The Special Ad Categories setting on Facebook applies to ads that promote content related to sensitive areas such as:

- **Housing** — Ads about real estate, home rentals, or housing-related services.
- **Financial products and services** (expanded category as of October 2024) — Ads offering credit cards, loans, or other financial services such as insurance and investment advice.
- **Employment** — Ads about job opportunities or career services.
- **Social issues, elections, or politics** — Ads related to social causes, political campaigns, or elections.

## Why the Setting Exists

This feature ensures compliance with laws and regulations that prohibit discrimination based on factors like age, gender, ethnicity, and location. If your company’s ad falls under these categories, Facebook limits your targeting options to promote fairness and equal opportunity.

### Key impacts on ad targeting

- No targeting by age, gender, or ZIP code.
- Location targeting is restricted to a 15-mile radius.
- No Lookalike Audiences: You cannot use Lookalike Audiences for Special Ad Categories. Instead, Facebook offers “Special Ad Audiences” based on similar behaviors but without using protected characteristics.
- Custom Audiences are allowed, but with additional restrictions: you can still use custom audiences, but you must certify compliance with [Facebook’s advertising standard](https://transparency.meta.com/policies/ad-standards/).

## Facebook’s October 2024 special ad category announcement

In October 2024, Facebook announced a few significant changes to special ad categories:

1. A new special ad category “[Financial products and services](https://www.facebook.com/business/help/510724041294968)” was introduced, replacing the previous Credit ads category. This expanded the type of companies and ads that were subject to special ad category limitations.
2. Certification that any ad set based on a customer list complies with Facebook’s advertising standards.
3. [Restriction on usage of shared custom lists](https://www.facebook.com/business/help/1452187872132363) for custom audiences.
   1. Ad accounts within the same business portfolio that share customer list custom audiences should make sure that all people with permissions to manage ad campaigns have the same business email domain. If the domains don’t match, then those shared customer lists will be unavailable for ads about housing, employment, and financial products and services.
   2. Usage of shared audiences from different business portfolios will not be permitted for ads about housing, employment, and financial products and services.

### Use of Faraday in Compliance with the Special Ad Category

Since the emergence of the Special Ads category, Faraday has supported businesses whose ad sets are subject to its rules and restrictions. The primary impact of Facebook’s October 2024 announcement is to increase the type of businesses that will be subject to these rules and restrictions.

If you are a business, whose Facebook ad sets are labeled a special ad category, Faraday recommends the following steps to ensure compliance of any custom list generated through the Faraday platform:

1. **Segmented audiences** — If you are using Faraday cohorts to generate a custom list, you should avoid building cohorts that specifically segment on restricted traits like age and gender, restricted locations like zip code, or other [prohibited information.](https://www.facebook.com/business/help/361948878201809?id=188852726110565)
2. **Modeled audiences** — If you are using a Faraday outcome (aka predictive model) to generate a modeled audience, you should:
   - Use trait blocking when you [build your outcome](https://faraday.ai/docs/academy/dashboard/making-predictions) to ensure that all traits associated with protected classes (e.g. age, gender, ethnicity, religion marital status, etc.) are not considered when building a predictive model. This ensures that your models will not knowingly consider any data that would violate Facebook’s policy against discrimination.
     ![Screenshot of Faraday dashboard](/images/blog/facebook-special-categories-1.png)
   - Use Faraday’s built in bias mitigation to account for implicit bias. Simply blocking traits from model training, does not account for the fact that there is often a correlation between other Faraday traits and certain protected class traits. [Faraday’s bias mitigation,](https://faraday.ai/blog/faraday-bias-mitigation-machine-learning) allows a user to neutralize or even correct for historical bias within their lead and customer base.

![Screenshot of Faraday dashboard](/images/blog/facebook-special-categories-2.png)

Please see [this blog post](https://faraday.ai/blog/how-to-optimize-ads-for-compliance-in-finance), for a more in depth look at common Faraday supported ad strategies that comply with financial regulations.

How to use Faraday for ads when you're in a special category

Facebook special ad category overview

Faraday’s [customer behavior predictions](https://faraday.ai/features/customer-behavior-predictions) have a wide variety of use cases. From enhancing your customer engagement through personalization via [bespoke personas](https://faraday.ai/blog/how-they-did-it-bedding-brand) to embedding predictive lead scores in your [field sales app](https://faraday.ai/blog/salesrabbit) to increase rep productivity and drive growth, there’s no shortage of applications.

