What Makes a Renter’s Score Smarter?

Patience Hackler

Hi, everyone. Thanks for joining. I want to welcome you to today's webinar where we'll talk about what makes a renter's score smarter.

My name is Patience Hackler. I have been with TransUnion for 13 and a half years. I started out as an account manager working directly with our customers back in the day when we were on a product called Credit Retriever 5.

I moved over to the product team for a few years when we launched our new product, which was Resident Screening at the time. That's our current screening suite of solutions. And I now work as a strategy advisor for all things resident screening in the multifamily and single-family housing space. So that includes Resident Screening, Resident ID, the verification of income and what we're here to talk about today, which is Resident Score 5.0.

Whether you're a customer or you're evaluating screening solutions or you're simply interested in how predictive models are evolving within rental housing, I’m glad you're here. Today's discussion is called What Makes a Renter's Score Smarter?

It's not going to be a deep dive into mathematics or model development because that's a little above my pay grade. Instead, we’re going to focus on practical questions like "Why do different scores exist? What makes the score relevant to rental housing? How should housing providers think and evaluate predictive models?" We’ll then connect these concepts directly to Resident Score 5.0, which is our newest generation of an industry-specific rental scoring technology. And by the end of today's webinar, my goal is you’ll leave with a simple framework, not just to evaluate our score but to evaluate any score. Because ultimately, the question isn't whether a score is newer. The question is whether it's appropriately aligned to the decisions you’re trying to make.

Before we really get into it, a couple of housekeeping items. First off, we can have questions in the Q&A within the webinar, so please feel free to drop questions there. If I don't have the opportunity to get to the questions, I’ll follow up with you post-webinar. Additionally, this webinar will be available as an on-demand resource once the webinar is complete, and we’ll be sending out a recap email that includes a link to the on-demand version. And then next, before we jump in, a quick note from our friends in legal. If you'd like the complete legal language, it's on the screen for you to read in all its glory. But to summarize, any characters you see today are fictional. If you happen to know someone who sounds suspiciously similar, that's purely coincidence. Everything we're covering is intended for informational purposes and not legal advice. So, for any specific regulatory requirements, please consult with your legal counsel. Now that that's taken care of, let's get on to the good stuff.

Here's our roadmap for today's conversation. We'll begin with some foundational concepts around screening. Then we'll look at how models actually convert information into risk ranking. From there, we'll address one of the most common questions about scores: "How can the same person have multiple valid scores?"

Once we've answered that question, we'll look at why rental housing often benefits from an industry-specific model, and then we'll introduce Resident Score 5.0. And finally, we'll talk about what customers can expect as we prepare for the migration and rollout.

Throughout today's presentation, there's one theme I'd like you to keep in mind: A model is only as useful as the question it was designed to answer. And that will come up a lot today.

As we dig into models and the output of any model, which is generally a score, it's important to make sure we focus on the right thing. It's not about the score; it's about the outcome. When people see two different scores for the same applicant, the instinct is often to assume one is right and one is wrong, but that's not necessarily true. A score is not a universal grade. It's the output of a specific model built for a specific purpose. A mortgage lender may care about something different than a credit card issuer. An insurance company may care about something different than a housing provider. The real question isn't which score is highest. The real question is: What outcome am I trying to understand? Because a model is simply an analytical framework designed to answer a question. If you change the question, you may change the model. If you change the model, you'll change the score. And that's not an error. That's intentional. Different models are designed to fit a specific outcome and a population. To see why that's true, we first need to distinguish between a model and a score.

When people talk about scores, they're often talking about the final number. But the score is actually the end of the process. First, information exists. Common attributes from a credit report include information that demonstrates how someone manages their financial well-being. Things like: Do they make regular on-time credit card payments? Do they have a mix of different types of credit like car loans, mortgages, store cards versus universally accepted logos like MasterCard or Visa? How long is their credit history and how is their credit history trending?

Then a model evaluates that information. The model looks for relationships between all of the attributes available and determines how those attributes work together to inform how a consumer may behave in various situations, like: How will they manage their credit card debt? How about paying a car loan? Do they prioritize their open collection records? How are they paying rent?

Then a score is produced. Think of the model as a recipe and the score is the finished dish. Most models have access to the same credit information or, in the case of a recipe, the same ingredients. However, how those ingredients are used determines what your designed outcome is.

Are there any other food nerds out there that watch "Chopped?” It ends up being four chefs who all get a mystery basket that contains the same four items. They have access to these items in the same pantry. At the end of the show, they've produced four different dishes, but they had access to the same items upfront.

How they use those ingredients is what changed the outcome. In the case of models to scores, which model you use changes the numerical ranking of relative risk. And that's the word: relative. The risk is relative to the item you're trying to produce or predict.

