Predictive AI: How You Should Implement Propensity Modeling to Maximize ROI

Thank you for joining us today at Windfall's webinar series about predictive AI. And today, specifically, we're gonna be talking about how you should implement propensity modeling in order to get the most out of your model. Joining us today, there's me. I'm Chris Ferrioli. I'm the data science lead at Windfall working alongside my colleague, Matt Donahue. He's the head of solutions engineering here. Both of us have been at Windfall for quite some time, and we've been working together on these models and with customers like you for at least five years for both of us. So quick overview of what Windfall is. We are a company based in San Francisco. We are on a mission to help to provide data driven nonprofit organizations and to help you more efficiently engage with your constituents and donors. What this means for you is you're going to be trying to get more donations, and we are going to help you to do that. We've identified more than twenty million, affluent households in the United States. That's out of more than one hundred million total households that we, are able to provide data on. We currently work with more than fifteen hundred customers, and we have one hundred trillion dollars of wealth that we have, attributed to different households within our, within our dataset that we're able to work with you with. So before we get into the crux of what we're gonna talk about today, and I'm gonna hand over to Matt for the first part of the conversation, I do want to just highlight one thing that we are offering. It's called Windfall Academy, and this is to give you a little bit of a hands on demonstration of what we're doing behind the scenes and to give you a little bit more experience, like, what modeling actually is. So we are offering something called Windfall Academy, predictive AI for nonprofits. I'm going to drop the link to this in the chat, so it is available there. You can also message us afterwards to get more information about it. But this is a three part course, and this actually goes into the nitty gritty of machine learning, how we do it, how we prepare the data, what the algorithms are, and even how you would promote this, a model and talk to your, supervisors and and the people that are would be approving the use of a model within your organization. It's a three part course. It ends with a certification. There's homework that goes alongside of it, and we are starting the next cohort on August fourth. I just dropped the link, but it is here in the slide deck as well. We are also offering a hundred dollar off promo for anybody who has attended this webinar. We just finished up one a few weeks ago. We got some great feedback. I do think it is a really interesting course, especially if you're interested in machine learning. You don't need to be a super technical person to come into this. We are trying to gear this towards people who are just getting started or just interested so that you can really have the conversations and start to build your own models if you want. So check this out if you're interested, and we'll be happy to see you there. As for what we're gonna be going through today, Matt's gonna kick off. He's gonna start off with the windfall overview, and then he's gonna start to talk about the what predictive I AI is, challenges that your nonprofits may be experiencing, and how we can help them. We're gonna talk about our approach to predictive AI, what we're doing on our side to to provide solutions to those problems. Then we're gonna talk about the best practices and actually implementing AI workflows. That workflow wording is key right here because if we have a model, but we don't know what to do with it, the model is just not that useful, and you're not going to get strong ROI off of that model. So then we'll talk about how we're gonna leverage propensity scores within your CRM so that you can activate them. Then we're gonna provide a demo of our application, which is really how we can, like, turn these numbers into actions. And then we'll wrap with some q and a at the end. For now, I'm gonna turn over to my colleague, Matt Donoghue, and he can get started. So, Matt, why don't you take it away? Thanks, Chris. Yeah. As we're kicking things off here today, we always like to start with a customer success story or case study that really highlights the importance of well screening as well as predictive modeling. So for this example, we're featuring a very well known medical center and health care organization within New York City who's leveraging Windfall for wealth screening to identify affluent donors and patients and specifically partnered with Windfall to help improve their grateful patient program. I think the the results sort of speak for themselves here in terms of the value of using data driven solutions and servicing hidden gems within the CRM file. We were able to identify a very large cohort of individuals for the major gift officers to pursue, and this really helped inform their portfolio allocation. So as they're thinking about where they're spending their time each day and stewarding larger transformative gifts, you know, making sure they're leveraging data to make some of those decisions and really extending beyond, you know, some of that first party data they have today. You have a really good understanding of who the major gifters are within their ecosystem. But there are a lot of folks, right, who have certainly additional capacity or interest or intent to make a major gift that partnering with, you know, Windfall can help you surface for your organization as well. And so we're able to help identify a number of hidden gems here who donated over a hundred thousand dollars to the organization and really help change how they're allocating portfolios, both in adding new folks to portfolios as well as making the hard decision to reallocate time, as well. And so as we're diving into today's webinar, you know, really aimed here to provide insight to nonprofit professionals who wanna use data and predictive AI to achieve some of those outcomes that we've seen with, that particular health care organization and many more clients that we partner with today. In terms of what you'll learn, you'll learn a little bit more about Windfall and our mission, high level overview of our products, how we leverage predictive AI, within our solution, and how you can leverage it in your workflows to drive more effective fundraising and examples of what an actual implementation looks like. You know, as Chris mentioned, you know, having a model that can help predict an outcome is super helpful, But where the value is really unlocked is getting that in the hands of your frontline fundraisers or your major gift officers and really beginning to act on some of the scores that you're receiving. And then lastly, we'll round things out with a preview of the Windfall application. We have some new enhancements that are coming out shortly that we'll take a look at today as well as a live demonstration of how you can really begin to use a PTG score within our application to begin to ID on your strategies as well. What