One application that’s proving to be more and more popular with our clients is geotargeting audiences based on the locations of physical stores and branches to enable personalized campaigns with a focus on retail shoppers. What this means is that these clients are able to specifically find the most likely buyers within a certain geographic radius around a point of interest, then run campaigns to draw people to their outlets, branches, and other physical locations.

In this guide, we’ll walk through how to create a geotargeted audience with Faraday.

## How to geotarget with Faraday

### Connect your data

To start, you’ll want to make sure that you have a [connection](https://faraday.ai/docs/abstractions/connections) created to wherever your data is stored. Doing this allows your customer predictions to seamlessly deploy back to your stack. If you’ve already got your connection created, double check that your [dataset](https://faraday.ai/docs/abstractions/datasets) is ready as well, as that’s how you’ll describe your data to Faraday–who’s who and what pieces of data mean what.

Both of these resources can be created with Faraday's API via the [createConnection](https://faraday.ai/docs/reference/createconnection) and [createDataset](https://faraday.ai/docs/reference/createdataset) requests, respectively. While these aren't required in order to complete the steps in this guide, they're required for you to be able to deploy your predictions directly back to your stack, or to add additional event-specific parameters mentioned in the [cohort section](https://faraday.ai/blog/how-to-create-a-geotargeted-audience-api#create-your-cohort).

### Create your place

In Faraday, you can create a point of interest–a **place**–around which you can create radius-based cohorts, or groups of interesting people. To do this, you’ll need your Faraday API key, which can be found in the _account_ tab of your [settings page](https://app.faraday.ai/settings/account).

[![Location of API key in Faraday](/images/blog/product/api-key-location.png)](/images/blog/product/api-key-location.png)

Once you’ve copied your API key, head over to the [createPlace](https://faraday.ai/docs/reference/createplace) request in API reference docs. Here, you’ll paste your API key into the `Token` field, then update the `addresses` field to include the address of the point of interest.

[![Location of createPlace API request in Faraday docs](/images/blog/product/geotargeted-audience/createplace-request.png)](/images/blog/product/geotargeted-audience/createplace-request.png)

A place can include more than one address, so if you’d like to take advantage of a radius around _all_ of your locations, you can add them all to a single place request within the `address` parameter. Here’s what your request might look like for a single address, though:

```bash
{
  "addresses": [
 {
   "city": "Burlington",
   "house_number_and_street": "431 Pine Street",
   "postcode": "05401",
   "state": "VT"
 }
  ],
  "name": "My Storefront"
}
```

Once your address info has been entered–don’t forget to give the place a name, like “My Storefront” above–then click **send API request** to create the place.

In the response window, grab the **place ID**, which you’ll use to create your radius-based cohort around your new place.

[![Location of createPlace API response in Faraday docs](/images/blog/product/geotargeted-audience/createplace-response.png)](/images/blog/product/geotargeted-audience/createplace-response.png)

### Create your cohort

With the place you’d like to geotarget your predictions around ready, you’re now ready to create the [cohort](https://faraday.ai/docs/abstractions/cohorts) of people that you’ll target with that place. To do this, you’ll use the [createCohort](https://faraday.ai/docs/reference/createcohort) request.

[![Location of createCohort API request in Faraday docs](/images/blog/product/geotargeted-audience/createcohort-request.png)](/images/blog/product/geotargeted-audience/createcohort-request.png)

In your request, you can optionally add parameters for things like `min_count` / `max_count` or `min_value` / `max_value` if you've created a connection and a dataset, but we’ll focus on the `place_conditions` parameter as that’s where you’ll specify the place you just created.

You can customize the cohort with `recency`, `frequency`, and `value` parameters based on the event you use to define the cohort, but for this example we’ll keep it simple and create a cohort using the following parameters:

1. A **name** for the cohort.
2. A **distance** of 10,000 meters around the place created above. Note that this value _must be in meters_.
3. The **place ID** saved from the createPlace request.

```
{
  "name": "Leads within 10km of storefront",
  "place_conditions": [
	{
  	"distance": 10000,
  	"place_id": "b1ab0efb-f732-478f-8310-6882383c4107"
	}
  ]
}
```

You can also optionally include an [event stream](https://faraday.ai/docs/abstractions/events) using the `stream_name` parameter if you’d like this radius-based cohort to also filter people that have completed an event of some kind.