When we talk about relative risk, we talk about how applicants are ranked compared to one another for a specific outcome the model was designed to evaluate. It's important to remember a score is not a prediction of what will happen to an individual applicant. A score does not tell you this applicant will or will not be evicted. It provides a relative ranking within a population. And that brings us back to the important distinction. The score is the output, but the model is what determines how that information is interpreted. The score itself doesn't tell you how the relationships were evaluated. The model does.

And that's why two models can look at similar information and still produce different rankings. They're applying different analytical frameworks to answer different questions. So, what determines how a model interprets information in the first place?

And it all starts with that question: "What outcome are we trying to predict?"

Once that question is defined, the model can be built. And once the model is built, we can produce a ranking. The order matters. Question first, model second, score third. If you change the question, the factors that matter may change. If the factors that matter change, the ranking will change. The idea can sound a little abstract, so let's take it visual.

Let's forget about credit scores for a minute. If I asked everyone to rank these athletes from best to worst, how would you go about doing it? Is there information you’re missing that you would need in order to rank these athletes? Might it be best to start with a question? It would probably be important to know "best at what?"

And that's exactly the right question because the answer changes if the question becomes "who is the best basketball player?"

Suddenly, the answer feels much clearer. Why? Because we know what skills matter: free throw shooting matters, court awareness matters, basic basketball skills matter. Swimming ability doesn't suddenly become irrelevant as a life skill. It's just not relevant to what we're trying to predict.

So, let's just change one thing about this question. Let's ask now, "who is the best swimmer?" The athletes didn't change, the talent didn't change, but the question changed. And now we're asking, "who is the best swimmer?" And immediately a different set of strengths rises to the top.

And that’s the same concept that applies to predictive models. Different outcomes require different signals. And that's the key insight I want you to remember. The athlete remains exactly the same, only the question changed. And when the question changes, the model changes. And when the model changes, the ranking can change. That doesn't mean one ranking is wrong. It means each ranking was designed to answer a different question. The same principle applies to consumers and credit scores.

Now let's move from athletes back into scoring. In the same way that basketball and swimming reward different combinations of skills, mortgage, bank card and rental models may evaluate similar credit information differently because they are designed around different outcomes. The consumer has not changed. The purpose of the evaluation has. Every industry has its own business question. Mortgage lenders have one, auto lenders have another, utilities have another, housing providers have their own.

So, when someone asks, "what score is best?," the more meaningful question is, "best for predicting what?"

For rental housing, we're not evaluating the likelihood of repaying a mortgage. We're not evaluating revolving credit performance. We're trying to better understand rental-related outcomes.

And that's where an industry-specific model can provide value. Not because the data is magically different but because the interpretation of that information is aligned to the rental decision. This is a subtle distinction, but it's the difference between using a general tool and a purpose-built tool.

When we're evaluating rental risk, we want a rental-specific lens. A smarter score isn't simply about having more data. It's about aligning the model to the decision you are trying to make. Are we evaluating the right population? We want to make sure that we're using relevant population to predict and answer the question we're trying to answer. And can the model meaningfully distinguish relative risk? Does performance remain stable over time? Resident Score was purpose-built with these concepts in mind, translating approved information into a rental-focused ranking that can be used as one input into a broader screening policy. A rental-focused model does not necessarily mean every input is rental-specific. It means the relationship among the available attributes is developed and evaluated against a defined rental outcome.

A smarter score isn't just about having more data. It's about better alignment to the decision you're trying to make.

So when you're evaluating a model, you want to be able to understand four things: Are we predicting the right outcome? Are we evaluating the right population? Can we meaningfully differentiate relative risk, meaning does somebody who has a high score have a meaningfully different risk evaluation than somebody with a lower score? And does the performance remain stable over time?

No single metric answers all four questions. Credibility comes from the combination of all four and that's exactly the framework we've applied when evaluating Resident Score 5.0.

So, with that in mind, let's just take a quick look at how Resident Score builds credibility.

What are the inputs? When it comes to building the model, we looked at two key components: the population of US renters and eviction data from across the US. The model doesn't look at all consumers because not all consumers are renters. If we want to know how renters are going to perform and how renters manage their financial well-being, we have to look at renters.

Because we’re building a rental-focused predictive model on a renter population, it's key we understand the outcomes available to renters: lease fulfilled, lease unfulfilled and renter evicted.

To understand that we need to be able to see historically who has been evicted to inform the model. We use a 24-month performance window to understand what a renter looks like when they apply to a community and how that journey ended for them, either lease fulfilled or eviction. By understanding the outcome, we can see what indicators there were at the time of application to inform the model of what attributes are most predictive of an eviction.