we won't be covering, you know, information regarding our underlying data and methodology, we do have a separate webinar series called why wealth data is really hard that we definitely recommend you take a look at if you're interested in learning more. We're not gonna dive into the the nitty gritty on how different CRMs, connects to Windfall, but we're certainly happy to discuss that, with your organization following the call as well as more detail on your specific modeling goals or what pricing might look like for your organization. So we're gonna kick things off here with a poll. What is the biggest pain points for your organization's fundraising efforts? So one moment, you'll see that pop up here. So our options for this poll are identifying your best prospects, missing hidden gems in your database, prioritizing outreach, as well as fragmented data across multiple datasets. As responses are trickling in here, it's interesting. There's a pretty even split across a couple different categories. One is really around identification. You know, how do we find the best prospects within our ecosystem to pursue? And that's coupled with fragmented data. That can certainly make it a lot more challenging, right, to find those prospects. And so leveraging a solution like Windfall with unlimited screening can really help you identify those prospects at scale. And so with that, we're gonna cover a brief windfall overview, just to give you a sense of what we're all about. And so as we're out in the the market here, you know, talking to prospects as well as existing clients, you know, we hear a lot of different questions as they're thinking about their overall fundraising strategy. And these are the sum of the most common ones that we hear. Right? So what is the wealth of my constituent base? You know, how many hidden gems are there? Can you help me figure out how to prioritize who I'm reaching out to? And who really has the ability to give in the first place? Just because someone's wealthy, you know, they may not necessarily have the likelihood to make a gift. How do I increase my donor's gift size? What are the best ways to really engage with prospects? And all this really comes down to how do you get the most value out of data, and how do you put that data into action? So creating a workflow using wealth screening or predictive modeling. And what we see with organizations is it really this data maturity curve. As you're building out a a data driven machine, right, you may be in sort of the earlier adoption phase where you're really establishing your CRM, you know, unifying some of those fragmented systems into a central hub. You're really beginning to segment based on historical transactions. So beginning to understand things like recency, you know, largest gift someone's made, to really begin to inform how you are going about assigning folks to a portfolio or even thinking through things like event planning or your upcoming gala. And in this layer, you know, typically, we see organizations, you know, identifying that well screening would be really beneficial to their their team. You know, they built that foundational layer within CRM, and they're just missing some context on folks in the file that could certainly help them get more creative. So they're thinking about how to segment that file and construct different campaigns or outreach. And so that's where our partners like Windfall come in with our unique third party dataset to help provide you additional context into household level net worth and other characteristics of your donors or constituents. Now once you have that foundation in place, you can really begin to drive a lot of different outcomes for your organization. This can start with building predictive models, which is the topic of today's webinar. But there are a number of other use cases for this data. You know, as we're thinking about improving the efficiency of your direct marketing, maybe the quality of pieces that you're sending out or the frequency of outreach, data can drive really important outcomes for your organization there in increasing response rate or the size of donation by investing in, you know, the right constituents or donors for a particular activation. It can also be used to identify, you know, folks who may be in portfolios today. They actually have the capacity to contribute a lot more to your organization. And this is where, you know, outside of, prospects that are new to file that are unassigned, you know, there are likely other hidden gems where you have existing relationships that you didn't realize there's an opportunity to continue to steward that relationship for a larger gift. And, lastly, you know, once you have that in place, you can begin to orchestrate different tactics. This could be, you know, dashboards and reports within your CRM, you know, alerts as we're thinking about changes in the data. You know, has someone had a recent life event or recent liquidity event or timing, you know, maybe more, important for outreach. We saw they just got promoted to their first VP job, or we saw they recently moved. Could be a good opportunity to reach out to that individual. And really beginning to create some automated workflows using the data. We have many partners who leverage us for, programs like programmatic direct mail, which can really begin to help nurture relationships at scale, you know, versus doing one time selects for direct mail. And, you know, beginning to embed data further into other parts of your organization as well if you're building your own models, for example. And so with that, a little bit more on our vision here. You know, we really wanna change how organizations understand and engage their donors. You know, regardless of where you are on that data maturity curve, data can play a really important role in how your organization is fundraising. And we're trusted by over fifteen hundred nonprofit organizations across the country, of all different shapes and sizes, you know, leading health care systems and universities to, you know, cause and care related organizations, you know, local bicycle chapters. If you're looking to drive your fundraising outcomes, you can partner, you know, with OneVal to meet that end. And how organizations typically use our data, you know, the first step is really identifying those highest value donors and prospects within your ecosystem and helping develop workflows to prioritize those individuals. The second is, you know, better understanding these constituents and donors. You know, what characteristics do your top donors have in common? You know, if we apply that heuristic to the broader population of your CRM file, where are there other opportunities? Maybe we could take a look at those that have made several gifts to your organization and have a trust associated with their household and a, you know, decent net worth profile. That could be a really good fit for a planned giving portfolio. So by taking a closer look at the data, it can really begin to inform some of your strategy. And then lastly, you know, how do we engage with these donors or constituents? Using the the career data or life events can really