What do nonprofits need to know about AI and data science?

Farra Trompeter: Welcome to the Smart Communications Podcast. This is Farra Trompeter, co-director and worker-owner at Big Duck. Today we’re going to ask the question, “What do nonprofits need to know about AI and data science?” I know this is a question on many people’s minds and I’m really excited to get into it today. We’re going to meet and talk with Dr. Ashutosh Nandeshwar, who uses he/him pronouns and is senior vice president of data science and analytics at CCS Fundraising. I’ll tell you more about him in a moment. You may recall hearing other members of the CCS fundraising team on Big Duck’s podcast series before. In fact, Eric Javier was joined by Sarah Durham back on episode 54 asking how do you ask for really, really, really, really, really big donations? So that might be an episode you want to listen to and check out. And in fact, Eric reached out to me recently and said, “Hey, I think this might be a great topic for the Smart Communications Podcast,” so we’re excited to host this.

Farra Trompeter: So again, Dr. Nandeshwar has education in systems engineering, artificial intelligence, and design thinking. He’s built solutions to improve fundraising results at several higher ed institutions, including University of Southern California and University of Michigan. He’s written multiple books including co-authoring a book called Data Science for Fundraising and holds a PhD and MS in industrial engineering, specializing in machine learning and an MS in design thinking. Ashutosh, welcome to the show.

Dr. Ashutosh Nandeshwar: Thank you, Farra, for having me.

Farra Trompeter: Well, this is a meaty topic and before we get into it, I’d actually like to talk a little bit more about you. On your personal website, you say that your childhood made you who you are. Particularly, you shared that you and your friends spent countless joy filled hours imagining things, what you couldn’t get you designed and created. And I was just wondering if you could talk a little bit more that and how it’s actually led you all the way to today and your work at CCS.

Dr. Ashutosh Nandeshwar: One of the really good examples from my childhood, and I don’t take any credit for this, this was all my friends in my neighborhood and we just used to hang out outside of our houses for hours and hours after school and just imagining things, doing things. One real good example from early nineties when India had not opened its economy yet and it was slowly opening up to the foreign investment and we got cable channels for the first time in that time period. Before that, it was only a couple of channels. Some of your listeners might remember where you could twist the knob and change a couple of channels. So that was the scenario. But then cable channels started coming to India and then people started getting the service. And at that time, some of us could not afford the service. So one of my friends thought of this clever way of using old speakers from the audio systems, and trying to create small dish antennas from those to try to catch some of the signals and then hooking that magnets essentially to their TVs. And, it was not the best of the signals that it was catching, but it was watchable. So, those are some of the examples. And, I remember from childhood just this genuine curiosity of how things work, how to figure out things when you don’t have all the resources, but you still have your thinking and whatever tools are available in front of you, trying to make the most of it. And that stayed with me for a long time, even today where, first of all, trying to find the simplest solution to a problem, creating subtasks of a large problem so that every task can be solved by itself, and then that’s how you solve the larger problem. But the pursuit of creative solutioning has led me to my current role where that every job that I have taken is again, trying to find the simplest solution, the most even sometimes elegant solution. And building things out of those solutions, and speaking about those solutions led me to my current role at CCS Fundraising.

Farra Trompeter: That’s great and I think that’s certainly something we both personally, and our company, share this idea, the spirit of both curiosity and problem-solving. And now with that in mind, let’s dive into our topic and I want to zoom all the way out and talk about data, and how data can help nonprofits make good decisions. And also please help me understand, what is machine learning?

Dr. Ashutosh Nandeshwar: Well, there are so many layers of how data can help fundraising operations become more effective. At the simplest level is, you can even look at it, this kind of IBM had this analytics maturity curve where they say your journey in analytics, where you are, it starts with maybe at ad hoc level where you just have lots of different spreadsheets or maybe some papers here and there and some numbers recorded somewhere. But then as you start moving up in your trajectory, you become more sophisticated and then you have databases, you have reporting mechanisms. But what also happens when you increase your awareness as well as your practices of storing and recording data, your journey also moves. For example, in the beginning, it might be more looking back, you are trying to find what happened and your reporting also. Simplest example would be, what was your total annual fundraising last year? That’s looking back, that’s reporting. Some organizations might measure their major gift officers and their activity, so how many visits they had, how many proposals did they submit? But that’s all looking back. But then also again, and that’s a parallel with your analytics maturity is that then you start asking the question, okay, so what’s next? We can now confidently tell what has happened, can we now plan out our future as well? And can we predict some of those activities? That’s where predictive modeling as well as forecasting and machine learning comes into play. And then obviously a generative AI that we have been hearing about so much in the past year or so, they also play a role.

