## How direct mail lead suppression works at Faraday

In the highly competitive financial services industry, direct mail remains a powerful tool for customer acquisition. However, without precise targeting, companies often waste significant resources sending mail to leads who are unlikely to convert. This inefficiency inflates marketing costs and dilutes campaign effectiveness.

Faraday’s predictive [lead suppression](https://faraday.ai/templates/lead-suppression/recharge) uses custom predictive models built on 1,500+ rich, consented datapoints from the Faraday Identity Graph to remove low-intent leads from direct mail campaigns. The process follows a clear, structured approach:

### Step 1: Data ingestion & enrichment

Faraday begins by securely ingesting the client’s lead data. Using the  [Faraday Identity Graph](https://faraday.ai/docs/abstractions/traits) (FIG)—which includes over 1,500 attributes on 240 million U.S. consumers—our platform enriches this data with key demographic, behavioral, and property insights.

### Step 2: Predictive modeling & trend analysis

Our machine learning algorithms analyze historical conversion patterns to identify datapoints associated with high-intent buyers—specific to the brand. This predictive analysis allows us to assign a probability score to each lead, ranking them from most to least likely to convert.

### Step 3: Lead decile segmentation

Leads are grouped into ten deciles based on their likelihood of conversion. This segmentation provides a clear understanding of which leads offer the highest return on marketing investment and which are unlikely to engage.

### Step 4: Establishing a suppression threshold

We work with clients to establish a **cut-off threshold**, ensuring that leads below a certain probability score are removed from direct mail outreach. This helps reduce wasted marketing spend while increasing the efficiency of each campaign.

### Step 5: Optimized mailing list delivery

Once the suppression model has been applied, we provide the client with a refined, high-value lead list, ensuring that marketing efforts are directed toward the most promising prospects.

## How lead suppression adds value to financial services clients

For financial services companies—especially those in debt consolidation, lending, and wealth management—**precision targeting** is critical. Predictive lead suppression delivers multiple key benefits:

### Reduced marketing waste

By removing low-intent leads from direct mail campaigns, financial institutions significantly cut costs associated with printing and postage, maximizing their marketing budget.

### Higher conversion rates

Focusing outreach on high-intent consumers improves conversion rates, leading to greater customer acquisition efficiency.

### Enhanced customer insights

Predictive modeling not only identifies which leads are least likely to convert but also provides valuable insights into the characteristics of high-intent buyers, allowing for more refined targeting in future campaigns.

### Increased return on investment (ROI)

By reallocating marketing spend to the most promising prospects, companies see a higher ROI on direct mail efforts, turning a traditionally broad-channel marketing approach into a high-precision acquisition strategy.

## Case study: Debt consolidation company saves over $100K per month

### Challenge: High direct mail costs with low returns

A leading [debt consolidation company](https://faraday.ai/stories/debt-consolidation-company) relied heavily on direct mail for customer acquisition but was struggling with inefficiencies. A large percentage of their outreach was going to low-intent consumers, inflating costs and reducing campaign effectiveness.

### Solution: Faraday’s predictive lead suppression

The company partnered with Faraday to implement a [custom lead suppression model](https://faraday.ai/templates/lead-suppression/recharge), which analyzed their historical customer data alongside external consumer attributes to pinpoint the least-likely-to-convert leads.

### Implementation: Multi-segment predictive modeling

While the initial suppression model provided immediate savings, Faraday’s data science team took optimization further by identifying four distinct customer segments within the company's audience. Instead of using a single predictive model, we developed four tailored models, each optimized for its respective segment. A rigorous validation process confirmed that this approach yielded significantly better results.

### Results: Immediate cost savings & higher revenue

By implementing Faraday’s predictive suppression approach, the company achieved:

* **600,000 fewer** pieces of wasted direct mail per month  
* **$100,000+ in monthly savings** on direct mail costs  
* **An additional $20,000 per month** in revenue due to higher targeting precision  
* **10x ROI** on their predictive modeling investment

This company didn’t just reduce waste—it transformed how they acquire customers. By leveraging AI-driven predictive suppression, they turned direct mail into a high-precision growth engine, ensuring that every marketing dollar was spent on reaching the right audience.

### In conclusion

For financial services firms, predictive lead suppression is a game-changer. By eliminating low-value leads and refocusing efforts on high-intent prospects, companies not only reduce costs but also increase customer acquisition efficiency and boost revenue. With Faraday’s AI-driven insights, financial institutions can turn their direct mail campaigns into powerful, data-driven marketing strategies that drive sustained growth.

Want to harness this predictive power for your brand? [Sign up today](https://login.faraday.ai/u/signup?state=hKFo2SBYd0tiZ1hRSmU4UlhCV052VFI4dzBzUWx5bU9pejM1a6Fur3VuaXZlcnNhbC1sb2dpbqN0aWTZIFhhbDZFLW9YSWxJd2NBWmNmYWFld0hXdUR5WGFJakNxo2NpZNkgMDF6SWFodVNaV203SmRVMFNtWmVNUHMzZ2ZHcklqWTI)!

Faraday’s predictive lead suppression leverages AI/ML models to remove low-intent leads from direct mail campaigns, helping financial services companies quickly maximize marketing efficiency and ROI by only targeting high-value prospects.

Direct mail lead suppression for financial services clients

You know how valuable predictive analytics could be to your business, but you haven’t signed up yet. So, what’s holding you back?

For many businesses, and especially those in notably security-conscious industries like healthcare and finance, data security is the biggest concern. You want the insights that AI and predictive analytics can provide, but you need to know your customer data is safe. At Faraday, we understand that concern. That's why we’ve built our entire platform around safeguarding your sensitive data, from collection to analysis.

In this blog, we’ll walk you through the steps we take to ensure your data remains secure, private, and ethically handled, so you can trust us with the insights that power your business growth. But first, we want to talk about our Faraday Identity Graph and why we take our security so seriously.

### The Faraday Identity Graph, and why we take privacy so seriously

The [Faraday Identity Graph (FIG)](/docs/abstractions/traits) is at the core of everything we do. It’s a powerful tool that [allows us to generate accurate, actionable insights from U.S. consumer data](/blog/how-faraday-enriches-first-party-data-with-fig), helping brands understand their customers at a deeper level (we also sell this data with [data appends](/blog/data-appends-blog)). But with great power comes great responsibility. We recognize that the data we work with is highly sensitive, and we take every precaution to handle it with the utmost care.

FIG isn’t only concerned with unlocking insights; it’s also designed with privacy as a top priority. [From the very beginning, we’ve built it with robust security features to ensure that consumer data is kept safe and secure.](https://faraday.ai/security) Whether it’s protecting personal information or ensuring compliance with privacy laws, our mission is to uphold the trust you place in us by following the strictest data handling practices.

So let’s start at the top and take a look at some of the foundational security practices we follow to safeguard your data.

### SOC 2 compliance

[SOC 2 (System and Organization Controls 2\)](https://secureframe.com/hub/soc-2/what-is-soc-2) is a security framework designed to ensure that organizations protect customer data from unauthorized access and security incidents, and is often viewed as the gold standard for this purpose. It evaluates five key areas: security, availability, processing integrity, confidentiality, and privacy.

At Faraday our compliance with SOC 2 standards means we undergo rigorous audits of our security practices every year, with an independent third party verifying that our data protection measures meet the highest industry standards.

Our security policy is built around important principles like the "principle of least privilege" (where users only have access to the data they absolutely need), prevention of supply chain attacks, and strict access controls. We also participate in  [SOC 2 Type II](https://secureframe.com/hub/soc-2/type-1-vs-type-2) audits, which not only check that security controls are in place but also assess their effectiveness over time. This ongoing process ensures that we continuously meet top-tier security protocols and maintain a secure environment for your data.

### Compliance with privacy regulations

To protect your consumer data, we comply with key privacy regulations, including [HIPAA](https://www.cdc.gov/phlp/php/resources/health-insurance-portability-and-accountability-act-of-1996-hipaa.html), [CCPA](https://oag.ca.gov/privacy/ccpa), and [GDPR](https://gdpr-info.eu/), to ensure your data is secure and handled responsibly.

For CCPA and other state privacy laws, we offer an easy opt-out system, allowing individuals to remove themselves from our datasets. What’s more, we extend this opt-out through a reciprocal process to our vendors, so when either Faraday or a vendor receives an opt-out request, the data is removed from both systems—ensuring full compliance across our ecosystem.

As for GDPR, since our Faraday Identity Graph focuses on U.S. data, we proactively filter out non-U.S. data at the "perimeter," ensuring we only process data we are authorized to use.

When handling Client data, Faraday operates as a Data Processor/Service Provider, meaning we will only use client data for the specific contracted purpose and will support our client’s own compliance with state data privacy laws. 

### Data vendor standards: permissioned and ethically distributed 

Ensuring responsible data handling goes beyond secure storage and processing—it also means sourcing data from reputable vendors. At Faraday, we take extra care in selecting the data providers we work with.

We only collaborate with vendors who explicitly affirm in their terms of service that their data is properly permissioned, legally obtained, and authorized for use and distribution. We are committed to maintaining the highest standards of ethical data usage, ensuring that every piece of data we use aligns with the proper permissions and legal requirements so you can confidently leverage accurate, responsible insights.