begin to set up, you know, repeatable workflows for your fundraising and development team, or perhaps you're creating specific lists for outreach. You have upcoming, capital campaign or upcoming events. You know, how do you identify who you should be engaging, to hit those objectives you have for your organization? And so Windfall has built a platform that's really for data driven development organizations. And there are a couple different products that we offer that help you better understand your constituents as well as improve your fundraising outcomes. And so the first one is wealth screening and wealth screening premier. This provides you insight into your constituents and donors so we can better understand their net worth, where they work, and other context that's helpful for stewarding larger gifts. We then also have propensity modeling, the topic of today's webinar, so leveraging machine learning and data science to really prioritize those records within your CRM and really surface those that are most likely to make a gift at a certain level. And then lastly, we have data link. If you have multiple databases or fragmented datasets, data link can be super helpful, in identifying shared records across those systems, like a a golden record, if you will, and layering in additional insight into predictive models. For example, in the the grateful patient example, you have sort of your patient data and your your donor fundraising data from CRM. We wanna consider, you know, both of those datasets as we're thinking about scoring a household. And something unique about our business model is pricing is not based on the number of records or credits. We provide unlimited syncs for all of our products. And the objective here is really to help you identify those hidden gems within your CRM and provide you timely insights as our data's updating every week. And just a a recent statistic here, we have over a thousand models in production today with customers. So certainly a very popular product outside of our our wealth screening package. And so as you think about, you know, the the value of Windfall's unlimited model relative to some of the other legacy players in the space, you you don't have to pick which constituents or donors you're screening. So if you're leveraging a partner today who has a credit or consumption based model, it can be really difficult to know how you want to spend those credits. And there's likely a lot of affluent hidden gems within your CRM file that you're not screening. So with removing the need to curate a very specific list and test in small batches, you're able to achieve, you know, data enrichment and inform workflows at scale using Windfall's data. And this is where, you know, wealth can be super helpful, but it's not the only factor as we're thinking about prioritizing constituents. And this is where, you know, artificial intelligence and additional prospect research, you know, certainly goes a long way. And so if we consider two profiles here and their their net worth profile, this can help us better understand their potential capacity to make a transformative gift. And what is not considering, though, is that constituent's relationship to your organization and their historical giving. So this is where, you know, your first party data in concert with WinFall's third party context can build a highly predictive model for your organization. So we wanna take into consideration, you know, what's their level of engagement with your organization? What gifts have they made in the past? And this can provide a lot of context on their likelihood to give in the future. And then there are other characteristics, right, outside of just net worth that can provide us insight into the lifestyle or life stage of that particular constituent. You do they own a boat? Are they philanthropic and supporting other organizations? Do they own multiple homes? And so if we're using, you know, net worth alone, you know, both of these prospects look pretty qualified. You know, they have some degree of engagement with our organization, and they have the the ability to make a a gift of a a decent size to our organization as well. But with a predictive model, we can really help you optimize your workflows in cases where it may not be the wealthiest prospect that is the best one to reach out to. And this is where the machine learning can pick up on a lot of hidden patterns in your data to help us surface those that are most likely to convert. And so if you think about your database, you've acquired a lot of, you know, prospects or constituents within your CRM, you know, maybe ticket buyers, you know, folks who've attended events, your galas, etcetera. What we're hoping to do here is to help you identify those individuals that are most likely to donate to your organization. And with traditional wealth screening models, you kinda have to flip over each of these dots here and assess, you know, are they likely to make that donation to your organization? And third party data can certainly help you prioritize your database, but it can be difficult to look across a number of different attributes at scale and make decisions. And so this is where Windfall's predictive models are designed to help you identify those that have the highest likelihood to to donate, and convert for your organization without the need to, you know, flip over each stone here. And so as we think about your your data, you know, we really wanna leverage, you know, Windfall's proprietary data set and predictive AI to automate this process for your team and get them focused on building meaningful relationships versus having to interpret this complex data within their CRM. And some of the use cases we see for predictive AI across our nonprofit portfolio, you know, really depends on the organization's objective. We've highlighted a a handful of different types of predictive models that we've built with organizations over the years. We actually just hit our ten year anniversary over the weekend. So we've been in this business for quite some time. And so as we look at the different types of predictive models, you could think of, you know, increasing annual fund participation. These may not be the wealthiest constituents in your file, but helping prioritize those that are likely to participate in the annual fund or those that may be likely to make a leadership gift. And so these would be two separate models if you're looking to really grow your annual fund performance. And then the third type here is our most popular, you know, identifying those major gifters. So establishing within your CRM, you know, who's made a major gift historically, what can we learn from those trends, and apply it to the broader broader dataset. And then we also support other types of models, you know, grateful patient, we touched on briefly, as well as planned giving, and then other types of models as well as we think about response for direct mail and the likes. And so with that, I'm gonna pass things over to Chris after this next poll, to begin to discuss our process in more detail. But our next poll here is how do you use predictive AI within your organization? Give