Dr. Ashutosh Nandeshwar: Couple of quick examples. First, yeah, you have to measure what has happened because if you don’t know, you really cannot make any progress or you don’t know if you have made any progress. So it’s critical that you measure what has happened. Second, as you move a couple of steps to plan your resources to find the prospects, you can employ machine learning. Machine learning is a subset of artificial intelligence where you take stock of all the input data that you have, all the variables you have, and try to squeeze as much as information as possible from those data points to predict a certain outcome. Some common examples are Amazon, Netflix, where they’re able to predict which movie you would like or which next book you would want to buy. Those are all built upon using machine learning mechanisms. And a similar example for us in fundraising would be who are your next likely major donors based on their previous giving history or certain attributes or interests and their participation with the nonprofit organization.

Farra Trompeter: That was helpful and I think in that you gave some good examples related to acquisition. So yeah, I really want to talk a little bit more about this idea of retention. And I’m curious how have you used data science and machine learning to solve problems, whether that is student retention, donor retention? When we were preparing for this conversation, I know you shared some examples of things you’ve done. You know, I’m curious not just how can we predict when somebody might, let’s say donate or not donate, but also can we use that information to make decisions around when and where we might communicate? So yeah, retention data, science, machine learning, tell us all about it.

Dr. Ashutosh Nandeshwar: Student retention was my dissertation research topic, so I spent good amount of time there. One of the things that I learned from that experience and that research was just the reemphasis of the statistical saying that “correlation is not causation“. So it is sometimes very hard to pinpoint which activity caused a future activity, but all of these different data points, just like how you mentioned using machine learning, predictive modeling gives you directional guidance and tells you maybe there are some things here that show us what could happen. And then maybe there are some steps, especially in student retention, there were these interventions that you could take, especially if you knew that there were some characteristics of some groups that made those students at risk, then you could proactively plan for an intervention and help them maybe provide some additional support in their registrations or maybe additional tutoring in some difficult subjects. But whatever that might be, you could actually be proactive and plan those steps.

Dr. Ashutosh Nandeshwar: And when you bring that to the donor retention problem, similarly, you may not know why a donor stopped giving unless you ask everybody and they tell you truthfully why they stopped. But there are some data points so we can study their recency of giving, their frequency of giving, and obviously the dollar value, the gift amounts. And using all those data points you can almost forecast for every donor, their future journey. And based upon those, again indicators and those values, if they stop or they slow down their giving or they give at lower amounts, maybe they’re at risk of completely dropping out and stop giving. So then again, maybe you can have some special programming, some special communication that could be directed to that segment specifically. And again, as we know to acquire customers or donors, it’s pretty hard, but it’s slightly easier to retain your existing donors. So as long as you can identify those segments using the methods that we just discussed, then again, creating good segmentation, and appropriate programming for those segments could help you retain more donors.

Farra Trompeter: Right. And beyond thinking about how this works with fundraising and your communications, I’m curious about an organization’s internal communications or operations or culture. And I’m curious if you have any examples related to how nonprofits can use machine learning to strengthen their teams and support their organization. Maybe you can share a little bit about when an organization might also incorporate this work into their staff, such as when to hire an in-house data scientist or folks to focus on analytics. So again, like Big Duck does some work in the area of teams and thinking about how we can strengthen teams. So I’m just curious how you can bring this to that conversation.

Dr. Ashutosh Nandeshwar: Yeah, as we were speaking earlier about analytics maturity and as organizations become more mature and there they start seeing results and they start seeing efficiency gains or higher revenue, that’s when they can think about certainly adding more stuff, augmenting their current capacities or even upskilling. There’s definitely a value in being data aware, but also there’s value in being able to analyze the data. And again, as we were discussing, forecasting some trends and planning actions based upon those trends. Some of those skills are innate to some people, but also some of those technical skills have to be learned. So investing in your staff so that they can do some of those things by themselves is great, but at times maybe they need to hire additional staff to support their activities.

Farra Trompeter: Yeah, I think it’s a good point. A lot of organizations collect data, but they don’t always pause and look at it and use it to make those decisions. And we often say, you know, data is one entry point that can help you answer the question, what should you stop? Where should you experiment? So making sure that you are spending time looking at it, and being not just aware of the data, but using it is really important. Well, you mentioned generative AI earlier, and I know a lot of folks are asking us questions or thinking about ChatGPT and a lot of other tools, while it’s been out for a bit, it’s still relatively new in the nonprofit space. I know that some organizations are diving in and experimenting, others are a bit nervous, cautious, concerned. And I’m just curious, in your opinion, how should nonprofits approach AI? Do you have any use cases or tips that they can use as they’re trying to kind of figure out what works for them?

Dr. Ashutosh Nandeshwar: And it is such a broad topic as well to find, and this is not particular to fundraising as well. Lots of for-profit companies, especially venture capitalists or startups, they’re trying to find out the right use cases for where it seems like there’s this powerful tool out there that can change everything that we know and can offer us all of these efficiencies. But at the same time, people are still struggling to find the right use cases. And I think that’s what I would like to share with the listeners is finding the use case is more critical than which tool you have. We always have had tools to solve different problems, but it’s really your choice. Do you want to use a hacksaw or do you want to use a circular saw? And if the problem requires a circular saw, sure you should use that, but if it doesn’t, then your simple hacksaw would work.