With the createCohort request ready, click **send API request**, after which you should see a **201 created** response for a success.

Voilà, you’ve got a radius-based cohort created based on all your leads within 10km of your storefront.

With it, you can add it to a [pipeline’s](https://faraday.ai/docs/abstractions/pipelines) _population to include_ in order to target this group of people with your predictions. Add both their likelihood to convert via [outcome](https://faraday.ai/docs/how-to/predictions/likelihood-to-convert) as well as their customer persona via [persona set](https://faraday.ai/docs/how-to/predictions/customer-personas) to maximize effectiveness for your next direct mailer.

A guide on creating a geotargeted audience using Faraday's API.

How to create a geotargeted audience with Faraday’s API

In this guide, we’re going to walk through the technical steps Boll & Branch, a prominent D2C linens brand, took to configure Faraday to make the [customer persona predictions](https://faraday.ai/templates/thematic-personalization) they needed, and how they activated.

## Integrating with their BigQuery warehouse

Boll & Branch has a wealth of customer data in their BigQuery warehouse, and Faraday’s seamless [integration](https://faraday.ai/integrations/predictions-with-bigquery) with BigQuery was the first step in connecting their data with Faraday. The only info they needed on their end to get this connection set up was the ID of their project in BigQuery and the dataset’s name.

[![Boll & Branch's connection configuration in Faraday](/images/blog/howto/boll-and-branch/boll-and-branch-connection.png)](/images/blog/howto/boll-and-branch/boll-and-branch-connection.png)

From there, they created a Faraday [dataset](https://faraday.ai/docs/abstractions/datasets) to describe their data: which rows in their data represent what pieces of customer data? Creating this dataset served the brand two purposes, both key cornerstones in powering their customer predictions:

1. It enabled Faraday to identify people in their data through an identity set.
2. It pointed Faraday at the [events](https://faraday.ai/docs/abstractions/events) occurring in their data, which is how they’ll make their predictions–more on that later.

[![Boll & Branch's dataset configuration in Faraday](/images/blog/howto/boll-and-branch/boll-and-branch-dataset.png)](/images/blog/howto/boll-and-branch/boll-and-branch-dataset.png)

In their identity set, the team at Boll & Branch matched the people in their data–their customers in this particular use case–into the [Faraday Identity Graph](https://faraday.ai/features/faraday-identity-graph) by telling Faraday which rows in their data represented personally identifiable information (PII): name, email, physical address, and so on.

They then defined two events:

- “Customers,” which looked at the date of someone’s first order to define them as a customer.
- “Orders with product category,” which pointed toward an event property that specified which category in their catalog the product purchased is in. They planned to use this to see which category their customers were purchasing from so that they could offer recommendations.

## Defining their customers

With their data connected, the next step was to create [cohorts](https://faraday.ai/docs/abstractions/cohorts) of interesting people, or people that they wanted to make predictions on. As their team was initially focused on identifying who their current customers really are as people and consumers, their [_Customers_](https://faraday.ai/docs/how-to/cohorts/customers) cohort was where they began.

The brand defines their customers as people that have made a purchase, and so they chose the “customers” event that they created earlier on in their dataset to define their _Customers_ cohort.

[![Boll & Branch's cohort configuration in Faraday](/images/blog/howto/boll-and-branch/boll-and-branch-cohort.png)](/images/blog/howto/boll-and-branch/boll-and-branch-cohort.png)

Creating this cohort grouped everyone in the brand's data that had made a purchase together, and any time a new customer made a purchase, they’d automatically join this _Customers_ cohort thanks to the connection to BigQuery they made.

## Predicting customer personas

Now that their pieces were in place, Boll & Branch's team was ready to make their first prediction: [personas](https://faraday.ai/docs/abstractions/persona_sets). They navigated into [persona sets](https://app.faraday.ai/personas) and created a new persona set using the _Customers_ cohort they just created.

[![Boll & Branch's persona configuration in Faraday](/images/blog/howto/boll-and-branch/boll-and-branch-personas.png)](/images/blog/howto/boll-and-branch/boll-and-branch-personas.png)

In taking a look at the freshly-clustered persona set for their customers, they were able to quickly identify that their demographic didn’t skew quite as old as they thought it did. Additionally, they found that their highest-value and largest persona at 55% of their persona set, nicknamed “Jen,” were heavy social media users through Faraday’s relevant traits like `Active on social media` and `Instagram user`. This insight gave them clear direction as to where they should focus marketing efforts.