Then we have the output, which is Resident Score 5.0, a relative ranking of eviction risk. And finally, the application of Resident Score 5.0 to the leasing decision as a key component. A little bit of history. We first launched Resident Score 1.0 in 2013 as the industry's first specific model. Since then, we've launched a new model about every three years, with the latest model launched in 2022 but refreshed in 2025. So why launch Resident Score 5.0 now?

The answer is simple: The world changes. Applicant populations evolve, generations shift from renter to homeowner back to renter, the economic environments change, data evolves and our analytical methods improve.

A predictive model isn't valuable because it remains unchanged. It's valuable because it continues to remain relevant. Resident Score 5.0 represents that evolution. The rental-focus objective remains consistent. What changed was the modernization behind the model.

As we modernized Resident Score, two things happened simultaneously. Some components changed, some remained intentionally consistent. The model benefits from newer data, updated inputs and modernized development practices. What didn't change was equally important. It still has the rental focus. It still has a score framework. The scorable population approach remains the same, meaning that we analyzed renters to predict renters. And the overall purpose of supporting rental screening decisions remained the same. That's important because customers aren't being asked to adopt an entirely new philosophy, they're benefiting from a more modern version of the same rental-focused approach.

This is where we connect theory to evidence. One of the most important things from validation was stronger differentiation when risk matters most. So, follow along with me. There are going to be a lot of highs and a lot of lows in the next explanation. And the chart moves in the opposite direction than you would expect. But key to note, high scores equal low risk, low scores equal high risk. Again, higher scores represent lower risk, lower scores represent higher risk.

For this analysis, applicants were ranked from highest predictive risk to lowest predictive risk, meaning applicants on the left side of the graph have the highest predicted risk. Applicants on the right side have the lowest. We then examined where observed future eviction outcomes occur within those rankings. If a model is separating risk effectively, we would expect a large share of observed eviction outcomes to be concentrated among applicants identified as highest risk, that is lowest score. Those applicants on the left.

Looking at the highest-risk 20% of applicants, Resident Score 4.1 identified approximately 60% of observed evictions. Resident Score 5.0 identified 64% of observed future eviction outcomes within the same highest-risk 20% of applicants. In other words, Resident Score 5.0 demonstrated stronger separation by concentrating a greater share of observed outcomes within the portion of the population identified as the highest risk. Put simply, Resident Score 4.1 is a strong rental-specific model. Resident Score 5.0 builds on that foundation and showed improved differentiation in validation testing.

The comparison becomes even more meaningful when we look beyond rental-specific models. Compared with a broader credit risk score, which is used across all industries and built for a general credit well-being purpose, those models identified approximately 52% of observed eviction outcomes within the highest-risk 20% of applicants. Resident Score 5.0 identified 64%. So that's a 12-point improvement by moving from a general-purpose score to a rental-specific score.

Now it's important to be precise about what this does and does not mean. This is a population validation result. It does not mean that 64% of applicants in that group experience an eviction, nor does it predict what will happen to any individual applicants, and it certainly doesn't make a leasing decision. But it does demonstrate the model's ability to distinguish relative risk more effectively within the validation population.

That's the real takeaway. Not that the score makes decisions but that it provides a strong rental-focused risk signal that can be used as one input into a broader screening policy. And ultimately, stronger model performance only matters if it translates into practical value for housing providers.

So, let's talk about what this means in the real world. A smarter score helps housing providers do three things: Align the model with the rental question, better distinguish risk differences and apply screening policies more consistently. For customers, the good news is much remains familiar. The scoring range is the same. We’re still 350 to 850. The role within the screening remains the same. It’s still accessed through Resident Screening, whether you're coming through an integration partner, a property management system or our website. There's no technical implementation required to take advantage of Resident Score 5.0. The model is evolving, but the customer experience remains intentionally unchanged, which leads to the question many customers care most about: "How will this transition happen?"

Our objective is to make migration simple, supported and seamless. The upgrade is managed on our side, communication is planned throughout the transition, and the customers are not expected to complete any technical implementation work. The focus is helping customers realize the benefit of the updated model while minimizing operational disruption.

Migrations will begin Oct. 27 and run through Nov. 10. You’ll receive an email in the next week with your target migration dates, and your account manager will be able to provide full details on what this means for you. Additionally, this webinar will be available on demand if you would like to share any of this information with your teams.

So, let's bring this all together. We started today's webinar with a simple question: "What makes a renter score smarter?"

The answer is not simply more data, and it's not a bigger version model number. It's not a single performance metric. A smarter renter score is built around the right question, developed for relevant population, able to meaningfully distinguish applicant risk and supported by validation evidence. That's the philosophy behind Resident Score and that's the foundation behind Resident Score 5.0, a model purpose-built for rental housing, modernized for today's market and grounded in the same rental-focused mission that has guided Resident Score from the beginning.