Dr. Ashutosh Nandeshwar: So that’s what I would encourage the listeners to think about the use case and the problems. Is this a problem, first of all, worth solving? Are the benefits clear to us? And then second, what’s the right tool to solve this problem? Can I use mail merges> And people, people have been using mail merges for a long time and they work fine, but if you want hyper-personalization, does it make sense to use ChatGPT or related tools to create very targeted fields or acknowledgement letters or proposals? Do you really see the benefit from taking that approach or do you still have to verify all those different thousand various combinations? Then is it really solving a problem? So I would pose it from more first think and then solve. That’s how I would approach it.

Farra Trompeter: I really appreciate that. Like be really clear on what you need. And then the solution might not even be AI, it might be something else, and that’s a great way to frame that. Well, before we go, I’d really just love to talk about how all of this interfaces with issues related to diversity, equity, and inclusion. I know there’s been a lot written about the harm and bias connected to AI and machine learning, and I’m just curious what you have seen and what you think nonprofits should pay attention to as they’re entering into this world.

Dr. Ashutosh Nandeshwar: There are two or three potential sources of bias and the dangers with the bias that is introducing these models. Let’s talk about generative AI first. So again, AI became so well known after ChatGPT was introduced, which is based on large language models also called LLMs. They are built upon all the vast amount of data that users have created, people like us, humans have created, including websites such as Reddit. And Reddit could be toxic. And if the models have learned from that type of language and the models main thing is to predict the next word based on previous words, then the likelihood of those models saying something offensive are pretty high. Although some companies have tried really hard to filter out offensive language, this couple of weeks ago in Vegas, there was a conference for hackers to force some of these large language models to say something offensive, and a lot of them were successful in getting those models to say some things not only offensive, but there was a use case where the model actually gave a recipe “how to discriminate against certain candidates when they’re applying for a job”.

Dr. Ashutosh Nandeshwar: So again, the models don’t have thinking they’re just reproducing the words based on the words that came before and replicating essentially what we have said on the internet. So that’s one source of problem. Second source of problem is just the missing data. Whose stories have not been told yet that are not part of this large corpus of data that we have. If winners write all the history, where is the history of other people who were oppressed or who were quote-unquote “on the losing side?” So that’s missing. So it’s incomplete. So those are two sources. The first one is maybe slightly easier to solve for if you see offensive language, then you can filter that out yourself and the companies will also do a better job at filtering that out. But the second one is more challenging. You’ll have to get out of the large language models to ensure that the stories that you’re telling are comprehensive and inclusive.

Dr. Ashutosh Nandeshwar: When it comes to predictive models themselves. Choosing your variables carefully is critical because if you include some variables that even if you didn’t think about those as bias-inducing, they could have some inherent bias in themselves. For example, at some organizations, the trend used to be even if the gift or the check was returned by the female partner of the household, the credit would go to the male partner of the household. So the person who made the gift never got recognized, the male partner of the household got recognized. And if the data is stored that way, then when you build the models, it would seem as if none of the females have ever made a gift. And then your future models are going to predict based upon that. So of course that’s an easy problem to solve because you just don’t include prefixes, or gender, or sex as part of your models. But there are some other variables that could be deceptive. You would not think that they have some value, but they could actually be biased. So studying every variable carefully when you’re building your models is also very critical.

Farra Trompeter: Yeah, and in that example, of course, as someone who is married to a woman and identifies as a woman, those models also leave out people who may be in same-sex relationships or maybe people who don’t identify as male or female or use other identities and expressions. So I think the world has changed, not just with the tools we use, but also our understanding of what makes a family, or relationship. So there’s a lot to think about in these conversations and I think the idea of being curious, asking what question you’re trying to answer or think about the use case, but also questioning and making sure you’re not just taking the information from these tools at face value, but really pushing it forward.

Farra Trompeter: Well this was really fascinating. If you’re out there and you’d like to connect with Ashutosh, you can certainly see what he and his colleagues at CCS Fundraising are up to at CCSfundraising.com. You can also connect with him on LinkedIn. I’ll spell out his name if you’re out there listening. It’s A-S-H-U-T-O-S-H, last name N-A-N-D-E-S-H-W-A-R. So search his name and you’ll find him on LinkedIn. We’ll also link in to his page over on our website. If you go to bigduck.com/insights, we’ll link over to your LinkedIn profile as well as his website, which is at hislastname dot info or nandeshwar.info. So be sure to check that out. Well, Ashutosh, before we go, any other parting thoughts or words of wisdom you’d like to share with our listeners?

Dr. Ashutosh Nandeshwar: I think you captured it, Farra. Just be curious, be on the lookout, keep learning and keep growing. The world is certainly changing, but we don’t want to be left out. But also I would say extend some grace to yourself because it is too much, sometimes it’s overwhelming. Even some of these leading experts think this is too much. Every day there’s a new AI model or something happening. So give yourself some grace, but also be on the lookout and keep learning and growing.

Farra Trompeter: I love that. Well, thanks again for being on the show and everyone out there have a great rest of your day.