### Addressing bias in consumer data

As a data provider, Faraday acknowledges the reality that consumer data is inherently shaped by [historical and systemic biases](https://developers.google.com/machine-learning/crash-course/fairness/types-of-bias). We acknowledge these biases but take active steps to mitigate their impact on our models and predictions.

For example, mortgage lending and credit scoring have historically exhibited bias, with certain demographic groups, particularly minorities and women, facing higher rejection rates or higher interest rates due to systemic biases embedded in financial models. These models often relied on historical data that reflected past discriminatory lending practices rather than objective risk. 

To avoid perpetuating these systemic biases, Faraday takes a range of actions. For instance, we normalize data to reduce geographic disparities and offer bias detection and mitigation strategies to address disparities along sensitive dimensions such as age and gender. 

Responsible AI and ethical data practices are at the forefront of our methodology, ensuring that our models and insights promote fairness and inclusivity.

### Secure data handling practices

Faraday accepts data exclusively through secure channels—no emailed files or unprotected transfers. We also support hashed data ingestion and egress, allowing brands to share encrypted email addresses and phone numbers for added security. While hashing is not a foolproof privacy solution, it remains a valuable safeguard, and we ensure it is an option for those who prioritize additional data protection.

### Customer data is never shared between clients

One of the most common questions we receive is whether we share customer data across different client accounts. The answer is unequivocally no. Our terms and data processing agreement (DPA) explicitly state that client data is used solely for delivering services to that client, and it is never shared, pooled, or combined with other clients' data in any way. We believe that maintaining clear, stringent data boundaries is essential for protecting client trust and upholding our ethical standards.

### Conclusion: a commitment to ethical data handling and security

At Faraday, we don’t just meet industry standards—we set them. From rigorous security audits to responsible data sourcing and compliance with privacy regulations, we prioritize the safety, privacy, and ethical handling of your data at every step. Our commitment to transparency and secure practices ensures that your customer data is protected, and our proactive efforts to address biases help foster fairness in every insight we generate.

By choosing Faraday, you’re not just getting advanced predictive analytics—you’re gaining a trusted partner dedicated to upholding the highest standards of data security and privacy. We believe that responsible data use is the foundation of meaningful insights and lasting business success.

If you have any questions about our data security policies or compliance measures, don’t hesitate to reach out. We're here to help you navigate the complexities of data protection, so you can focus on what truly matters—growing your business with confidence.

Safe and responsible handling of data is a primary priority at Faraday, but what does that really mean? In this blog, we'll unpack some of the ways we ensure your data remains secure, private, and ethically managed throughout our platform.

How Faraday ensures secure and responsible customer data handling

*Note to reader: The following is the perspective of a lowly marketer who has followed these debates closely, but it’s worth noting that these discussions have passed the desks of far more knowledgeable people. So if you want to know what [Spinoza](https://homepages.uc.edu/~martinj/Spinoza_&_Hobbes/English/Spinoza%20-%20Complete%20Works%20$2002$.pdf) and [Descartes](https://yale.learningu.org/download/041e9642-df02-4eed-a895-70e472df2ca4/H2665_Descartes%27%20Meditations.pdf) really said, read their work\!*

Imagine a world where a computer can write a poem that ‘rivals’ Shakespeare, but struggles to understand the simple concept of a joke. This is the paradoxical reality we often find ourselves in when working with large language models (LLMs) like [OpenAI’s](https://openai.com/) [GPT-4](https://openai.com/index/gpt-4/). 

We’re all hearing a lot about LLMs—AI systems that generate text, answer questions, and even compose blogs (although we promise this one comes from a certified [real boy](https://giphy.com/gifs/uyG3UwkbgTaOevKe9C)). These models are rapidly reshaping workflows across industries, automating tasks that once required human effort—from customer support to legal document review to coding assistance. But as they become more embedded in our work and lives, it’s worth asking: Are they truly intelligent? And, more importantly, what do we even mean by ‘intelligence’ when it comes to machines?

To unpack this question, let’s compare two different popular approaches to AI: **conversational** and **predictive**. While both are powerful, they operate in very different ways. Let’s take a closer look at how Conversational AI like GPT-4 works and how it compares to Predictive AI like [Faraday](http://Faraday.ai).

### Conversational AI vs. Predictive AI

[Conversational AI](https://cloud.google.com/conversational-ai?hl=en), like GPT-4, excels at generating human-like text. Trained on vast amounts of data, it learns patterns in language by analyzing how words and phrases relate to each other across different contexts. At its core, it predicts the most statistically likely next word in a sequence—similar to the predictive text on your phone’s keyboard, but on a much grander scale. This allows it to mimic conversation convincingly, but it doesn’t necessarily understand what it’s saying. It can generate fluent responses, but it struggles with complex logic, [math](https://garymarcus.substack.com/p/math-is-hard-if-you-are-an-llm-and), or real-world reasoning, which requires understanding beyond mere pattern recognition.

Think of it like a high-speed version of Mad Libs. While it can generate coherent and even insightful responses, it doesn’t understand the meaning behind them—it’s simply identifying statistical patterns and filling in the blanks. On that basis, one could argue that its intelligence is primarily performative—more like an advanced pattern recognition game than genuine comprehension or thought. 

Of course, this raises a deeper question: What actually counts as thought? Philosophers, cognitive scientists, and AI researchers have debated this for decades, and we’re not here to settle it today. But for the sake of this discussion, let’s focus on the practical distinction between AI that emulates behavior and AI that augments decision-making.

[Predictive AI](https://faraday.ai/templates) (like [Faraday](http://faraday.ai)), on the other hand, isn’t built to simulate conversation. Instead, it’s designed to analyze complex patterns in structured data and forecast outcomes, helping humans to make smarter, faster decisions. While Conversational AI generates responses based on linguistic patterns, Predictive AI identifies meaningful relationships between behaviors, demographics, and other variables—surfacing insights that would be difficult or impossible to detect manually. In that sense, Conversational AI is designed to  *sound human*, whereas Predictive AI is designed to *enhance human decision-making* by providing precise, data-backed foresight.

Let’s unpack how predictive AI works with a parallel from science fiction.

### A mathematical model for predicting the future

In Isaac [Asimov’s 1940s Foundation series](https://en.wikipedia.org/wiki/Foundation), mathematician Hari Seldon develops a science called psychohistory—[a statistical method capable of predicting the future.](https://media.tenor.com/TrAsjYbL720AAAAM/chris-pratt-wow.gif) Psychohistory doesn’t predict individual actions but instead foresees large-scale societal trends with remarkable accuracy, enabling leaders to anticipate and alter the course of history.

If you’re already using Faraday, you probably see a connection here. Like psychohistory, predictive AI helps businesses affect the future by anticipating customer behaviors. By analyzing vast amounts of data, predictive AI can forecast outcomes such as which leads are most likely to convert or what offer will resonate with a particular customer. This ability empowers businesses to make precise, data-driven decisions that can directly influence future outcomes.

But unlike psychohistory, which only understood the broad movements of society, predictive AI is powerful enough to make accurate, actionable predictions at the individual level. And in this sense **our product actually beats a system designed to save the galaxy**  (sorry Hari)\!

### Turing, Descartes, and machines that learn

But we came here to talk about intelligence. 

And when discussing intelligent machines, our minds often turn to the famous, controversial, and (critically) often misunderstood [Turing Test](https://www.techtarget.com/searchenterpriseai/definition/Turing-test), a benchmark proposed by the great [Alan Turing](https://en.wikipedia.org/wiki/Alan_Turing) to measure whether a machine can exhibit behavior indistinguishable from human intelligence. 

“So it *basically* tested if the machines were sentient right?”

Well, not exactly. Turing himself viewed the test as more of an “[imitation game](https://plato.stanford.edu/entries/turing-test/)” than a definitive way to assess whether a machine truly “thinks”. His focus was less on the idea of thought itself, and more on the appearance of it—the ability of a machine to convincingly mimic human behavior and language.

The goal of the Turing Test wasn’t to prove that a machine could reason or understand in the way humans do, but rather to test if it could generate responses so human-like that an observer might be fooled into believing it was also actually a person. Passing the test would mean a machine could emulate the behaviors and language associated with human reasoning, but not necessarily that it possesses any true understanding behind them.

This idea interacts interestingly with the philosophy of [René Descartes](https://plato.stanford.edu/entries/descartes/), who famously said “[I think, therefore I am](https://www.britannica.com/topic/cogito-ergo-sum)”. Descartes believed  true human intelligence was defined by the ability to reason and, importantly, to use language to express reason. For him, language wasn’t just a tool for communication—it was the ultimate marker of conscious thought. He argued that only beings capable of reason, expressed through language, could be considered truly conscious. In this context, Descartes viewed machines and (sadly) animals, as [Automata](https://academic.oup.com/book/32221/chapter-abstract/268391060?redirectedFrom=fulltext)—[sophisticated mechanical systems that operated according to predetermined laws](https://www.amazon.com/Top-Race-Robot-Dog-Programmable/dp/B08G6PT5DC?source=ps-sl-shoppingads-lpcontext&ref_=fplfs&psc=1&smid=A17FAJD38FG3OS&gQT=1), but are incapable of thought or reason. 

