule Risk. All categories can be added or removed to ensure the model is relevant for your business.

To add a risk category, scroll to the bottom of the model model and click the 'Add Risk Category' button. This will bring up a modal of the risk categories you can now add. As mentioned above, the 'Rule Risk' and 'Custom Data' risk categories can be added as many times as you liked with friendly names, while the other categories, like 'Address Risk', can only be added once. If the 'Address Risk' category is already a part of your model, it will appear as an option to be added.


To remove a risk category click the red bin icon in the top right of the risk category.

Risk Inputs

Within each Risk Category, there will be one to many Risk Inputs. These are the inputs that will create the scores. To edit your risk inputs, click on any of the ‘Variables’ in the corresponding Risk Segment or click ‘Add Variable’ to add a new input.


Here is a detailed example how to update risk inputs:

  1. Click on a variable; e.g. ‘City’ in the ‘Address’ Risk Category.

  2. Use the ‘Add Value’ button to add a known risky city; e.g. ‘Monterey’ which is in on HIFCA list (reference: https://www.fincen.gov/hifca-regional-map).

  1. Alternatively, you can use the ‘Add Matchlist’ button. In this example, you can upload a matchlist of all cities on the HIFCA list (note, we would use zip codes for HIFCA when deploying a real model to avoid same named cities).

  2. Set the desired weight from 1-100. More on how weights are calculated in the ‘Weights’ section below.

In addition to the above example, here is an exhaustive glossary of items and actions that will be helpful when updating Risk Inputs.

Unknown Risk - this is when the data is not present. In this example, Unit21 has not received data that specifies the country for the entity. This is available for all Risk Parameters.

Unscored Risk - this is when the data is present but is not explicitly scored. In this example, the country for the entity is present (e.g. US) but that value is not in the ‘high risk places’ matchlist.

Add Matchlist - leverage matchlists for HIFCA & HIDTA lists of known risky locations. This can be used for any text value, like known risky device types. Using matchlists is recommended when you have a high number of risky values (e.g. a lot of countries or zip codes) or if the risky values will be dynamic in order to avoid re-deploying a new risk model to change values.

Add Value - this is used to add values one by one and is available for text fields. An example includes adding five risky zip codes with different weights of risk.

Add Condition - this is similar to ‘Add Value’ but for number inputs. Examples includes the number of SARs filed against an entity in the past 12 month or the number of months an account has been active.

Map External Risk Score - enables mapping internal or other external risk scores to the Unit21 risk model. This is available only in the Custom Data Risk Category.

Add Variable - adds a risk input to the corresponding Risk Category if more are available. This will appear in the bottom right of each Risk Category.

Default High - ensure high risk activities put entities into the high risk category. For any input, you will have the option to set it as a 'Default High' option. This will give a risk score of 100. This is especially beneficial for associated SARs, high risk Alerts & Cases, and PEP hits.


How Are Scores Calculated?

All entities will be given a score of 0-100 based on a weighted sum average. Each Risk Category has a customizable weight relative to other Risk Categories in the model. Within each Risk Category, the Risk Inputs have weights that are relative the other Risk Inputs in that Risk Category and corresponding Risk Category.


As an example:

  1. The example model has four risk categories: New Member Risk, Alert & Case Risk, Address Risk, and Suspicious Activity Risk.

  2. Each of the models are weighted equally; .5 in this example. This means each Risk Category is worth 25% of the score or 25 points out of 100. This is also shown the UI next to the weight in the model.

  3. After adding a Risk Input to the Address Risk model and giving a weight of 100 - e.g. an entity is located in Monterey - this will give the entity a risk score of 25 / 100.

  4. In this example, no other attribute in the Address Risk category will add to the score. The max contribution to the Address Risk category is 25. We do not add up the scores within one category. This is why we allow for many custom data risk and rule risk categories - so the scores across a variety of use cases, like Cash and Wire transactions, can be additive.

  5. If we changed the weight of the Risk Input to 50, the score would now be 12.5 / 100 for an entity that has an address in Monterey.

  6. Now we add a Risk Input to the Suspicious Activity Risk category and set it at a weight of 100. In this example, if an entity has one SAR associated with it and are based in Monterey, they will now have a risk score of 50 / 100.