Artificial intelligence is on the cusp of widespread adoption across the economy. Most Americans already interact on a daily basis with AI tools such as virtual assistants and advertising algorithms. Businesses are rapidly incorporating AI into their planning and processes.
So it only makes sense that charities and philanthropies would look to these same technologies to make their operations more efficient and to better connect donors with recipients.
A recent report funded by the Bill & Melinda Gates Foundation, “AI4Giving: Unlocking Generosity with Artificial Intelligence: The Future of Giving,” explores the current and potential uses of AI in the nonprofit sector, from enhancing fundraising operations to better understanding what motivates donors to support an organization. The report presents an array of opportunities—along with some ethical concerns—and we’re seeing its findings reflected in the evolving work of a number of nonprofits and tech companies.
AI4Giving came about when its authors, Beth Kanter and Allison Fine, approached the Gates Foundation with the idea to explore the benefits nonprofits could derive from AI. Parastou Youssefi, a senior program officer at Gates, said the foundation supported the project because it was “interested in learning more about the current applications, challenges and opportunities that emerging technologies present in the philanthropic sector.”
“Specifically, in what areas can technology and innovation lead to more efficiencies in fundraising or grantmaking practices? Are there ways that technology can help us to reimagine the entire fundraising model as it currently exists?”
What is AI, and What Tools Does It Offer Nonprofits?
In their report, Fine and Kanter write, “Artificial intelligence (AI) is an umbrella term used to describe different types of technologies. Though AI comes in many flavors and varieties, at its heart, AI is the use of computers to help perform tasks automatically that could previously only be done by humans.”
Darrell West, director of the Brookings Institution’s Center for Technology Innovation, defines artificial intelligence as “automated software that learns from data, text and images, and then makes intelligent decisions based on that learning. Its key distinguishing features are intelligence, learning and adaptability—these are what distinguishes AI from other past technologies.”
One of these primary AI capabilities, machine learning, involves the use of computer algorithms to process and understand large volumes of data to identify patterns and make predictions. Machine learning often employs natural language processing (NLP) to understand and generate human language.
Shawn DuBravac, an expert in emerging technologies who was not involved in the report, describes machine learning as “taking large amounts of data and discerning patterns humans can’t see. AI lets you identify a pattern you didn’t even know existed. For example, people who go to ice cream parlors on Saturday might have a higher propensity to give to your charity.” DuBravac indicated that as AI is democratized, it could level the field by allowing small organizations to do sophisticated analysis that formerly could only be done by charities with large staffs.
Machine learning is the most likely initial application for nonprofits, according to Mary Purk, executive director of AI and analytics at the University of Pennsylvania’s Wharton School, “given where we are now in terms of [nonprofit organization] skillsets.” Purk (also unaffiliated with AI4Giving) suggests that its best use is to automate laborious tasks, such as email communications, but cautions that to accomplish this, organizations need to ensure their data is clean and accessible.
Purk emphasizes that using AI requires that an organization’s team must also acquire the right expertise in order to implement machine learning effectively. She suggests organizations first try something relatively simple, like an automated email tool. “If you want a quick win in this area, demonstrate its value to your board or decision-makers. Then you can move on to more complex things.”
AI4Giving outlines a handful of ways that nonprofits are currently employing AI in their own operations to improve efficiency and, in particular, engage more effectively with donors. These include:
Donor matching and personalization engines
Donor prediction models and automated stewardship workflow
Online fundraising campaigns
Donor research and data cooperatives
Reporting and workflow tools
Donor Matching and Personalization
Donor matching is a way of applying machine learning to connect donors with nonprofits or causes via data on both the donor and the nonprofit. Fine and Kanter cite a number of applications that are already providing these capabilities. They identify Philanthropy Cloud, which was initially developed by Salesforce to serve its own employees’ desire to support charitable causes through giving or volunteering, as an example of a tool that incorporates donor-matching functions.
Nick Bailey, vice president for innovation and product management at Philanthropy Cloud, sees this type of AI innovation as providing mass customization for donors. For him, the key question is, “How do we connect people with the right kind of organizations?” He sees AI as enabling companies and their employees to become citizen philanthropists “at scale.”
“You can’t hire enough people to handle small and medium donors in the same way you handle large donors,” he said. “You can put systems in place that allow you to provide a similar experience to small and medium donors without having to have staff to do this.”
Nasi Jazayeri, Salesforce’s chief technology and product officer, uses the concept of “engagement ladders” to explain the giving process. He sees this process as changing an advocate into a volunteer, and then into a donor. “Understanding how you move people up the ladder is important.” The idea is to use AI to analyze donor or volunteer data and provide them with the right content and recommendations that will motivate and convert them.
A development officer could set up a repository of fundraising content and then employ natural language processing to construct an appeal that presents the most appropriate text, images or video based on what’s known about the potential donor, Jazayeri said. The ask itself could also be adapted to the donor’s giving history or demographics.
On the philanthropy side, the report found that AI shows promise in helping potential donors understand which organizations they should support to maximize the impact of their giving.
“We’re getting increasingly accustomed to having tailored recommendations, like with Netflix,” said Rhodri Davis, who leads Giving Thought, an in-house think tank at British nonprofit Charities Aid Foundation. “There’s going to come a point at which having to do research on what charities to support is going to seem frustrating and annoying, so that’s the direction we’re going to travel, and tailored charity recommendations will be more the norm.”
Davis cautions that using algorithms to provide giving advice has big and potentially negative implications for the field. “We could just end up with a small number of well-known organizations getting even more funding, along with a huge long tail of organizations that will probably miss out even more.” Charities will need to understand how the process works in order to ensure they are showing up in these sorts of tools, as with search engine optimization, he said.