me one second here to launch that. This is a bit of a an easier question to to answer. Do you use them today or not? So, yes, we currently do. No. We don't. Yes. We have in the past, or we're looking into it. Alright. Well, looks like we have a pretty inquisitive bunch here with us today. Around fifty five percent of attendees are looking into using machine learning, so we're glad to have you here for that journey. Roughly twenty percent use machine learning models and predictive AI today in their workflows. And then the balance either don't currently use predictive AI or have in the past. So a great opportunity to sort of revisit that for your organization. Okay. I think that means that it is over to me. Thanks for all that, Matt, and I will take over now. Hopefully, we can get all of you that are actively looking to use machine learning over to the yes category. And the remainder of what we'll be talking about today is if you do make that transition over or if you already are using, propensity modeling, how do you implement this? How do you make it work in practice? So, I'm gonna start to talk about just predictive AI in general and what exactly that means. So before I even get into how Windfall approaches it, I wanna give you a quick overview of, like, what we're talking about when we say predictive AI. So, like, what is what is AI exactly? As a term that's, like, gotten stretched to mean pretty much everything from whether it's ChatGPT to just, like, your autocomplete in your email or, like, your your iPhone, it can mean a lot of things. Let's try to ground that a little bit. It's not very helpful if you're trying to figure out what you need to use for a specific objective. There are a lot of things that come under the AI umbrella. First one is that we've listed out here is machine learning, and that's what we're gonna be spending our time on today because that's what we're actually using for the models that we're creating, and this is what people are calling predictive AI. We also have things like natural language processing and speech and voice recognition, which comes from things like deep learning models. And then, of course, you've most likely heard in recent years of this new thing that's coming out, this generative AI. That's things like Gemini and ChatGPT and Claude. That's the branch that's really made AI a dinner table conversation within the last couple of years. But but for fundraising, the branch that matters most is machine learning, and that's what we are using to build our models that are going to score your database. Now all of these types of AI generally do relate to each other, and I like to think of it like a hierarchy. And here's, like, the most simplistic hierarchy that we can keep in mind as we're discussing AI in general. Artificial in general artificial intelligence, I think of as just the largest envelope. Everything within it falls within artificial intelligence. What this is is any program that can do something like sense or reason or predict. And it it it's it's a broad enough category that even, like, Excel's goal seek function, if you're familiar with that, is technically an AI that we can use. Now within artificial intelligence, we have a few different categories. One of them is machine learning. And this is the family of algorithms that improve with more data over time. You give it, something that you want it to do. You give it data to do that, and these algorithms run through a couple of mathematical equations, and you get something out at the other end. Now deep learning is just a a slightly different version of that. It uses something called multilayered neural networks. This needs a lot of data and a lot of computing power to be useful. So this is why within, like, the last ten, fifteen years, that's when it's really started to take off. Machine learning has been around for a long time before that. Even probably familiar with some simpler things like, a linear regression model from your your college statistics course if you've taken that. That's a that's a pretty good analog of what machine learning is. Now more recently, we've gotten generative AI, which is just a further specialization of deep learning. But the the difference here with generative a lie AI versus deep learning or machine learning is that it's not specifically predicting something. It's actually generating new content. So when you go to ChatGPT and you're asked it to create a picture, it's going to create a picture. It's going to use pictures that were created and pictures that were to train the model, but it's going to be creating a new thing. Whereas previously, deep learning and machine learning, they're creating predictions. They learn from from data what a pattern looks like, and they understand what pattern comes next. So when somebody says that we're using AI for fundraising, what they usually mean or at least what they should mean is that they're using machine learning, and that's the layer that we're gonna focus on as we go throughout the rest of this conversation. There we go. Okay. So there are generally two big families that machine learning can be split into. The first one is unsupervised learning. This is when you don't give it a label. You just give the algorithm a whole bunch of data, and you ask it to find patterns within this data. So these are really useful because they answer questions like, what are the common traits of my top donors? You end up with clusters. You end up with groups, and you can develop personas from those groups of people that the algorithm determined were were groups. The other type of machine learning is supervised learning. This is something where you have a known label and you learn to fit the model towards that. So, like, what this person is a major donor. Yes or no. That's your label. And then the algorithm is going to learn to predict that label from all of the features that you give it. That's the rest of the data. So the question is, like, how will I predict who will be a top donor? And that is what our propensity scores are doing. So our propensity modeling, our predictive AI is all based on supervised learning. So the good news for everybody here, these algorithms are all pretty standard. We don't need to go into the depth on all of them. Random forest, gradient boosted trees, logistic regression, whatever they may be, they've been around for a long, long time. But what separates a good model from a bad model is not actually the algorithm that you choose. The algorithm can certainly help. There are there are times when you want to choose one algorithm as opposed to another algorithm. But the most important part for making a good model is that you have good data to feed in. So we spend a lot of time on the data side of this, and Windfall is, at its heart, first and foremost, a data company, and we're applying these machine learning algorithms on top of that data in order to fit your objectives. Sorry. Slide transitions are a little bit slowly. So a quick example of what Windfall can do when we're trying to work some clustering models, some unsupervised, some unsupervised learning. On this slide, there's an example of a k means