Even though Automata might perform tasks that appeared purposeful, Descartes believed their actions were purely mechanical, driven by physical processes rather than any understanding or intention. For Descartes, the difference between humans and these objects was clear: humans could reason and engage with the world in a way that reflected true thought.

*Fast forward* to the world of modern AI, and we now have machines that can produce remarkably human-like language, making it seem as though they are reasoning or thinking. But are they really doing so?

While LLMs can string together words and phrases in ways that sound intelligent, they don’t truly “understand” the meaning in the way humans do. In this sense, LLMs seem to align with Descartes’ view of machines as sophisticated [mimics](https://media.tenor.com/c4HtDmmYreUAAAAM/mimic-animation.gif)—producing outputs that seem intelligent but, in reality, are merely a performance of an extremely complicated script.

But this also presents a challenge to Descartes’ assertion that language is the definitive marker of reason, as they demonstrate the ability to produce seemingly intelligent language without underlying comprehension.

And to be fair, it is also of course worth noting that Descartes was writing in the [17th century](https://www.marxists.org/reference/archive/descartes/1639/meditations.htm). A time when machines were more likely to resemble windmills than the complex computers we know today. So his idea of “machines” was far more limited, even if his philosophical view of them as unthinking, mechanical objects persists in many ways today.

But Turing didn’t just talk about AI that could communicate, he also discussed those that could learn from experience. In a 1947 lecture, he stated, [“What we want is a machine that can learn from experience,” and that the “possibility of letting the machine alter its own instructions provides the mechanism for this.”](https://www.britannica.com/science/history-of-artificial-intelligence) Instead of just following pre-written rules, such a system would improve over time through experience and trial-and-error. This is where the concept of [heuristics](https://www.merriam-webster.com/dictionary/heuristic) in problem-solving comes into play. Heuristics are rules of thumb that allow people (or models) to quickly find solutions based on past experiences and probabilities of success. Rather than seeking the perfect solution, which could be time-consuming and resource-intensive, heuristics prioritize finding a good-enough solution more efficiently.

Predictive AI operates on this principle. When these models analyze new data, they adjust their internal calculations, improving their accuracy based on what has worked in the past. Unlike LLMs, which don’t fundamentally change after their initial training unless retrained on new datasets, predictive AI systems evolve dynamically. They continuously optimize their decision-making to deliver better, faster outcomes as they learn from new data and experiences.

### The future of AI: beyond emulation

In this author's opinion the future of AI should aim to augment, not just replicate, human behavior. 

Predictive AI, which excels in analyzing data and uncovering hidden patterns, serves as a prime example of this. By turning complex data into actionable insights, these models empower businesses to make more informed decisions. For instance, in marketing, predictive models can identify high-conversion customers and suggest tailored strategies, tasks that would be impossible for humans to undertake or scale efficiently on their own. Its continuous learning and evolution further refine its predictions, unlocking new levels of efficiency and impact over time.

But while the focus of this article thus far has been primarily on how LLMs focus on mimicking human behavior, that’s not the only thing they can do. For instance, LLMs help us interact with and quickly manage complicated systems, as is the case with AI agents. 

Consider a scenario where a patient needs to schedule a complex series of medical appointments. An AI agent, integrated with the hospital's scheduling system, can understand the patient's needs, check doctor availability, coordinate with insurance requirements, and send reminders, all through a simple conversational interface. This not only streamlines the process for the patient but also frees up hospital staff to focus on critical patient care. 

These tools can automate processes and save time by performing repetitive tasks or helping us interact more efficiently. They don't replace us or try to copy us; they help us do more and do it better. And they free up human effort for higher-value work, enabling creativity and decision-making. 

In this sense, both predictive AI and LLMs, despite their differing focuses, can amplify human abilities when used in the right contexts, showing that the key to successful AI implementation is using the right tool for the job.

### And really, what even is intelligence?

At this point in the article you might be asking, “Wasn’t this whole thing supposed to be about whether these AIs were intelligent?” 

The concept of intelligence, though, isn’t as simple as a binary choice between “smart” and “dumb”. As another enlightenment philosopher, [Baruch Spinoza](https://www.britannica.com/biography/Benedict-de-Spinoza) proposed with his philosophy of [monism](https://ndpr.nd.edu/reviews/spinozas-argument-for-substance-monism-why-there-is-only-one-thing/), consciousness and reason are not exclusive to humans. They are defined through an object's interactions with its surroundings, which exist on a spectrum throughout nature, from the simplest forms of interaction (those of a rock) to the complex reasoning of humans. 

And indeed, both predictive AI and LLMs, though different in their design and function, are part of this spectrum. Each possesses a distinct form of intelligence that, when used thoughtfully, can augment human decision-making, creativity, and problem-solving.

Ultimately, both tools—predictive AI and LLMs—are intelligent in their own right, but each excels in different areas. By approaching AI with the mindset of enhancing human abilities rather than simply emulating human behavior, we unlock its true potential. Predictive AI excels at providing insights and optimizing decision-making, while LLMs offer a powerful tool for generating human-like text, facilitating communication, and automating tasks (among a range of other applications). By understanding their strengths and applying them wisely, we can harness AI to not only complement human abilities but to push the boundaries of what is possible, opening up new opportunities for innovation, efficiency, and growth.

### Conclusion

Both LLMs and predictive models are intelligent, but in different ways. It’s about using them thoughtfully—recognizing the unique strengths each brings to the table.

In the end, the true value of AI isn’t in replicating human thought, but in extending our own potential. Predictive AI can enhance decision-making and strategic thinking, while LLMs offer powerful ways to facilitate communication and automate tasks. When used in harmony, these technologies can open up new opportunities for innovation, efficiency, and growth, creating a future where humans and AI work together to solve complex problems and drive meaningful change.

At the end of the day, it’s not much different from working with any other tool; drills are great for drilling, but don’t ask one to write an essay (that draft was terrible).

What makes a machine intelligent, and how can we harness its capabilities to improve our lives? This essay explores various approaches to that question, from Descartes to Turing.

AI to empower human intelligence, not just emulate it

What if you could embed AI-driven predictions into your platform—without the massive investment in data, engineering, and ongoing model maintenance?

Your customers want smarter insights: predictions on which leads will convert, which territories will perform best, and which prospects match their ideal profile. But building AI models from scratch is a massive undertaking—one that requires specialized expertise, vast amounts of data, and months (or even years) of development. For many SaaS companies, the cost and complexity make it an unrealistic endeavor.

So what is to be done?

[![SalesRabbit partnership with Faraday landing page](/images/blog/pbf-blog/salesrabbit-story-header-image.png)](/images/blog/pbf-blog/salesrabbit-story-header-image.png)


## Introducing Powered by Faraday

This is precisely why we launched **Powered by Faraday**—a turnkey solution that allows SaaS companies to integrate AI-powered customer predictions into their platforms without having to build or maintain machine learning models themselves.

We’ve spent over a decade refining predictive models that help businesses understand and anticipate customer behavior and with **Powered by Faraday**, we bring these capabilities directly into the hands of product teams, enabling them to enhance their software with advanced AI-driven insights quickly and cost-effectively.

## What Powered by Faraday does

With **Powered by Faraday**, companies can:

* **Enrich their customer data** with 1,500+ built-in consumer attributes, adding valuable context to their existing datasets.  
* **Predict customer behavior** by using pre-built machine learning models that score leads, prioritize outreach, and identify high-conversion opportunities.  
* **Accelerate product development** by embedding plug-and-play AI models, cutting multi-year build times down to weeks or even days.  
* **Optimize costs and scalability** with transparent pricing and predictable expenses, avoiding the uncertainty of large-scale computational costs.

For SaaS companies looking to offer predictive capabilities without reinventing the wheel, **Powered by Faraday** is a game-changer.

## Faraday’s competitive edge

Many companies offer AI solutions, but **Powered by Faraday** stands out due to several key advantages:

* [**Seamless integration with a robust API**:](https://faraday.ai/docs/abstractions/traits) Powered by Faraday is designed for seamless implementation. Faraday’s well-documented, developer-friendly API makes embedding AI-driven insights into a product straightforward and efficient.   
* [**FIG (Faraday Identity Graph)**](https://faraday.ai/docs/abstractions/traits): A vast, privacy-compliant third-party dataset with 1,500+ attributes, enabling richer predictions than first-party data alone.  
* **Forward-deployed data science & professional services**: Hands-on support from data scientists who fine-tune models and ensure seamless deployment.  
* **Ethical, privacy-first AI**: A commitment to consumer privacy, avoiding personally identifiable information while delivering accurate predictions.  
* **Industry-specific models**: Pre-trained, vertical-specific machine learning models that accelerate time-to-value.  
* **Flexibility to design tailored solutions:** Faraday enables companies to integrate predictive capabilities without the constraints of rigid, one-size-fits-all AI solutions—offering the perfect balance between ease of adoption and customization.

By combining rich third-party data with hands-on AI expertise, Faraday helps SaaS companies deliver world-class predictive features faster and with greater accuracy than in-house or off-the-shelf solutions.