## Deploying their predictions with a pipeline

Analyzing their persona set within Faraday’s Dashboard gave their team a high-level overview of their customer makeup, and they were soon ready to deploy their predictions back into their BigQuery warehouse for activation. To do that, they needed a [pipeline](https://faraday.ai/docs/abstractions/pipelines).

In their pipeline, they added their _Customers_ cohort to the **population to include**, as they wanted to make predictions on their customers. Next, they grabbed their _Customer personas_ persona set for their **payload** so that they could apply their persona predictions to all of their customers.

[![Boll & Branch's pipeline configuration in Faraday](/images/blog/howto/boll-and-branch/boll-and-branch-pipeline.png)](/images/blog/howto/boll-and-branch/boll-and-branch-pipeline.png)

While the pipeline was building in the background and persona predictions were being applied to their customers, the team moved on to deploying their predictions back to their stack with a [deployment](https://faraday.ai/docs/abstractions/deployments).

They knew they wanted to sync their predictions back into their BigQuery warehouse, so they created a referenced deployment, which was an easy way to sync data back into their stack via a reference key–a way for them to specify which column in their data Faraday should use to match their predictions to the appropriate person.

[![Boll & Branch's deployment configuration in Faraday](/images/blog/howto/boll-and-branch/boll-and-branch-deployment.png)](/images/blog/howto/boll-and-branch/boll-and-branch-deployment.png)

With their reference key set as _email_, every customer had their persona applied to them, and their data was then synced back into BigQuery based on the email column.

Now that everything was in place on the Faraday end for their persona predictions, they turned off [preview mode](https://faraday.ai/docs/abstractions/pipelines#preview-mode) so that their pipeline would constantly stay up-to-date as new customers entered their _Customers_ cohort.

## Making practical use of their predictions

In BigQuery, Boll & Branch’s analysts were able to manipulate their Faraday deployment to discover which of their customer personas over-indexed on what products by tallying products-per-persona, what colors of products they preferred, and more. This was key for them, as it enabled them to recommend personalized [next best offers](https://faraday.ai/templates/next-best-offer). Once they were satisfied with their findings, they pushed their predictions out to [Iterable](https://iterable.com) \- their goal was to launch an email campaign with their newfound personalization insights.

Drawing on the analysis they performed on each persona in Faraday’s Dashboard and the post-analysis they performed for product recommendations, they created a different [Iterable List](https://support.iterable.com/hc/en-us/articles/204780549-Lists-Overview) for each persona, then applied copy & creative based on the traits they discovered in Faraday’s Dashboard.

The result of applying these persona predictions to their email efforts was a **30% lift in conversion rate**.

Part of our "How they did it" series, this entry highlights exactly how Boll & Branch predicted their customer personas in Faraday, then deployed their predictions to Iterable for email activation.

How to: enhance your customer data with rich personas

Our [Powered by Faraday](https://faraday.ai/solutions/powered-by-faraday) partners work with Faraday to supercharge their features with customer predictions, ensuring their customers receive the best solutions on the market.

To accomplish this goal, they need a subaccount for each of their customers so that all data is siloed and secure. In this guide, we'll walk through the steps to create a subaccount.

## Finding your Faraday API key

The first thing you'll need in order to create a subaccount is your Faraday API key. You can find this in the Faraday Dashboard, under **Settings** in the left-hand nav, then **Account**. Click **copy** next to _production environment_.

[![Location of API key in Faraday's Dashboard](/images/blog/product/api-key-location.png)](/images/blog/product/api-key-location.png)

## Creating a subaccount with the Faraday API

Next, head over to our [API reference docs](/docs/reference), and specifically the [createAccount](https://faraday.ai/docs/reference/createaccount) request.

Here, you'll want to do three things:

[![Arrows pointing to three steps to create a subaccount in Faraday's reference documentation](/images/blog/product/api-request.png)](/images/blog/product/api-request.png)

1. First, paste your API key into the **token** field.
2. Next, replace "Acme Wind Turbines" with the name you'd like to give your subaccount. We suggest something like "Your Company Name - Customer Company Name."
3. Last, click **send API request**.

With these three steps done, your new subaccount will be accessible via the account switcher in the upper left corner of the Dashboard.