clustering algorithm, which is one of the most commonly used clustering techniques. You essentially tell it how many groups you want. Like, maybe you wanna find three groups in your data, which is what this illustration is showing, and it will find you three groups in your data. The way that this works is that the algorithm drops in a couple of random points. In this illustration, those are the points that are moving around, and then it calculates the distance of every data point from the closest randomly generated point, one of those three randomly generated points. It then measures all that distance and tries to minimize that. So it iterates through. It makes sure that the points are closer and closer and closer until they don't move anymore, and that is how a cluster is defined based on which point is closest to which one of these what are called centroids. So once you've assigned every data point or in the case of a nonprofit donation world, every donor is assigned to a cluster, you can profile the groups. So one cluster you might determine is younger and more digitally engaged. Perhaps those are the features that you put in, or maybe that's just something that came out based on other, information that was fed the model. Or maybe you see another cluster, and that's the wealthy, cluster that just gives infrequently. And those are the kind of things that you'll learn from clusters. So what you can do with clustering models is design outreach strategies based on each individual cluster. You let the model figure out the common traits of each group, and then each different group you may want to approach differently. So it's a different tool than propensity modeling, and it's useful for a different set of questions. So as for how we set up modeling for your organization, we generally go through the same workflow for any customer, any organization that we're working with. Our first step is to start with wealth screening where we set up the transfer of data between you, your organization, and Windfall so that your data can get enriched with our data, and we can provide that wealth information back to you. The next thing that we wanna do is understand the makeup of your constituent base. The reason that we wanna look at this and understand this is because different organizations, of course, have different constituent bases. A university foundation is gonna look totally different than something like an animal shelter. So we spend some time understanding your first party data alongside with our third party data to make sure that we understand how to best support your organization. The next thing we do after we've understood what your organization is and what your data looks like is develop and agree upon what model we actually want to, develop for you. In order to do this, we're going to use some rock chart. There's some things called rock charts, which is a data science term. You'll learn about those in the Windfall Academy if you choose to come to that. But we also look at things that are a little bit more tangible, like lift charts and gain charts tells you how much money, was attributed to, for example, the top decile of scores. So we understand how a model performs before it goes live. We make sure that you have alignment with us, you understand how the model is, and you're confident in the model. If it's not behaving, in the way that you'd expect it, we wanna hear about it at this point of the loop, so we can go back and make sure that we have the model working up to your standards. The next thing we do when we agree on everything, only then do we actually product productize the model and make sure that the scores get enriched back to you and make sure that they get into your workflow so that you're using them and you're able to get the most out of this model. Of course, we want to prove that the model works, so we are going to measure and iterate as time goes on. As I'm sure you know, major donors are a slow and small population. So we do need to look at this over time and understand how the model is changing over time potentially and performing over time. So this is actually a loop because every piece of new data that comes in gives a little bit more information to that model, which gives more accurate predictions. So here's what our propensity models are actually doing. We score every constituent in your database, and we give it a zero to ninety nine score. The higher score means they have a higher predicted likelihood of giving. So those ninety nines, they look a lot like what your current major donors look like now. Your current major donors should score high as well. So you should see a group of people that you know are your major donors along with some people that you might not have expected or are new major donors that are the ones that you want to start to cultivate. So the scale is calibrated on the model objective that you define. So perhaps one organization, a major donor, is a two thousand five hundred dollar donation per year. For another organization, maybe you're looking at a hundred thousand dollars. We totally understand that those two organizations and the donation behavior of the people that will give at those levels is totally different. So we want to build a model that is specific to you based on your specific target. So a high score for you means what it should. We then update the scores on a regular cadence. A donor score isn't just fixed today. There are wealth events. People move. People donate to your organization or other organizations, and we pick up on that data flow and those data changes. So just because somebody scores a certain score today doesn't mean they won't score a different score tomorrow. Data is live. Data is changing. We wanna make sure that you have the most up to date scores that you can take action on. So to make this a little bit more concrete, on the left here, we have demographic and predictive score information, that we are taking. These are the inputs into the model. They're the net worth, career signals, philanthropic history, and things like that. Now on the right, we're trying to figure out if that makes somebody a major gift candidate. The model is combining everything on the left and predicting whether somebody is going to be a good major candidate on the right. So the critical piece here is that the predictive scores need to be refreshed regularly, either weekly or monthly ideally, so that a donor who spikes an engagement, like, maybe they've come to an event or they've donated a couple of times in a particular month, you want to know immediately. You want to know when that happens, not six months later if you send an enrichment. So the cadence and looking at these things in action and regularly really helps pick up on those signals so you can strike while the iron is hot. So let's think about what Windfall can provide against the alternative, some of which you may already be implementing. If we look across the board at heuristics, RFM models, internal data scientists in your organization, or windfall, you're going to need fresh data across the board. Appropriate predictive modeling and even heuristics depends on fresh data. We're able