So, thank you for spending time with us today. I hope you got something from this. I kept it as high level as I could. It looks like I'm getting some questions, so I'm going to pause there for a second. With the intention that I could give you the foundational understanding of why we do rental score and what the benefits are of having score while assuring you the migration is seamless and supported throughout.

I have a question here: "So in addition to the Resident Score, will screenings include a risk recommendation based off of an applicant's credit patterns or trends?"

And the answer there is yes — because it's those credit patterns and trends that are feeding into Resident Score. So Resident Score is using credit attributes, including trended credit history. Those trended credit attributes include things like in their journey in credit, are they heading in the right direction or are they acquiring debt they can't manage? That's a trended credit history. It's not looking at a moment in time, it's looking at all of the moments up to now. And all of those moments are what are feeding into Resident Score and being evaluated to rank them in the relative risk.

I don't see any other questions. If you are a customer or a prospect and you are working with a TransUnion representative, please do feel free to share questions with them.

Oh, "So if we'll state high risk next to Resident Score, is a Resident Score the same as VantageScore?"

Okay, so first question: "Will it rank high risk next to Resident Score?"

The way this works is that we have decision-point thresholds that will provide a recommendation within your policies. So, on a scale range of 350 to 850, we have decision points that say, for instance, and not a recommendation, just throwing some numbers out there, that maybe somebody over a 700 would be a full accept. So what Resident Score is doing is it's giving you a score and then we have thresholds we set with you in our system that then provide a recommendation: accept, conditional, decline. We have a low accept based on those thresholds within the relative risk.

So, we're not necessarily identifying that this 550 is a high risk. We're saying this 550, in relative risk, has this estimated eviction risk, and based on your policies with us and those decision-point thresholds, this would be the recommendation that goes with it.

And then the second question: "Is the Resident Score the same as VantageScore?"

And no, it's not the same as VantageScore other than it’s using the same information as far as credit data, but it's answering different questions. So even within VantageScore, VantageScore has multiple models, FICO® has multiple models because they're all predicting something different. The most common VantageScore models out there ((which are models three and four) are used to predict general credit health and general ability to manage debts.

We’re very specifically predicting how they manage rental payments and rental outcomes. So, we're using the same credit attributes, but we're weighting them for a very specific outcome. So, it's not a straight correlation to VantageScore. I will say good credit is good credit. So generally, somebody who's above 800 is going to be above 800 on pretty much any model that’s looking to predict a bad financial outcome because good credit is good credit.

Where you're really seeing that difference is in anybody in those higher-risk categories because somebody who may look good on a general model may not perform great on an eviction model because of the mix of their credit and how it's weighted with a specific outcome in mind.

Vice versa, somebody who doesn't look good for a mortgage risk may actually be a very solid renter risk. So they may score higher on Resident Score because again, we're answering a very specific question. We're not answering, "How qualified is this person for any form of credit?"

We're saying, "How does this person manage their debts in a way that they prioritize their rental housing payments to avoid eviction?"

That's what we're trying to get to. So they may not look great for a mortgage, they may not look great for an unsecured personal loan. However, the way they're managing their debt does make them a good rental risk. So, they may score higher.

Then, let's see, "Does the report mention what attributes contributed most to the score for a particular resident?"

So, it will, you’ll get score factors. You’ll get score factors that tell you what attributes contributed to the score. Anybody who doesn't have a perfect score has to have what's called score factors. These are the four, usually the top four items that contributed to them not having a perfect score.

It's important to keep in mind they’re required on anything under an 850. For instance, if you've got somebody who's an 820 and their score factor is they have credit inquiries, we had to have a reason they weren't a perfect score. The weight of those varies based on what the rest of their credit history says.

There will be score factors that say things like: "They have too much revolving debt, they have too many inquiries."

For instance, if you go out and you apply for a bunch of credit cards this weekend, those inquiries are going to tank your score. Not recommended. Just what's going to happen because what we're seeing is somebody who's going out and trying to get a lot of debt simultaneously. That will impact your score. It will be a score factor. Trying to think of some of the more common ones, but there will be score factors to tell you what contributed to their score.

I don't see any more questions, and I do thank you for the questions. Like I was saying before, if you’re a customer or a prospect with TransUnion, please feel free to reach out to your account managers or your sales reps as they do have a direct line to me. They’ll for sure reach out with questions.

If you’re not and you're just interested, please let us know. You can respond to one of the emails. You can respond to a survey that comes out. You can go to transunion.com and find the rental-specific page for more information.

So again, I want to thank everybody for joining and I look forward to supporting you through your Resident Score migration. Thank you.