[![Location of account switcher in Faraday](/images/blog/product/account-switcher.png)](/images/blog/product/account-switcher.png)

A short guide on creating a subaccount using Faraday's API.

How to create a subaccount using Faraday's API

At Faraday, we’re on a mission to help [businesses of all kinds](https://faraday.ai/stories) predict customer behavior. Whether you’re a [credit union](https://faraday.ai/stories/advia-credit-union) that wants to predict the next best offer for your current customers, a [construction brand](https://faraday.ai/stories/window-manufacturer) looking to increase lead conversion rates, a [sales platform](https://faraday.ai/stories/salesrabbit) set on enhancing your product with customer predictions to increase rep efficiency–or anything in between–we’ve got you covered.

While we’ve taken great leaps in the 10+ years since our founding to make this type of data science more accessible to non-data-scientists, we know that it can still be a struggle to interpret what exactly you should pay attention to within your outcomes and recommenders so that you can use your customer predictions with maximum effectiveness.

To this end, let’s take a quick look at an example of the output of an outcome and highlight the exact pieces of info you can focus on so that you can understand how to activate on your predictions. For recommenders, check out our [accompanying blog post](/blog/how-to-interpret-recommenders).

## Using the lift table

When you’re viewing an [outcome](https://faraday.ai/docs/abstractions/outcomes), the **lift table** is one of the analysis options available to you, and a great place to start.

Every outcome has an _average conversion rate_ — this is the portion of every eligible person that attained the outcome. Given a lead conversion outcome, for example, the average conversion rate is just the average lead conversion rate you’ve experienced to date.

In the lift table, you’ll find performance broken down by probability range into 10 groups, or _deciles_. The table sheds light on how likely people in each decile are to attain the outcome, including in comparison to that average attainment rate. The comparison of a given decile’s attainment rate to the average attainment rate is called **lift**, which you can see in the rightmost column. The higher the lift, the better the performance.

[![Faraday outcome's lift table with decile and lift highlighted](/images/blog/product/lift-table.png)](/images/blog/product/lift-table.png)

As you review the table, the columns you’ll want to keep an eye on are **decile** and **lift**. You can look at the lift on a scale around the value 1.0, which represents no change against your average conversion rate. When the lift moves positively away from 1.0, that means you’re in a decile that will do better than normal, while moving below 1.0 indicates the opposite.

To illustrate, look at the top decile — this contains the 10% of all people who scored highest. Faraday expects that out of everybody who receives a score this high, about 95% of them (the **precision**) will attain the outcome. This is way higher than normal for this outcome — in fact, about 90% higher (1.896x lift).

## Choosing a cutoff

If you only pursue this top decile, you’ll have a great success rate, but you’d be leaving a lot of other potential attainers behind. When predicting customer behavior like this, an important job for a human is to decide where the cutoff point (“threshold”) should be.

In general, you can use this scale to determine which people in your pipeline’s eligible population you should focus more on, and which you should focus less on. In the above table, you can see a positive lift all the way to the top 50% of scorers.

When you create a [deployment](https://faraday.ai/docs/abstractions/deployments) for your predictions, you can match this by applying a filter on the outcome’s probability percentile at “\> 50,” ensuring you’re focusing on places where you’re likely to see positive lift.

Advanced users can leverage probability values, along with profitability, to determine a cutoff based on expected value (EV).

A short guide on how to interpret your predictive outcomes in Faraday.

How to interpret outcomes in Faraday

At Faraday, we’re on a mission to help [businesses of all kinds](https://faraday.ai/stories) predict customer behavior. Whether you’re a [credit union](https://faraday.ai/stories/advia-credit-union) that wants to predict the next best offer for your current customers, a [construction brand](https://faraday.ai/stories/window-manufacturer) looking to increase lead conversion rates, a [sales platform](https://faraday.ai/stories/salesrabbit) set on enhancing your product with customer predictions to increase rep efficiency–or anything in between–we’ve got you covered.

While we’ve taken great leaps in the 10+ years since our founding to make this type of data science more accessible to non-data-scientists, we know that it can still be a struggle to interpret what exactly you should pay attention to within your outcomes and recommenders so that you can use your customer predictions with maximum effectiveness.

To this end, let’s take a quick look at an example of the output of a recommender and highlight the exact pieces of info you can focus on so that you can understand how to activate on your predictions. For outcomes, check out our [accompanying blog post](/blog/how-to-interpret-outcomes).