## Real-world use cases

Many companies have already leveraged **Powered by Faraday** to enhance their platforms. 

Here’s how two of them—[AccuLynx](https://faraday.ai/blog/acculynx-lead-intelligence-powered-by-faraday-api) and [SalesRabbit](https://faraday.ai/blog/salesrabbit)—are making AI-driven predictions a core part of their products:

### AccuLynx: Smarter lead prioritization for roofing contractors

[AccuLynx](https://acculynx.com/), a leading provider of business management software for roofing contractors, needed a way to help their customers prioritize sales efforts more effectively. They integrated [Faraday’s prediction API](https://faraday.ai/docs/reference) into their new [Lead Intelligence feature](https://acculynx.com/roofing-sales-lead-intelligence/), which automatically scores leads within AccuLynx’s platform, based on hundreds of demographic, financial, and property data points.

The result? Roofing contractors using AccuLynx no longer have to guess which leads to focus on. Instead, they get **AI-powered rankings** that help them close more deals efficiently—without needing in-house data science expertise.

As their VP of Product [Mark Rattin](https://www.linkedin.com/in/markrattin/) said, “Lead Intelligence is a game-changer for roofing companies,” … “AccuLynx customers no longer have to guess which leads to call first, or only rely on their experienced sales people to quickly close deals. Now they can improve their sales win rates by focusing their efforts on the best leads.”

Sounds like a win to us\!

[![AccuLynx landing page](/images/blog/pbf-blog/acculynx-landing-page.png)](/images/blog/pbf-blog/acculynx-landing-page.png)

### SalesRabbit: AI-Driven insights for field sales teams

[SalesRabbit](https://salesrabbit.com/), a leader in field sales software, wanted to take their [DataGrid](https://salesrabbit.com/datagrid/) solution to the next level. While their platform already provided canvassers with property and homeowner data, they saw an opportunity to go further—predicting which homeowners were most likely to convert based on behavioral and demographic insights.

By integrating [Faraday’s customer prediction models](https://faraday.ai/templates), SalesRabbit was able to:

* Guide sales reps toward high-potential neighborhoods and properties.  
* Optimize canvassing routes for better efficiency.  
* Improve app performance by caching predictive insights in advance.

By leveraging **Powered by Faraday**, SalesRabbit transformed DataGrid into DataGrid AI, offering their customers a predictive lead scoring tool in a fraction of the time it would’ve taken to build in-house.

[![SalesRabbit DataGrid AI landing page](/images/blog/pbf-blog/salesrabbit-datagridai.png)](/images/blog/pbf-blog/salesrabbit-datagridai.png)

## **Conclusion: AI-Powered predictions without the headache**

For SaaS companies looking to bring predictive insights to their customers, building AI models from scratch is often too expensive, time-consuming, and resource-intensive. Powered by Faraday removes those barriers, making it easy to integrate best-in-class consumer predictions into your product without the heavy lift.

Whether you want to help your customers prioritize leads, optimize sales territories, or enhance customer insights, Powered by Faraday lets you do it faster, more affordably, and at scale.

Want to see how it works? Check out the full stories from [AccuLynx](https://faraday.ai/blog/acculynx-lead-intelligence-powered-by-faraday-api) and [SalesRabbit](https://faraday.ai/blog/salesrabbit), or reach out to [Dave](https://www.linkedin.com/in/davidtoddsmall/) or the [Faraday](https://faraday.ai/talk-to-sales) team to explore how Powered by Faraday can enhance your platform.

AI is revolutionizing how SaaS companies can leverage predictive insights, and at Faraday, we’re making it easier than ever to integrate powerful, ethical AI-driven capabilities without the hefty overhead.

AI-Powered insights for SaaS companies, the fast way

In today's data-driven world, businesses can amass a wealth of first-party information, but this presents only a partial picture of their current and potential customers. Gaps in their data can limit the effectiveness of their marketing strategies, market research, and predictive models. That’s where Faraday’s data appends come in. By enhancing their existing datasets with high-quality third-party datapoints, brands can unlock deeper insights to make smarter, more informed decisions.

## What are data appends, and why do they matter?

Data appending is a process by which businesses can purchase and integrate external data into their own datasets to supplement their first-party information. These attributes might include demographic information, behavioral patterns, or even lifestyle indicators, datapoints that are generally not available through first-party channels alone. By enriching your data with these additional insights, you can gain a fuller picture of your customers and their behaviors. 

But Faraday’s data appends go beyond simple enrichment. Our data is rigorously vetted, privacy-compliant, and built for seamless integration into your existing workflows. Whether refining market research, improving segmentation, or even powering your own predictive models, these appends help businesses make smarter, data-driven decisions. With Faraday, you’re not just adding data—you’re adding intelligence.

## How businesses use data appends

Data appends address various business needs. Here are some common ways companies leverage them:

### Exploratory research

Understanding your market often requires more than just your existing data. Take, for example, a roofing company. They might want to analyze roof sizes across different regions to gauge demand. Instead of sourcing this data themselves (which would be slow and expensive) or purchasing a generic dataset that requires extensive processing, they can use Faraday to buy and append only the relevant geographic and property data, seamlessly. This allows them to conduct targeted research efficiently and make informed decisions—like leveraging predictive targeting with Faraday next.  

### Supplementing existing datasets

Businesses often lack key customer details. Take a retailer that has always relied on email marketing. They would have extensive email lists but likely never collected addresses or phone numbers. If they want to expand into direct mail or call center outreach, they’ll need that missing data. This is where Faraday’s data appends come in, providing the necessary information without the hassle, time commitment, or high costs of sourcing it themselves. 

### Building custom predictive models

 An advanced use case for our data appends is supporting businesses that prefer to develop their own machine learning models. Just as we use this data at Faraday, these companies leverage our high-quality, cleaned, and structured datapoints to enhance their predictive outcomes. In this case, Faraday acts as a trusted data provider, delivering the precise insights needed to model a variety of customer attributes—such as likelihood to churn, lifetime value, or purchasing behavior—ready to be seamlessly integrated into customers’ [Python](https://www.python.org/) environments.

## How Faraday’s data appends work

The Faraday Identity Graph (FIG) is the backbone of our data appending services. This robust dataset combines multiple, meticulously cleansed, and feature-engineered data sources into one comprehensive dataset. [With over 1,500 datapoints covering nearly all U.S. adult consumers and their households](https://app.faraday.ai/traits), it offers a deep well of insights into things like demographic trends, property ownership, life events, and financial indicators, just to name a few.

The process to access this data as appends is simple and streamlined:

1. **Select datapoints**: Choose the datapoints that align with your business goals (e.g., income levels, household size, location-specific behaviors).  
2. **Match & append**: Faraday will match these datapoints to your existing customer records, providing a more complete profile of your audience.  
3. **Integration**: The enriched data is delivered in a clean, structured format that’s easy to integrate with your models and business processes.

[![Data-appends diagram](/images/blog/data-appends-diagram.png)](/images/blog/data-appends-diagram.png)

## Why not just find the data yourself?

Some companies attempt to enhance their marketing efforts by purchasing third-party data directly from various vendors themselves, rather than using packaged and processed services like Faraday’s data appends. However, this approach often presents significant challenges:

**Contracting complexity**: Sourcing offline data can involve lengthy negotiations and large upfront commitments.

**Security risks**: Handling sensitive data requires strict compliance frameworks, such as SOC-2 certification. Unlike many vendors, all Faraday data appends are privacy-compliant, ethically sourced, and adhere to high standards of data protection, ensuring your customers' personal information remains secure while providing actionable insights.

**Regulatory challenges**: Companies storing offline data must navigate consumer privacy laws, including opt-out and deletion requests. Faraday manages these complexities for you, so you can focus on what matters—leveraging new insights to drive revenue.

**Data science overhead**: Raw offline data often requires significant processing, feature selection, and bias mitigation before it can be effectively used. This can result in costly time and resource investments as companies try to integrate third-party data into their models.

Faraday eliminates these challenges by providing ready-to-integrate data that is pre-processed and structured for immediate use. Our data appends come with built-in privacy compliance, so you don’t have to worry about regulatory hurdles or the risk of handling sensitive information. Additionally, the data is feature-engineered and bias-mitigated, saving you the time and resources typically needed to prepare raw data. With Faraday, you can access high-quality, actionable insights without the complexities and overhead associated with managing third-party data directly. This allows your team to focus on refining your marketing strategies and driving better business outcomes.

## How to get started

Getting started with Faraday’s data appends is simple. Once you’ve identified the datapoints you need, we’ll work with you to integrate them into your existing systems. Whether you’re looking for a one-time data boost or ongoing data enrichment, our flexible pricing and easy integration make it simple to scale.

To learn more or request a demo, **contact us today**.

Faraday’s data appends enrich your existing customer data with third-party datapoints, helping you fill gaps, refine targeting, and make smarter decisions.

Leveraging the power of the Faraday Identity Graph through Data Appends

One of the most critical aspects of model development is ensuring that our models can generalize well to unseen data (i.e. when the predictions are deployed). While it's relatively straightforward to train a model that performs well on training data, the real challenge is creating models that maintain their performance when faced with new, previously unseen examples. This is where cross-validation (CV) comes into play to validate our models' performance and make more informed decisions during the model development process. Throughout this post, we will review different CV variants and how Faraday uses them to train its model and forecast return on investment (ROI) for our clients.