to provide that to you. A lot of these, you want to understand whether it takes a lot of setup or a little bit of setup. Heuristics are pretty straightforward, but they might not be the most predictive things in the world. RFMs and internal data sciences teams, those are a lot harder to set up. Windfall, on the other hand, we do all of that for you. We make your model. We productionize your model, and we make sure that we advise you on how to actually implement it. Now machine learning is an important way to make better predictions. So heuristics and RFM, they are not going to be leveraging machine learning, so they're not going to be as powerful and as predictive. Internal die data scientists certainly can do that, and Windfall does do that as well. Next question is it's easy to update. Heuristic's easy to update. Maybe you have has somebody donated a thousand dollars in the last year as a heuristic? You can update that to fifteen hundred dollars if you want really easily. RFM models take a little bit of updating. Internal data science team, if you want to change what they're doing or how they're doing it, that's not going to, be the easiest thing to update. But windfall, we have a lot of products. We have a lot of different ways that we can support you. So we're we're flexible, and we can make sure that we, change models to keep, keep appraised of, like, your current needs. Next question, of course, is cost. An internal data scientist is going to require salary. They're gonna require a lot more time and attention and budget than something like Heuristics or RFM or Windfall. So what we are trying to do, what all this comes down to is that we are trying to democratize predictive AI for all nonprofit organizations. To make sure that you get all of these things, all of which are all the requirements for having informative predictive AI, we do need to have all of these things. Windfall is able to provide these things where some of the other solutions, struggle a little bit. So now let's actually talk about the best practices for implementing AI into workflows. You have a model. The model is being enriched back. What does that mean for you? What are you gonna be able to do with it? So a quick poll before we move on there. Let me get this fired up. Sorry. One moment, Matt. If you could help oh, here it is. Alright. So, yeah, if you are interested in Dictive AI, do you know how to implement it correctly? And the answers are oh, this is looks like the answers are not populated correctly, but let's interpret this as yes. We know how. No. We don't. And we're trying to figure out how to how to implement it correctly. So it looks like so far what's coming in is that most people are, interested in doing it. Maybe you have some scores, but you're trying to figure out how to actually do this correctly. That's what we're gonna be going through right now. So now we're gonna talk about the best practices. Everything we've covered so far is about building model and what Windfall is able to provide that maybe other things can't. Next section or this section, we're gonna be talking about how we actually use it. So there are three moves that we have seen that our successfully implemented customers have in common compared to those who might not see the results that they're looking for. And we're happy to advise on how to get all of these three things and make sure that you can get the best, performance out of the model as possible. The first thing that we see our successful customers do is that they segment. They don't just look at the raw scores. They combine them with not only other data attributes, but with information that you have about the relationship that might not make it into the model. We don't want to take away from your personal relationships with your known donors and and your other constituents. We know that you are the experts there, but we want to use that along with these scores to help you discover new ones. So we want you to segment based on your information and your know how and not just treat the scores as the only thing to use. We don't want you to tell your officers, call this person that you've not called because they have a high PTG score. We want you to make a story and create these segments that say, this is why we're gonna call these people, and we're going to use our intuition as well. The second thing that we see our successful customers do is that they are actually going out and activating these scores. They get the scores in front of your frontline advertisers, whether that's directly in your CRM or if it's just through, like, a report or dashboard or something like that. But, critically, educating your team on what the scores mean. Whoever is interfacing with our customer success team, they generally have a good idea what's going on and how the models work. But we need to really educate the team as well. And this is for understanding things like maybe a score of seventy to somebody might not sound good, but in a model, a score of seventy could be really high. So if your if your officers don't understand what makes a good score and even how scores were calculated, they might not understand how to read the scores. They won't trust them, and they won't use them. So activation through education and just getting these numbers in front of your frontline officers is really important. Finally, we do need to run tests. We wanna compare high score outreach against low score or, like, other segments and make sure that the people who are scoring high are actually donating more or achieving your objective more than the other, segments that you're creating. Now what is important here is that we need to track the pipeline, generate the GIFs by score track the pipeline and, look at different score buckets and understand what GIFs are coming in. So this is how you're building trust with the team. You wanna actually show we implemented the model, we did this test, and it did show performance. Now all this, again, is a loop. Once you've gone through it once, you create new segments or may maybe your segments were really good. You go with them again. You activate again. You measure again. So all of this is a loop of implementation to make sure that everything works. So a lot of what the activation of the scores comes down to is understanding what your current score distribution looks like because this can trip people up a little bit. So the chart on the screen shows a real customer score distribution, and the dashed vertical lines show percentiles. In this case, it's the ninety ninth, the ninetieth, and the fiftieth percentile. So for this particular org, the ninetieth ninety ninth percentile shows a score of ninety four or higher. What that means is one percent of your population is going to receive a score of ninety four or higher. The ninetieth percentile, so the top ten percent of your scores, get a seventy seven or higher, and then you see a big drop down to the fiftieth percentile, which is only a nine or higher. So the key takeaway here is that we're not grading on, like, grade school, grades here. Good scores don't need to necessarily be in the eighties or nineties depending on what you're