## Understanding AROOP

A [recommender](https://faraday.ai/docs/abstractions/recommenders)’s _metrics_ tab will show you the average rank of observed property (AROOP), which represents the average ranking of the true next event during training. In other words, the average number of recommendations it took to recommend the correct actual behavior.

What’s an observed property? Many brands use a Faraday recommender to identify the so-called “next best offer.” In this scenario, the event would be “transaction” and the property would be “product.” A recommender’s job is to predict which _product_ property will occur in the person’s next _transaction_.

Let’s look at an example for a fruit seller. In the below table, John’s actual behavior landed in one of the “top 2” recommendations from Faraday, while Jane’s was in the top 1\! The overall AROOP for the recommender would be the average of these ranks: **(<span style="color:red">2</span> \+ <span style="color:green">1</span>) / 2 \= 1.5**.

<table>
  <thead>
    <tr>
      <th>Name</th>
      <th>Recommendations</th>
      <th>Observed property</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>John</td>
      <td>
        <ol>
          <li>
            Apple
          </li>
          <li>
            <strong><span style="color:red">Orange</strong></span>
          </li>
          <li>
            Kiwi
          </li>
          <li>
            ...
          </li>
        </ol>
      </td>
      <td><strong><span style="color:red">Orange</strong></span></td>
    </tr>
    <tr>
      <td>Jane</td>
      <td>
        <ol>
          <li>
            <strong><span style="color:green">Kiwi</strong></span>
          </li>
          <li>
            Apple
          </li>
          <li>
            Orange
          </li>
          <li>
            ...
          </li>
        </ol>
      </td>
      <td><strong><span style="color:green">Kiwi</strong></span></td>
    </tr>
  </tbody>
</table>

## Using the accuracy table

Looking at a recommender’s **accuracy table**, accuracy refers to how often the observed property is contained in the “top X” recommendations.

For example, if the accuracy for a threshold of 3 is 0.56, then this Faraday recommender correctly predicts the observed property to be in the top 3 recommendations 56% of the time.

[![Faraday recommender's accuracy table](/images/blog/product/accuracy-table.png)](/images/blog/product/accuracy-table.png)

Depending on your use case and the medium you’re using recommendations in, you can use accuracy up to the appropriate threshold that makes sense for that medium. Choosing the threshold is the equivalent of finding the balance between the accuracy of the recommendation and overwhelming information. If you recommend _everything_ to the customer, you’re guaranteed that their next action will be one of your recommendations, but that’s not very helpful\!

If you want to take a deeper look at how each trait (feature) impacted your predictive models, check out our blog on understanding [model reports](https://faraday.ai/blog/faraday-model-reports).

A short guide on how to interpret your predictive recommenders in Faraday.

How to interpret recommenders in Faraday

Let's take a look at some of the latest features in Faraday.

## Event stream analysis

[Events](https://faraday.ai/docs/abstractions/events) are the most critical pieces of data used in Faraday–they help define nearly everything you do, after all. When an event stream that you’ve created stops receiving data for one reason or another, it could be frustrating to discover that you’ve been working with data that is no longer up-to-date.

To help alleviate this frustration, we’ve updated the events index view to display the number of events being used in each event stream, when the most recent event was added, and the status of the event stream. Additionally, each event stream now includes an analysis view that indicates these same event counts.

[![Screenshot of event stream analysis in Faraday](/images/blog/product/event-stream-analysis.png)](/images/blog/product/event-stream-analysis.png)

Overall, this change should help you stay on top of your data flow into Faraday and ensure that your predictions are always up-to-date using your latest events.

## Enrichment precision filtering

In identified and referenced [deployments](https://faraday.ai/docs/abstractions/deployments), an enrichment precision column is included for each trait provider your account is subscribed to. Enrichment precision tells you how well Faraday was able to assign traits to identity information you provided, and is broken down into two levels: person and residence.

- Person enrichment precision means that Faraday is able to assign person-level traits to an identity, like age, consumer habits, or marital status.
- Residence enrichment precision means that Faraday is able to assign address- or property-level traits to an identity, like home value or number of bedrooms.
- Region (postcode) enrichment precision means that Faraday is able to assign region-level traits to an identity, like average commute time or average property tax.

## Lookup API support for identity sets

For our [Lookup API](https://faraday.ai/docs/features/lookup-api) users, you’re now able to use multiple identifiers in a single identity_sets array to find a match for a person. If the root-level identifier does not find a match, Faraday will use each included identifier until a match is found.