## What is cross-validation

At its core, cross-validation is a resampling procedure used to evaluate machine learning models on limited data samples. Instead of using a single train-test split, cross-validation provides a framework for multiple evaluations of our model using different portions of the available data.

Cross-validation involves partitioning the available data into complementary subsets, training the model on some subsets (training set), and validating it on the remaining subset (validation set). This process is repeated multiple times, with each data point getting the opportunity to be part of both training and validation sets.

The key principles of cross-validation include:

1. Data Independence: Ensuring that training and validation sets are strictly separated in each iteration
2. Comprehensive Evaluation: Using all available data for both training and validation through multiple iterations
3. Statistical Robustness: Generating multiple performance metrics to assess model stability

### K-Fold cross-validation

K-fold cross-validation is the most common and widely used variant of cross-validation. In this approach, the data is divided into k equal-sized folds, and the following process is repeated k times:

1. One fold is selected as the validation set
2. The remaining k-1 folds are used as the training set
3. One model is trained on the training set
4. Predictions on the validation set are saved

Faraday uses a 5-fold CV to evaluate its models. This means that for any given propensity model, Faraday builds 6 variations of the model (5 for the k-fold validation and one final version used as the deployed model.)  

[![K-fold cross-validation diagram](/images/blog/cross-validation-blog/regular_cv.png)](/images/blog/cross-validation-blog/regular_cv.png)


Once all five models are trained, each sample in the data has been scored once. This means that the cross-validation allows us to assign a prediction to each person, allowing us to compute the performance metrics using the reconstructed scores.

Special variants of k-fold cross-validation include:

- **Stratified K-Fold**: Maintains the same class distribution in each fold (useful for imbalanced datasets)
- **Leave-one-out cross-validation**: Sets k equal to the number of data points (useful for very small datasets)
- **Time series cross-validation**: The folds are defined using a rolling window for time-based data.

[![Time series cross-validation](/images/blog/cross-validation-blog/time_cv.png)](/images/blog/cross-validation-blog/time_cv.png)

## The benefits of cross-validation

From the previous section, it is clear that CV implies more computation than training a single model. However, cross-validation offers several crucial advantages that outweigh the added costs and complexity, making it a mandatory step in the training process. Let's explore the key benefits that make this technique so valuable.

### Preventing overfitting

One of the primary benefits of cross-validation is its ability to help prevent overfitting. When a model learns the training data too well, it starts to capture noise and specific patterns that don't generalize to new data. Overfitting leads to poor performance in real-world applications. In the charts below, we compare the effects of three different model architectures on the same dataset:

1. The first model is too simplistic to capture the trend in the data.
2. The third one tries too hard to capture every data point and adds a lot of noise between them.
3. The second model does a good job of capturing the overall trend in the data without necessarily trying to capture small variations in the location of the data points.

[![Underfitting vs overfitting curves](/images/blog/cross-validation-blog/blog_3_curves.png)](/images/blog/cross-validation-blog/blog_3_curves.png)

Cross-validation helps detect overfitting by evaluating the model's performance on multiple subsets of the data. If a model performs significantly better on the training data than on the validation sets, it's a clear indicator of overfitting. This early detection allows data scientists to make necessary adjustments to the model's complexity, feature selection, or hyperparameters before deploying it to production.

If we compute our model performance (ROC, accuracy) on the training data, the third model is the best option. However, when using CV, the second model performs better.

### Computing reliable performance metrics

Traditional train-test splits can sometimes provide misleading performance metrics due to the random nature of the split. Cross-validation addresses this limitation by:

- Providing multiple performance estimates using different data splits
- Enabling the calculation of confidence intervals for performance metrics
- Reducing the variance in performance estimation
- Offering a more robust assessment of model stability

These multiple evaluations give us a clearer picture of how our model is likely to perform in real-world scenarios, helping us make more informed decisions about model selection and deployment.

The content of our outcome report and UI relies entirely on the cross-validation but two charts explicitly depict results per fold. Our “folded ROC” plot shows the ROC curve for each fold. This is useful to ensure the model is stable (i.e. the results across folds are nearly identical).

[![Folded ROC curve](/images/blog/cross-validation-blog/Folded_ROC.png)](/images/blog/cross-validation-blog/Folded_ROC.png)

The lift chart illustrates the volatility of lift, particularly for those with higher scores. This information is crucial for lead generation models, which aim to identify potential customers. Faraday assigns a score to each individual within a specified population, and the client uses only the top X% of scorers.

[![Lift curve](/images/blog/cross-validation-blog/lift_curve.png)](/images/blog/cross-validation-blog/lift_curve.png)

## Technical dive

### Formal definitions

**Dataset definition**  

 - Let $\mathcal{D} = \{(x_i, y_i)\}_{i=1}^n$ be our dataset where:  
  - $x_i \in \mathcal{X} \subseteq \mathbb{R}^d$ are feature vectors for sample $i$  
  - $y_i \in \mathcal{Y}$ is the label for sample $i$  
  - $n$ is the number of samples  
 - Define $\mathcal{P}_{XY}$ as the true joint distribution of $(X,Y)$  


**Model space**   
 - Let $\mathcal{F} = \{f_\theta: \mathcal{X} \rightarrow \mathcal{Y} | \theta \in \Theta\}$ be the hypothesis space  
  - $\theta$ represents the model parameters  
  - $\Theta$ is the parameter space  

**Loss function**  
 - Define $\mathcal{L}: \mathcal{Y} \times \mathcal{Y} \rightarrow \mathbb{R}_{\geq 0}$ as the loss function  
 - The expected risk is:  
```math
R(f) = \mathbb{E}_{(X,Y)\sim\mathcal{P}_{XY}}[\mathcal{L}(f(X),Y)]
```
$R(f)$ is interpreted as the error of the model $f$ on the dataset.  
 - The empirical risk on dataset $\mathcal{D}$ is the average loss over $\mathcal{D}$:  
```math
\hat{R}_{\mathcal{D}}(f) = \frac{1}{n}\sum_{i=1}^n \mathcal{L}(f(x_i),y_i)
```
**K-Fold partitioning**  
 - Partition indices $\{1,\ldots,n\}$ into $k$ disjoint subsets: $\{\mathcal{I}_1,\ldots,\mathcal{I}_k\}$  
 - For each fold $j$:  
   - $\mathcal{D}_{val}^{(j)} = \{(x_i,y_i) | i \in \mathcal{I}_j\}$ (validation set)  
   - $\mathcal{D}_{train}^{(j)} = \{(x_i,y_i) | i \notin \mathcal{I}_j\}$ (training set)  


**Cross-validation estimator**  
The k-fold CV estimate of the expected risk is:  
```math
CV_{(k)}(f) = \frac{1}{k}\sum_{j=1}^k \hat{R}_{\mathcal{D}_{val}^{(j)}}(f_{\theta_j})
```
   where $f_{\theta_j}$ is trained on $\mathcal{D}_{train}^{(j)}$  


### Statistical properties

**Bias of CV estimator**  
The bias of the k-fold CV estimator can be expressed as:  
```math
Bias(CV_{(k)}) = \mathbb{E}[CV_{(k)}] - R(f^*)
```
   where $f^*$ is the true optimal function and $\mathbb{E}[CV_{(k)}]$ is the expected risk over all the folds.  

**Variance decomposition**  
The variance can be decomposed into:  
```math
Var(CV_{(k)}) = \frac{1}{k^2}\sum_{j=1}^k Var(\hat{R}_{\mathcal{D}_{val}^{(j)}}) + \frac{2}{k^2}\sum_{i<j}Cov(\hat{R}_{\mathcal{D}_{val}^{(i)}}, \hat{R}_{\mathcal{D}_{val}^{(j)}})
```
Where $Var(\hat{R}_{\mathcal{D}_{val}^{(j)}})$ is the variance of the risk estimate over a fold. It does not refer to the variance of the risk itself (which is just a scalar value). Instead, it refers to the variance of the estimated risk based on multiple possible samples of the validation set within that fold.  

Therefore, as k increases, the overall variance of the cross-validation risk decreases leading to a less volatile estimation of the model performance post-deployment. 

# Holdout test

### What is it?

A "holdout test" can refer to any process where a model is being evaluated on unseen data (e.g. cross-validation above). At Faraday, a holdout test specifically refers to the process of choosing a date in the past $t_0$, fitting a propensity model with data prior to $t_0$, and evaluating the model on data after $t_0$.

On one hand, a holdout test is just a special case of time series cross-validation with 1 fold, but as we will see there are a few nuances to consider when configuring a holdout test and analyzing the results to inform a business decision.

### Configuring a holdout test in the Faraday platform

A typical scenario where a holdout test can be used is lead scoring. We will walk through how to run a lead scoring holdout test in the Faraday platform, but the same principles apply to a holdout test for other use-cases (e.g. predicting churn).