doing and how many people you want to reach. A score of sixty might put a donor in the top decile, and they might be perfect for whatever you're trying to do at that moment. So we wanna care about the percentile, not the absolute number. So the whole point of this slide is to prevent your team from just dismissing a score because it's fifty five or sixty or whatever it may be because it sounds low because in context, it actually may be pretty high. So thinking about percentiles again, this is how we try to drive these conversations of score prioritization. Once you know the shape of your distribution, the question becomes more, how many prospects can your team actually handle? So for example, maybe you look at your top one percent. This is a major donor model. They should be your major gift portfolio. Your top five percent, maybe they're the group below that, your mid level givers, and your top twenty five percent are your annual giving campaign audience. And you might want to have outreach to all of these different groups, but do it differently depending on what score range they're actually in. Now the right cutoffs are going to to depend on your team capacity, and, of course, that's gonna be different across any organization. If you're a large organization with a ton of staff, your major gift team will be able to work a wider distribution, a larger distribution than a smaller organization. Like, a a three person shop, they might not be able to reach out to a thousand people, but a larger team could. So setting your come your cutoffs based on your team, throughput is how we need to think about this and how we actually turn these scores into action. So a quick reminder too that for prospects that you can't assign to a gift officer right now, maybe you don't have that throughput, you shouldn't just drop them. You should have rules of engagement of, like, what category people should put into. Maybe you can do an email nurturing campaign or a direct mail or or digital retargeting or something like that. There's different ways to use these scores that will allow you to get different reach. So the method of how you contact people can vary based on what your team can support and what your team is trying to do. So we can also look at things like ratings because they can really help to simplify scores, for all of your teams. So what we return back is a score between number of between zero and ninety nine, whereas a rating is just a tier. Ratings are a lot easier for your nontechnical teammates to act on. This, again, goes back to the percentiles. Is a sixty a good score or a bad score? Well, that depends on what the percentile is. But if we work on ratings and provide those ratings to the frontline officers, things can be a little bit easier to understand. So there's three things that you can do with ratings. You can reassign prospects. Like, maybe at the end of each month or each quarter, you take your highest rated prospects who have donated recently, have some kind of segment based in there, and put them into a qualification pool. Or maybe you just put them directly into a portfolio if their score is high enough and their donation is large enough. We can also look to prioritize the existing portfolios. Perhaps you could build some reports that will raise up the top rated pros prospects in each of your gift officer's books, and then that officer is able to focus their energy on those. You can also keep an eye out for who are your next best prospects. The great thing that we're able to offer is that you can do daily screening, and that will help you flag new opportune opportunities as they're coming in. You can prioritize the ones with the top scores or or whatever other data interesting data bits that they have. Maybe if somebody comes in who's very wealthy, has a high score, has a large recent donation, whatever it might be. So all of these come from the same underlying scores, but a different way of presenting those scores to the team so that your team knows how to act. So let's turn that into something a little bit more tangible. Let's just build this out real quick. Maybe you define something as a hot prospect if they have a propensity score that is in the top one percent and a gift in the one in the last six months. That means they're engaged with you. They have the capacity. They have the interest in donating. You know that. That's a great prospect. Now an a rating, maybe there's somebody that's still in the top one percent of the score, but they haven't given in the last six months. B rating or sorry. Tier tier three rating, in the next two to five percent and so on. Now what's interesting, if you look down here at tier six, you can put qualifiers on this as well. If you're looking at that, second quantile of the scores, they're the top fifty percent of scores, but not in the top twenty five percent of scores. If you put a qualifier that they must have a net worth of ten million and be in those scores, well, they could be an interesting, group of people to nurture as well. So depending on what you're trying to do, what your goals are, these can be moved around. You can put different ways, different heuristics on top of the models in order to create these tiers and to share these with your teams. So these aren't windfall defaults. These are rules that just fit this particular organization, this example organization based on this imagined team capacity and the objectives. Your rules will look different, but the point is that we need to actually define them, make your teams aligned, and then have them applied consistently and work them the same way. So a score is only one part of that. Coming up with how to use the scores is just as important as the model itself. Now let's look at this at, like, actual in an actual data level. We have a handful of people here, and we'll see how that breaks down into different tiers. If you look at Linda, Linda has given a donation in the last six months. She's in the top one percent of scores. She has a she is quite wealthy. So this person is excellent to have a hot lead rating. We should have her in a portfolio. We should be developing a relationship with her. Now compare this to somebody like Harry on the full right side. Very wealthy individual, scores decently well in the top forty percent. Why it's probably not higher with that wealth is because his last donation was only three years ago, and that recency does matter. So maybe this is somebody that falls into that bottom tier that they are interesting, and we should follow-up with them, but maybe don't prioritize them over some of these, higher rated prospects. So there's three examples that we can use to really make this a little bit more concrete. Let's say you have a donor who doesn't who hasn't given in five years, but they show up in the top two percent of your PTG scores. Maybe this is somebody that you can assign to a cultivation pool. This is gonna be like a business rule on your side that you would set up that will notify your prospect research team if this person does something like makes a gift or RSVPs to an event or something like that. So you know they're scoring high, and then you have