[![Screenshot of identity sets used in an API request](/images/blog/product/lookup-identity-sets.png)](/images/blog/product/lookup-identity-sets.png)

With this update, your Lookup API workflow should be simpler and more streamlined when it comes to finding individuals in the Faraday Identity Graph.

---

Check out our [changelog](/changelog) for a dated list of updates.

Explore the latest Faraday updates, including tracking of event stream change over time and filtering based on Faraday enrichment type.

What's new: event stream analysis, enrichment precision filtering & more

It’s no secret that [third-party data](https://faraday.ai/blog/first-party-vs-third-party-data-marketing) is one of the most critical components of effective, data-driven AI/ML projects. It adds additional layers of depth and complexity to predictive models to help them perform better, provides rich insight into who your customers and leads are outside of their engagement with your business, and more.

In Faraday, responsibly sourced third-party data is [built in](https://faraday.ai/features/faraday-identity-graph), and we recently pushed our biggest-ever update to the Faraday Identity Graph (FIG). 827 brand new traits alongside fresh updates to hundreds of existing FIG traits means that your customer predictions are now even more powerful–without any action required on your end, as Faraday automatically considers these new traits when building models for you.

From lifestyle [traits](https://faraday.ai/docs/abstractions/traits) like `Environmentally focused household` to financial segments like `Affluents`, Faraday’s algorithms have even more lenses through which your customer data can be analyzed, resulting in even more powerful predictions.

For example, a financial institution might find that a prominent feature in their [outcome](https://faraday.ai/docs/abstractions/outcomes) is another new trait, `Likely to use an investment broker`, which could indicate that they should leverage their brokerage service more heavily. They could append this trait to their pipeline that includes a likelihood to convert outcome so that, maybe instead of sending an email to people for whom this trait is true, they reach out directly via phone to talk about investment services.

Similarly, a business looking to focus their ad spend might use `Instagram user` to find their leads that are most likely to be active on Instagram. With this knowledge, they can then target their ads specifically on Instagram as opposed to using a widespread approach to see what sticks.

Predicting customer behavior is nothing without a strong foundation of first and third-party data. Faraday’s data is regularly refreshed with both updated and new traits so that your predictions are as accurate as possible. If you’re interested in learning more about these new traits, contact [support](https://support.faraday.ai) or your account manager.

The Faraday Identity Graph recently received its biggest-ever update, with over 800 brand new traits and updates to hundreds of existing traits.

1,000+ new and updated traits now live in Faraday

At Faraday, we automatically detect and correct for geographic bias in our training data through a process we call “geonormalization”.

## Why geonormalize?

It's easiest to explain through some examples, so let's walk through two.

**Example 1: to re-balance an unbalanced training population**

Say you have a business with two customer bases: California and Nebraska. The business started in California, so its customer base in California is 10x what it is in Nebraska. Without geornormalization, because the business has more customers in California and those customers look totally different to those in Nebraska (e.g. higher household income in California), everyone in California is scored higher than those in Nebraska.

**Example 2: to allow for a prediction to generalize over geographical areas not covered in the training data**

If your business started in New York, so all of your data–demographics, financials, and so on–is New York-based, and now you're trying to find _brand new_ business in California, the predictive model won't have been able to learn about what makes California unique in terms of that data, so the model will struggle to learn and its predictions will be unreliable.

Neither scenario is particularly helpful for growth!

Geonormalization adjusts both of these types of geographic bias with geography-based scaling instead of a raw value: Jane Doe is in the top 10% of scores in her state instead of the top 10% of scores for everyone.

Without geonormalization, you might define someone as "wealthy" if their net worth is $500k. Having a definition this static can present problems when comparing people from different demographic and financial backgrounds. With geonormalization, and keeping the above examples in mind, the definition of wealthy becomes more dynamic: "they have a net worth over twice as much as the average in their county."

Just because someone lives in a location that’s different from your customer data doesn’t necessarily mean that they should score lower than everyone that resembles your customer data.

## How does Faraday geonormalize?

Hence the need for a correction factor, to ensure that we are able to find the most likely customers in all of the regions our client wants to expand into. Our approach requires a detection algorithm and a correction algorithm

### Detection

We use a simple, but effective, algorithm. We compare the market penetration (customers / total possible customers) among each distinct geographic area at the desired level of analysis. If the largest market penetration is greater than 100x the smallest market penetration then we perform our correction algorithm during training.