#### What is needed for a holdout test?

To configure a holdout test a user needs lead and conversion records with PII, dates, and a way to distinguish between the two groups. The setup for this can vary, but perhaps the simplest form is a single table that includes the following:

- PII columns
- a `lead_created` date that captures when a record became a lead
- a `conversion_date` that captures when that lead record became a conversion

NOTE: In our working example our conversions are called `tier_1_leads`.

In addition to the raw data, a user needs to have some prior knowledge of the typical attribution window for conversion which we will discuss in more detail when configuring the cohorts.

#### Datasets and events

Given a dataset described above, we need to configure the PII as one or more identity sets and two event streams, one for leads and one for conversions.

Configuring the leads event can be as simple as mapping a single date column  
[![Lead event](/images/blog/cross-validation-blog/lead_event.png)](/images/blog/cross-validation-blog/lead_event.png)

Similarly for the `tier_1_leads` event  
[![Attainment event](/images/blog/cross-validation-blog/attainment_event.png)](/images/blog/cross-validation-blog/attainment_event.png)

At this point our resource graph has a dataset table that emits two event streams  
[![Datasets graph](/images/blog/cross-validation-blog/datasets_graph.png)](/images/blog/cross-validation-blog/datasets_graph.png)


#### Cohorts

Next, we need to configure groups of people (cohorts) that partition the data into a training and validation set (recall this is time series cross-validation with 1 fold).

The inputs to determine this are:  
[![XYZ categories](/images/blog/cross-validation-blog/xyz.png)](/images/blog/cross-validation-blog/xyz.png)

[![XYZ description](/images/blog/cross-validation-blog/xyz_descriptions.png)](/images/blog/cross-validation-blog/xyz_descriptions.png)

The "prior period" is the training set for the holdout test, and the "holdout period" is the validation set. The amount of data available can influence how long/short these time windows are, but in general, we want to have enough data in the prior period to train a model and enough data in the holdout period to be confident in the results. For example, it would be bad to have zero conversions in the holdout period.

The attribution window is the time it takes for a typical lead to convert. Depending on the situation, this could be a few days to months.

The data lag is the time delta between today and the most recent date of any lead or conversion.

Once these have been determined we can create three cohorts

- prior leads: all records that became leads in the prior period
- recent leads: all records that became leads after the prior period
- conversions: all conversions (`tier_1_leads` in the working example) without any date filtering

Prior leads example configuration  
[![Prior lead configuration](/images/blog/cross-validation-blog/prior_leads_config.png)](/images/blog/cross-validation-blog/prior_leads_config.png)

Recent leads example configuration  
[![Recent lead configuration](/images/blog/cross-validation-blog/recent_leads_config.png)](/images/blog/cross-validation-blog/recent_leads_config.png)
 
Our resource graph now looks like  
[![Cohort graph](/images/blog/cross-validation-blog/cohorts_graph.png)](/images/blog/cross-validation-blog/cohorts_graph.png)

#### Outcome

Once cohorts are configured it is time to build a model and get some output. In Faraday, a propensity model is defined by an "outcome" which takes two cohorts as an input.  

[![Outcome configuration](/images/blog/cross-validation-blog/outcome_config.png)](/images/blog/cross-validation-blog/outcome_config.png)

Note that using the prior leads from the prior period implicitly restricts the conversions (tier 1 leads) to the prior period.

#### Pipeline

A pipeline applies an outcome (propensity model) to a group of people specified by a cohort. In this case we want to apply the outcome trained using data from the prior period to score recent leads records. Additionally, we want to have conversion information available in the output to analyze performance.  

[![Scope configuration](/images/blog/cross-validation-blog/scope_config.png)](/images/blog/cross-validation-blog/scope_config.png)

Here is the full resource graph  
[![Resource graph](/images/blog/cross-validation-blog/resource_graph.png)](/images/blog/cross-validation-blog/resource_graph.png)

This pipeline produces a table whose rows correspond to the recent lead records and whose columns include:  
[![Payload columns](/images/blog/cross-validation-blog/payload_cols.png)](/images/blog/cross-validation-blog/payload_cols.png)

Any of the `fdy_outcome_` columns can be used together with the `fdy_cohort_*_member` column (which is true when a row is a conversion) to measure model performance.

### Analyzing results

The output from the above pipeline can be used for custom analysis as desired by the user. One example is the automated reporting that Faraday produces to analyze holdout results and potential ROI.

#### Holdout results

First, we look at the internal performance metrics of the model built using the prior period.  

[![Training data used](/images/blog/cross-validation-blog/training_data_used.png)](/images/blog/cross-validation-blog/training_data_used.png)

These metrics are based on cross-validation described above and distilled down into a single plot.  

[![Model results](/images/blog/cross-validation-blog/model_results.png)](/images/blog/cross-validation-blog/model_results.png)

Next, we look at the pipeline output to see how the model performed on the recent leads.  

[![Data used for holdout test](/images/blog/cross-validation-blog/data_used_for_holdout_test.png)](/images/blog/cross-validation-blog/data_used_for_holdout_test.png)

Again, we can produce conversion rates by decile. 

[![Holdout test results](/images/blog/cross-validation-blog/holdout_test_results.png)](/images/blog/cross-validation-blog/holdout_test_results.png)

The last step of the holdout analysis is to compare the observed results from the holdout test with the model performance computed at training time via cross-validation.  

[![Comparing model report to test results](/images/blog/cross-validation-blog/comparing_test_results.png)](/images/blog/cross-validation-blog/comparing_test_results.png)
 
If these results are comparable, then it gives the user confidence that the performance metrics observed at training time are reflective of the model performance moving forward.

#### ROI calculation

When additional business context is known, the holdout test results can also be used to estimate ROI. For example  

[![Lead rejection ROI calculator](/images/blog/cross-validation-blog/roi_calculator.png)](/images/blog/cross-validation-blog/roi_calculator.png)

## Conclusion

Cross-validation can be used to prevent overfitting and evaluate a model's performance at training time. A holdout test is just a special case of cross-validation in which a single fold is determined by a point in time. Combining these two approaches with business assumptions enables the user to deploy models confidently, with a sense of expected performance and insight into expected ROI.

Faraday uses cross-validation to evaluate model performance, ensuring accuracy and value by training on multiple data subsets and validating on others. This allows us to make informed decisions and deliver reliable ROI for our clients.

How Faraday validates our predictive models

As businesses race to integrate AI into their software, many turn to first-party data modeling techniques like [RFM (recency, frequency, monetary value)](https://en.wikipedia.org/wiki/RFM_$market_research$) to build predictive models. These models typically rely on user interactions, including website navigation, email engagement, and purchase behavior to anticipate action. But first-party models alone have significant limitations, lacking the depth, speed-to-deployment, and predictive accuracy that businesses need to get ahead in today’s competitive market. 

Faraday’s models solve these problems by integrating rich third-party data from the [Faraday Identity Graph (FIG)](https://faraday.ai/blog/unlocking-the-power-fig), out of the box. The FIG enhances our models by integrating high-quality, consented offline insights into your first-party data, creating a more complete and actionable dataset. By enriching your existing data with verified external datapoints—such as demographics, lifestyle indicators, and purchasing behaviors—the FIG fills in gaps, improves predictive accuracy, and speeds up deployment to help businesses reach the right audiences at the right time with the right product.

## Why first-party data alone falls short

First-party data has its advantages, but it also has critical limitations that can hinder AI-driven personalization and prediction.

First-party data is information that businesses collect directly from their customers and audience through owned channels. This includes website activity, purchase history, and email engagement. Because it comes straight from the source, first-party data is highly accurate, privacy-compliant, and essential for personalized marketing and predictive modeling. However, it can be limited in scope, which is why many businesses enhance it with external data for deeper insights.

One major challenge to overcome with first-party data alone is the [cold start problem](https://www.sciencedirect.com/topics/computer-science/cold-start-problem). When a new customer visits your business, there is no past activity to analyze, making it impossible to generate useful predictions without supplemental third-party data. A first-party data-based model will most likely only improve over time as the customer interacts with the brand, but this delay can mean lost opportunities, particularly in industries where real-time data is crucial like financial services.

Even as first-party data accumulates, it often remains noisy and incomplete. Activity data is unpredictable, with various factors influencing when a user engages with your site or makes a purchase, meaning first-party models must quickly adjust to these fluctuations. External factors like age or income level are also essential for understanding customer behavior, but they’re typically not captured in first-party data. Additionally, first-party data only reflects [activity, not affinity](https://faraday.ai/blog/why-first-party-data-and-rfm-is-not-enough). For instance, knowing that someone purchased a pillow tells you little about their broader lifestyle, financial capacity, or long-term interests.

## How Faraday enriches customers’ datasets to power our models

Our data enrichment process uses identity resolution to securely match first-party identifiers—such as email addresses or hashed customer IDs—with the FIG’s extensive dataset before they are fed into our models. This process ensures a seamless connection between your existing data and additional insights, following three key steps:

1. **Secure matching** – Faraday uses identity resolution techniques to match first-party data with offline records, ensuring privacy and compliance along the way.  
2. **Data enrichment** – Once linked, Faraday appends high-quality demographic, behavioral, and lifestyle datapoints to each record.  
3. **Model integration** – These enriched profiles are seamlessly fed into our AI models, enabling more accurate predictions and personalized customer experiences. 