some other action, donation, first time donation, or I guess it would not be a first time, but the day of RSVP'd to an event, something that would trigger some kind of outreach. You might not want to ask them for a large gift now, but you might wanna warm them up first. You might wanna become more familiar with them and start to develop their relationship. Now the next is somebody who's a first time four figure donor. They're in the highest percentile of scores, top one percent of scores. What should we do with this person? This person, given their high score, their large initial donation, maybe this is somebody that you wanna put directly into a portfolio. Start with a personal follow-up on the gift. The model is telling us that this person can do more. Maybe this person is your next best prospect that we need to develop that relationship with. And finally, maybe you have a previous principal principal donor. They have not donated in quite some time. But because of their large donation history, they still end up in the top fifteenth percentile scores. The model isn't showing strong scores that they're ready to give again, but you do know that this person has given largely in the past. If you still have a relationship with them, you should still continue that relationship to build that relationship. And when they are ready to give, make sure that you have a gift officer ready for that. So the same tools of the scores are given to every gift officer, but we're applying them a little bit differently based on what the model and your first party data are saying. So, again, the model is a tool alongside with your expertise to figure out exactly how to apply this. So now we're gonna look at a little two by two frame to help you come up with what we might want to do with a high score or a low score. If we look at the vertical axis here, that is the PTG score. We've just broken it into two buckets, high or low. Again, the definition of high and low will depend on your objectives and your team. The horizontal split is whether they are currently assigned to a gift officer or not. So that gives us four quadrants. We have an unassigned high score. These pre people should likely be moved to a portfolio. These are gonna be your hidden gems. Maybe they're not on your radar yet, but the model is telling you you should be, that they should be. Maybe this is the person that has donated five dollars every month for the last five years, but is also shown to be philanthropic more broadly, and they have high wealth. Maybe that's somebody that would be quite interested in giving a larger amount to your organization. We also have the assigned high scores. This is something that maybe you confirm the ask amount. The gift officer knows about them already, but the model is validating that this is indeed a real opportunity. That person is where they should be. So make sure that the mask make sure that the ask for the gift matches their potential. So we validated this person as well. You think so. The model thinks so. Let's just try to get that ask right. Then you might have people that are assigned to portfolios, but they have a low score. So this is where you should review portfolio qualifications. Maybe there's a legitimate reason that they're in their portfolio. Like, there's a board relationship or just something else that the model can't see. So you shouldn't just remove these people, but you do need to question whether they should be in a profile or not. If there's a reason to be in the profile, by all means, keep them there. But if there's no strong reason, then we should focus our resources elsewhere. Finally, we have our unassigned low score quadrant. These are for low touch campaigns, direct mail, email campaigns, that type of things. Don't spend your, major gift officer's time there. So this matrix here is a really simplistic but powerful way to actually implement the scores. It's gonna surface new opportunities, validate existing one, and it'll help you to rationalize your portfolio assignments whether they are in the right bucket or not. It's also easy to explain, which is half the battle when you're trying to roll out, a new predictive AI model. So finally, before I turn back over to Matt, rolling out scoring the model is really only part of the job. The other part, maybe even the harder part, is proving your team that it works and that they should use this. That may take some time and experimentation to understand exactly what sticks specifically for your team, but we do have advice on how we can proceed there as well. First of all, we need to design what we're going to do, and what that design could be is split your list into two groups. Maybe new prospects are ones at the model services that nobody was working with before. Existing prospects are already in someone's portfolio. You want to understand both groups and to tag both groups at the beginning. Say when we started a model, these people were newly assigned. These people were already there. Next, we want to actually implement, our outreach. So you're distributing these new prospects across the team. You wanna tag them so that they can find them again in six months. This most organizations skip this. And then the question of does the model work, we don't know. Finally, you wanna monitor, which allows you to measure the model. And if you see the model's performing, that's good. You keep going. If not, you iterate. So there are ways that we can help you out with this and make sure that you're doing this correctly. But with that, I'm going to hand back over to Matt to bring us through how to do this in your CRM. Awesome. Thank you, Chris. We are running a bit short on time, so I will be brief here. But, really, you wanna design workflows with your key stakeholders in mind. There are likely different consumers of your model within your organization. You'll wanna build reports and dashboards that are going to service each of these personas. And this could look like a a high level, dashboard that's just showing the lay of the land. You have different KPI cards showcasing opportunity within the CRM, and perhaps those that had recent life events or recency of donation, where you have the ability to prioritize using net worth as well as your propensity to give score. Secondly, you can also set up report subscriptions. There's typically a functionality available in most CRMs and just gives your team further insight into the highest priority prospects within your CRM on that recurring basis. And so you can set that to send out, you know, every week, every month, and really make looking at this data part of your team's rituals in terms of driving your fundraising goals. Then lastly, we have a a number of features available, you know, within our application. If you have more questions on how to leverage the Windfall application, you know, we're certainly happy to dive in as part of a a separate call from today's webinar. But it allows you to quickly prioritize prospects using that PTG score if that's included in your subscription as well as your wealth screening data and really begin to weaponize this information to make your teams as productive as possible.