For detections we start at the largest geographic area of interest, the state level, and work our way down to the smallest geographic area of interest, the postcode level. If at any point during the detection, the 100x threshold is met we perform the correction for that geographic level. Therefore, we are always applying the correction at the largest level possible.

### Correction

We use a two-pronged approach to correction: instance weighting for the training examples, and bias correction for the training features.

#### Instance weighting

We assign a weight to the positive examples calculated from the penetration rate of those positive examples in their respective geographic area. This in essence levels the playing field and gives a boost to the positive examples from areas where there are relatively fewer examples (unsaturated markets) and gives a penalty to the positive examples from saturated markets.

From a modeling perspective, we are tricking the model to believe that the good examples are just as likely to exist in any of the geographic segments. This leads to less bias amongst the features used to distinguish good examples from bad examples.

#### Feature correction

Our other method we employ in tandem with the instance weighting, is to transform features so that we level the playing field across the geographic segments. Instead of providing the model with raw features, we transform them so each person is represented by the percentile they occupy within their segment. Again, this one is probably easiest to explain via an example.

Instead of providing raw household income values of \$84k for CA resident A and \$66k for NV resident B, we provide the model with how they rank amongst their peers within their respective states. I’ve chosen the median income for each state, so the model would receive 50 for resident A and 50 for resident B.

By correcting the features, we provide the model with fewer discriminatory features with which to determine that everyone from the more saturated geographic segment is a better fit than everyone from the less saturated segments.

## Conclusion

Geographic bias correction, aka “geonormalization”, means that when using Faraday to expand your business into new markets, you can do so confidently knowing that our models will not just default to recommending people from markets in which you already have a significant presence. Grow your business confidently with predictions from Faraday.

At Faraday, we automatically detect and correct for geographic bias in our training data through a process we call “geonormalization”.

Geonormalization in Faraday

Predictive patterns change over time. When it comes to predicting customer behavior, it’s entirely likely that a person who’s been a lead for say, three months, is going to behave differently than someone who just signed up for your mailing list. The former has been receiving your email drip for a few weeks, is aware of your offerings, and probably has a decent idea of what they’re looking for. The latter is much closer to the top of the funnel, and is just getting to know your business better.

Why does this matter? Well, the person that just subscribed to your mailer might be less likely to convert compared to someone who’s been interacting with your business for 3 months - or they might not be! If the behavior “visited your site 3x” tends to be highly predictive of conversion, that’s only relevant to leads who’ve had the time in the funnel to actually visit your site more than once, so a different model should be used for fresh leads. Additionally, aggregated data is used to capture the situation of a person at the time of prediction. For example, has a new email been sent since we last produced a prediction?

With dynamic prediction, we create predictive models for several tenure bins, or periods of time, that someone has been in the eligibility cohort that you’re using for your prediction (i.e. their funnel stage). Then, we use the best possible model to make predictions on an individual based on that tenure bin. In the below example, you can see two strategies applied (A and B) across four tenure bins (0, 10, 20, and 30 days). In it, strategy A scored higher for the bins of 10 and 20 days, while B scored higher for 0 and 30 days, which means the optimal scoring combination is to use strategy A to score people that have been in the cohort for 0-10 days, strategy B for those in it for 11-20 days and 21-30 days, and A again for those in it for 30+ days.

[![Graphic of dynamic prediction tenure scoring](/images/blog/dynamic-prediction/dynamic-prediction-1.png)](/images/blog/dynamic-prediction/dynamic-prediction-1.png)

To illustrate this point further, below is an example of a pipeline that uses tenure bins to select the best model for each tenure group. In it, you can see the **pipeline population** group is broken down into three tenure bins, each of which has a “best” predictive model that’s used for scoring.

[![Graphic of dynamic prediction pipeline](/images/blog/dynamic-prediction/dynamic-prediction-2.png)](/images/blog/dynamic-prediction/dynamic-prediction-2.png)
_For further reading on Faraday model calibration and probability scores, check out [this blog post](/blog/faraday-probability-scores)._

In short, dynamic prediction enables us to provide scores that update with your data, ensuring accurate, powerful predictions for all stages of the customer lifecycle.

[Sign up free](https://faraday.ai/signup).

Predictive patterns change over time. Faraday uses dynamic predictions to choose the best model for you based on subject tenure.

Product