[![Data-appends diagram](/images/blog/first-party-merged-w-fig.png)](/images/blog/first-party-merged-w-fig.png)


## Powering better AI with enriched data

Our models’ ability to access the **Faraday Identity Graph (FIG)** solves numerous problems experienced by first-party-only models. Here are some examples:

### 1\. Solving the cold start problem

With the FIG, even first-time visitors are no longer unknowns. By matching first-party identifiers to offline, third-party data, Faraday provides insights into likely interests, purchasing power, and household composition, allowing our models to make informed predictions from the very first interaction.

### 2\. Boosting model stability and accuracy

AI models trained solely on first-party data must constantly adapt to new patterns in an ever-changing environment. The FIG introduces stable, high-quality datapoints, reducing reliance on transient, high-variance signals like click-through rates. This results in more durable models that perform consistently over time.

### 3\. Enhancing AI with affinity data

Instead of just looking at past transactions, the FIG integrates data that reflects who customers are, not just what they’ve done. This includes household income, education level, homeownership status, and even media consumption habits.

## The best of both worlds

Faraday’s FIG takes you data to the next level. By seamlessly integrating high-quality, consented third-party insights with your existing first-party information, Faraday empowers you to access more accurate and predictive models through our platform. With this rich array of demographic, lifestyle, and behavioral datapoints, you can unlock deeper insights into your customers and make smarter, data-driven decisions. The Faraday Identity Graph goes beyond the limitations of first-party data alone, giving you a clearer, more holistic view of your audience.

Want to see how the FIG can supercharge your AI models? **Let’s talk.**

The Faraday Identity Graph (FIG) allows users to create more accurate, stable, and predictive AI models within our platform with deeper insights into customer behavior and preferences.

Creating better predictive models with the Faraday Identity Graph

In the world of predictive analytics, data is everything. The more accurate, comprehensive, and up-to-date your data is, the better your models perform. That’s where the [**Faraday Identity Graph (FIG)**](https://faraday.ai/docs/abstractions/traits) comes in—a massive, carefully curated dataset designed to help businesses make smarter decisions with ease.

## What is the Faraday Identity Graph?

The Faraday Identity Graph is a powerful combination of multiple data sources, meticulously cleansed, de-duplicated, and feature-engineered to create one comprehensive dataset. It consists of [1,500 datapoints](https://faraday.ai/features/faraday-identity-graph) covering virtually all U.S. adult consumers and their households, offering a deep well of insights into demographic trends, property ownership, life events, financial indicators, and more.

And what’s even better is that it comes built-in with our services.

## So where does the data come from?

We partner with **best-in-breed data providers** like [Epsilon](https://www.epsilon.com/us) to compile a rich dataset that encompasses:

* **Demographic information** (age, education, income, etc.)  
* **Property details** (homeownership, property value, etc.)  
* **Life events** (moving, marriage, new child, etc.)  
* **Financial indicators** (credit attributes, equity, etc.)  
* **Buying patterns** (how they spend across dozens of categories)  
* **Shopping behavior** (e.g. in-store vs online)

By combining multiple premium data sources, we eliminate the need for businesses to juggle multiple vendors—saving them time, effort, and money.

## How frequently is the data updated?

The Faraday Identity Graph is updated regularly throughout the year as we receive new data from our vendors. The frequency of updates varies based on data type, but customers can rest assured that the data contained in our graph is as up to date and relevant as possible to ensure the most accurate and useful predictive datapoints.

## Can I buy the data directly?

Yes\! We offer direct data licensing, allowing businesses to purchase cleansed, consolidated, and competitively priced data. If you're looking for a variety of data sources, without the hassle of amalgamation and the time and expense of working with a range of vendors, we should talk—it’s likely we can save you all that time and money by bundling everything you need into one seamless package, and since we already made that package, you could be accessing that data tomorrow\!

## How can businesses use this data?

Our clients use the Faraday Identity Graph in a variety of ways but here are two core use cases:

* **Data appends:** Businesses can purchase specific datapoints to enhance their own internal models and processes.  
* **Predictive datapoints:** We can combine our data with your first-party data to generate powerful datapoints that answer key questions, such as: 
  * **Where are your best customers?** Read about our [market opportunity analysis service](https://faraday.ai/docs/abstractions/MOA).  
  * **Who is most likely to buy?** Check out our [lead scoring quick start guide](https://faraday.ai/docs/quickstarts/leads/lead-scoring).  
  * **What are they most likely to purchase?** Dive into our [product recommender](https://faraday.ai/docs/abstractions/recommenders) system, which predicts the best products or services for each individual.

Whether you're looking for raw data or full-service predictive modeling, Faraday makes it easy to harness the power of data-driven decision-making.

## The Faraday advantage

What makes this data even more valuable is access to our team of forward-deployed data science experts. Our team works alongside you to ensure that you are accessing not just the highest quality data, but the right data for your specific needs. With the addition of this service, whether you’re just looking for raw data or full-service predictive support, Faraday gives you everything you need to unlock smarter decisions.

Additionally, Faraday delivers a broad suite of other services like real-time scoring, giving you instant access to dynamic, data-driven insights that enable rapid decision-making and model explainability, ensuring you can trace every prediction back to its data-driven roots—this transparency enhances model trust and guides smarter adjustments. 

What's more, with immediate access to this rich data from day one, you can avoid the cold start problem, quickly deploying predictive models without sacrificing accuracy. Lastly, Faraday’s flexible API integration allows you to seamlessly incorporate our data into your existing systems, providing a scalable solution that evolves alongside your needs and analytics workflows.

## Questions?

Sounds pretty great right? We think so too. If you want to learn more, reach out to talk to our sales reps or even our data team to learn how Faraday’s data can push your business to the next level\!

The Faraday Identity Graph (FIG) is a massive, meticulously curated dataset of 1,500 attributes covering virtually all U.S. adults, enabling businesses to enhance their models, streamline data sourcing, and make smarter decisions with ease.

Unlocking the power of the Faraday Identity Graph (FIG)

At Faraday, we often hear from clients about their experiences working with our competitors. These conversations give us valuable insights into how our platform stands out in the market, and where we have room for improvement. Recently, one of our clients shared a specific experience that really resonated with us, and we wanted to take the opportunity to highlight it to our audience.

This client, who recently worked with [Think Unlimited](https://thinkunlimited.com/) before moving to Faraday, pointed out some key areas where we outshine our competition. As she put it:

*"The transparency of the data science behind the modeling \[at Faraday\] was refreshing. There were no secrets. It was our data, and it was given back to us in a digestible fashion. We could also easily access the experts behind the science to ask questions."*

We believe that transparency is crucial, and we’ve built our platform with the goal of making data science accessible and understandable for our users, so this was rewarding to hear.

This client also observed a significant difference in the quality of predictions between Faraday and Think Unlimited, noting that Think Unlimited's data science just didn’t seem as robust under the hood. This aligns with our commitment to providing high-quality, validated predictive models that you can trust.

But those observations are just the start. Let’s take a deeper dive into some of the key differentiators we've observed ourselves and how they contribute to Faraday’s unique value proposition.

## Faraday vs. Think Unlimited: Our core differentiators

### 1\. Real-time scoring with Faraday's Lookup API

One of Faraday's core strengths is our ability to score leads in real time. Thanks to our powerful [lookup API](https://faraday.ai/docs/features/lookup-api), users can get immediate feedback on lead quality, before bids are even placed. By receiving immediate feedback, businesses can optimize decisions in real-time, leading to better outcomes in high-pressure or fast-moving situations, such as the [ping post process](https://faraday.ai/blog/so-what-is-a-ping-post-anyway). Think Unlimited, on the other hand, lacks this level of real-time scoring capability, which can slow down the workflow, impact decision-making timelines, and in the case of the ping post process, result in lower quality leads being purchased. 

### 2\. A more robust data set for better predictive power

Our built-in Faraday Identity Graph is a rich third-party [data set](https://faraday.ai/features/faraday-identity-graph) that goes far beyond basic metrics by incorporating a wide range of insights that enhance the accuracy and depth of our predictive models. In addition to traditional datapoints, we include rich demographic, lifestyle, and behavioral data, and this combination of static and dynamic data allows us to understand not just who a customer is, but how they engage with products, services, and brands. Using these third-party insights, Faraday’s models can create a dynamic, multi-dimensional picture of potential customers, leading to more accurate predictions of future actions, such as conversion or product interest.

Faraday’s robust data set also helps solve the [cold start problem](https://www.sciencedirect.com/topics/computer-science/cold-start-problem) by providing this rich data from day one, meaning that brands don’t have to wait for a build-up of their own first-party data before accurate scoring can commence. It also allows them to immediately target new customer segments or groups that they don’t have existing data for, facilitating expansions and allowing them to quickly roll out new campaigns and products.

The depth of our data enables Faraday to generate more accurate, quick, and actionable insights compared to our competitors, empowering businesses to make data-driven decisions with confidence. 

### 3\. Score explainability: know exactly why leads get their score

Faraday emphasizes clear [score explainability](https://faraday.ai/blog/faraday-score-explainability) with all of our models. This means that users can actually know exactly why each lead receives a particular score. This level of transparency is crucial for businesses that need to understand the "why" behind their results. Having this clear visibility into the model’s scoring also helps businesses gain confidence in their decisions and ensures they can optimize their approach based on a better understanding of the process.

That earlier client specifically highlighted this point as being one of the biggest advantages she appreciated with Faraday compared to Think Unlimited, where such transparency can be harder to come by.

### 4\. Flexible integration across use cases

Faraday isn’t just limited to lead scoring. Our platform is adaptable and can fill a range of other use cases, including [product recommendation](https://faraday.ai/docs/how-to/use-cases/customers/next-best-offer), [lead suppression](https://faraday.ai/templates/lead-suppression), and even [churn prediction](https://faraday.ai/templates/churn-scoring), [plus a bunch of other options](https://faraday.ai/templates). This versatility allows businesses to leverage Faraday’s insights across different departments and applications, and also sets them up for success when expanding their business or launching new types of campaigns they haven’t undertaken before. Think Unlimited’s capabilities are more narrowly focused, which can hinder adaptability and expandability.

### 5\. Access to experts

At Faraday, we take our data science seriously. But we also understand that predictive modeling is inherently complex and that just presenting where data comes from doesn't necessarily mean that the client will be able to understand it. 

Score explainability measures are a critical first step but sometimes, the best way to ensure a score is actually explainable is pretty straightforward: just have an expert explain it. That’s why we made the game-changing choice last year to forward deploy our entire data science team, making them directly accessible to our clients—our CTO even works with a range of client accounts to ensure that they’re getting the most possible value out of their investment. This level of open communication and transparency is something Think Unlimited lacks and is the ultimate failsafe to ensure users are able to understand their predictions, and the models behind them.

### 6\. Fully API-based for seamless integration

[Faraday is fully API-based](https://faraday.ai/docs/reference), meaning it can be integrated into any system programmatically, offering unmatched flexibility and customization. This structure not only simplifies integration but also provides the flexibility to adapt the models into any custom workflows, making it an ideal choice for companies looking for a solution tailored to their unique needs. Our API is also well-documented, which facilitates integration into clients’ existing tech stacks and ensures easier operation throughout the customer lifecycle. 

These features dramatically increase speed-to-deployment, with some clients seeing major cost savings or revenue growth [within their first month](https://faraday.ai/blog/optimized-direct-mail-outreach-in-one-month) after signing on. Think Unlimited, in contrast, lacks the API-based framework that makes Faraday so adaptable to a wide range of use cases and workflows.

## Quality predictions and transparency make the difference for our clients

At the end of the day, the quality of the predictions you rely on is what matters most. Our predictions are not only proven to be accurate and value driving, but they’re constantly monitored and revalidated to make sure we’re delivering as much value as possible to our clients. Additionally, Faraday’s data science, transparency, and flexibility make it the clear choice for businesses seeking high-quality predictive analytics that can scale and adapt to their needs, all while remaining understandable and accessible. Think Unlimited may have its place, but when it comes to powerful, explainable models and real-time scoring, Faraday leads the way.

Faraday outshines competitors like Think Unlimited with real-time scoring, a richer data set, transparent predictions, flexible integrations, expert support, and a fully API-based platform.

Faraday vs. Think Unlimited: Our core differentiators

[Direct mail remains a core acquisition strategy for many financial services providers](https://www.lob.com/blog/the-benefits-of-direct-mail-marketing-in-todays-economy), but it comes with a hefty price tag. Each mailer represents an investment, and when sent indiscriminately, costs can spiral with little return.

When this client first came to us at the end of 2024, like many in the industry, they relied primarily on instinct-based direct mail strategies, meaning a significant portion of their outreach was reaching consumers who were unlikely to convert. They knew they were overspending but lacked the tools to pinpoint which mailers were worth sending—and which could be cut without hurting revenue.

This client needed a solution that could optimize their outreach, reduce waste, and maintain strong enrollment numbers. Most of all, they wanted confidence in whatever decision they made—all without disrupting their existing workflow. So we provided them a custom predictive datapoint that did exactly that.


## Lead suppression with Faraday

And that’s where Faraday came in. As soon as we identified their challenges, we quickly devised a targeted strategy to help. 

The answer? Faraday’s [lead suppression](https://faraday.ai/templates/lead-suppression/recharge) template.

To enact this strategy, we started by inputting the provider’s recent lead data into our system (maintaining rigorous data privacy standards like HIPAA, GDPR, and SOC-2 compliance throughout the process). This data was then enriched through our [identity resolution](https://faraday.ai/blog/identity-resolution) process with 3rd party data from the built-in Faraday Identity Graph (FIG), which [includes over 1,500 attributes on more than 240 million U.S. consumers and their households](https://faraday.ai/features/faraday-identity-graph), and features nuanced demographic and behavioral indicators that influence consumer actions—ranging from property ownership to life events and purchasing habits.

From there, new predictive models were created and trained on this historical enrollment data (now enriched with 3rd party insights), identifying patterns and trends that distinguished high-value leads from low-value ones. Then, using advanced machine learning techniques, the system assigned each lead a predictive score, ranking them into ten distinct deciles based on their likelihood to convert. With this scoring system in place, a [cut-off threshold](https://medium.com/swlh/determining-a-cut-off-or-threshold-when-working-with-a-binary-dependent-target-variable-7c2342cf2a7c) was established to determine the point at which the cost of printing and sending a mailer outweighed the expected return, ensuring that outreach efforts remained cost-effective.

Finally, Faraday delivered an optimized, high-fit mailing list back to the provider and armed with this refined list, they were able to move forward with confidence, ensuring that every mailer sent had a greater likelihood of driving enrollment. For a visual breakdown of this process, check out our recent blog on lead suppression, [here](https://faraday.ai/blog/holdout-testing-with-faraday).

So how did it go? 


## The results (so far)

If you noted the timestamp on this article, you might be thinking to yourself, “Well, they reached out at the end of 2024 and this was published on Feb 21, 2025, so did they even get the models rolled out yet?” 

Well, despite the short timeframe, they not only deployed the models and sent direct mail based on their predictions, but have also already seen a significant reduction in costs, all while maintaining strong engagement with high-quality leads. 

Indeed, by suppressing low-likelihood recipients, they were already able to send over **80,000 fewer mailers**, producing **a net savings of nearly $40,000**, demonstrating the immediate financial benefit of a more targeted approach.

These early financial wins were certainly affirming, but it had only been a month. How did we know that they were actually the result of our model and not just an uptick consumer interest this year?

### Ongoing model validation

At Faraday, we take the [validation](https://faraday.ai/docs/reference/getscopeefficacy) of our models very seriously. We do this before predictions are deployed through our [holdout testing process](https://faraday.ai/blog/holdout-testing-with-faraday) and continue testing on an ongoing basis even after successful implementation. In this case even though we were already seeing positive results, we wanted to see data that confirmed (or challenged) our assumptions. So we conducted further analysis on our models after we started getting results back to confirm that our predictive scoring was actually aligning with real-world outcomes. 

And we found what we were looking for: Leads with the lowest predictive scores showed below-average conversion rates, validating the suppression strategy. Meanwhile, leads in the highest-scoring segments consistently outperformed the average, reinforcing the value of prioritizing the right audience. And what’s even better is that this correlation is only expected to strengthen as more data flows through the model, further refining predictions and maximizing efficiency.

While these financial results would be impressive on their own, they also highlight how quickly Faraday’s models can be deployed and optimized. The contract was signed in late-December, and within weeks, the provider was already seeing meaningful returns. This quick turn around, combined with the ability to continuously refine and improve targeting over time, positions them for even greater efficiency and cost savings in the months and years ahead.

## Next steps

With direct mail suppression already demonstrating clear value, the provider is now looking to further optimize their customer outreach efforts. A key next step? [Call center list scoring](https://faraday.ai/stories/call-center).

Similar to how Faraday’s predictive models helped refine direct mail targeting, they can also help [prioritize leads](https://faraday.ai/templates/lead-prioritization) for call center campaigns. By focusing call center agents’ efforts on high-likelihood prospects, the provider can reduce wasted calls, improve conversion rates, and increase overall campaign efficiency.

We can’t wait to see how they use this technology to generate even more value for their business\!

## In conclusion

In just a short time, this financial services provider has revolutionized its direct mail strategy, reducing unnecessary costs while maintaining strong engagement with high-value leads. Faraday’s predictive modeling has enabled them to move beyond broad, instinct-driven targeting to a data-backed approach, ensuring each mailer sent delivers greater value. And with tens of thousands of dollars saved in only a few weeks, the provider has already realized the immediate financial benefits of a more targeted outreach strategy.

What’s more, looking forward, the provider is well-positioned to refine their targeting approach even further. By continuously updating and optimizing their predictive models, and exploring additional use cases like call center list scoring, they can maximize the efficiency of all customer acquisition channels. 

For this company, this is just the beginning of a smarter way to decide who’s worth reaching—and when.

If you'd like to know how Faraday can produce similar results for your business, **sign up for free**!

In their first month with Faraday, this financial services provider used lead suppression to cut 80,000 low-value mailers and save nearly $40,000.

All posts