AI could also enable better and more comprehensive rating of nonprofits, something that donors increasingly value. Michael Thatcher, CEO of Charity Navigator, indicated that the nonprofit rating site is looking at developing “reputational scores” using AI, which could mine social media data to analyze public sentiment toward a group.
Thatcher emphasized that corporate donors in particular care a great deal about reputation, and AI will enable “more subtle and refined evaluation choices.” Charity Navigator’s early explorations into evaluating reputation include conversations with CultureX, which uses machine learning to interpret GlassDoor workplace reviews through sentiment analysis.
Donor Prediction Models and Automated Stewardship Workflow
Fine and Kanter say AI has the potential to improve understanding of donors and prospects, along with managing the donor relationship more efficiently. “Using an algorithmic approach to donor stewardship is more efficient than a human combing through thousands, if not hundreds of thousands, of donor data points to identify the best potential donors to focus a fundraiser’s limited time on,” the authors write. “The algorithm is much faster than a human at discerning patterns in a large ocean of donor data, and unlike a human, does not get interrupted by other tasks or overwhelmed.”
Georges Smine, vice president of product marketing for Philanthropy Cloud, suggests that the use of AI in evaluating donors is “really about better understanding their giving propensity. Doing this more effectively can help a fundraiser maximize his or her engagement with a prospective donor.”
As another area with immediate potential, the report cites the use of AI to automate donor stewardship tasks, like routine communications. The authors highlight an email automation app, Gravyty, as a tool for automating tasks like donor management so frontline fundraisers can focus on higher-value donor interactions. Kevin Leahy at Gravyty refers to these sorts of tools as creating “the illusion of personalization,” but stresses that this is “not a bad thing.”
Adam Martel, Gravyty’s CEO, similarly emphasizes that the company’s intent is not to replace people. However, Martel noted that as a result of COVID-19-driven revenue shortfalls, many organizations have reduced staff across the board, including in development. In his view, that makes it critical for organizations to maximize efficiency, and AI could help.
Martel suggests that if an organization is forced to lay off someone who was managing 200 donors, those relationships are no longer being managed, and “technology has to step in at some point.” He also notes that AI tools could allow a fundraiser to manage more relationships, so a lower-level donor could receive the kind of personalization typically reserved for more generous givers, and thereby maximize giving potential.
Online Fundraising Campaigns
AI technology also manifests in tools, such as automated chatbots, that interact with donors and assist in online fundraising campaigns. Fine and Kanter suggest that such tools “not only help improve the retention and conversion rates for everyday donors by personalizing messaging via email, social media and landing pages, but also bring in new everyday donors.”
Engaging online donors is critical these days, and AI can help an online campaign tailor its messages. A recent report by Classy states that “organizations must capture visitors’ attention as soon as they land, either to immediately convert or to keep learning more to eventually convert.” The longer visitors stay on the site, the more likely they are to give.
AI can also help fundraisers optimize online campaigns by improving the timing of their efforts to get a donor’s attention. Bill Rand, a business professor at North Carolina State University, points to an American Red Cross “text to donate” initiative he helped develop as an example of how AI can optimize messaging.
“We built a model to try to predict when people would be most likely to donate. How do you make the decision to push a message out? For example, when raising funds for Hurricane Irene relief, how can you best measure when interest reaches a high level?”
Rand studied the relationship between social media activity about the hurricane and the text-message donations the Red Cross received during those hurricanes, and determined that tweets referencing hurricanes were a good prediction of the level of donations the Red Cross would receive. According to Rand, “this is because the primary driver of donations to the Red Cross and tweets about hurricanes is the same, namely, the level of the population’s awareness of and interest in the disaster.”
Donor Research and Data Cooperatives
In order for AI to work well, DuBravac stresses the need for “deep data” in applying machine-learning tools. “Data is the fuel for artificial intelligence algorithms, so the more data you capture, the more ability your fundraiser has to understand what drives additional giving.”
AI can help broaden an organization’s base and generate deeper engagement from those who are already committed to your cause by identifying patterns and helping fundraisers understand who’s already giving. “If you look at philanthropy, a lot of costs are around giving campaigns. That’s a big challenge for philanthropies, since they want to spend as efficiently as they can—data can help them understand who gave and why they gave, and recognize patterns in giving in order to target new givers.” AI provides tools for “deciphering all this information you’re sitting on, where you don’t know exactly what you have.”
AI4Giving cites Giving Tuesday’s Data Collaborative as one effort to create larger data sets on giving and its drivers and impacts. Woodrow Rosenbaum, the group’s chief data officer, said, “We built our data commons to help nonprofits, and we consider our data assets to be essentially a public utility.”
Fine and Kanter note that Giving Tuesday gathers data on donations from payment processors, giving platforms, 990 data, workplace giving, and social media. “It became apparent that there was an opportunity to learn much more about giving—the drivers behind it, the behaviors around it, and what might inspire more of it—not just on Giving Tuesday, but year-round,” they report. For example, by analyzing more than 600,000 tweets, GivingTuesday identified the top 50 hashtags that were tweeted along with #GivingTuesday, and was able to determine which of those words produced more social engagement.
GivingTuesday used natural language processing to analyze a database of Form 990 submissions and automatically mapped their National Taxonomy of Exempt Entities (NTEE) codes to the 17 United Nations Sustainable Development Goals (SDGs), data used by funders and others to assign funding and measure impact.
Paul Domanico, senior director of Global Health Sciences at the Clinton Health Access Initiative (CHAI), offered another example of this type of data analysis. He’s working with North Carolina State University on a project to support his organization’s fundraising efforts by doing a deep dive into the web to identify donors with the potential to grant more than $5 million to the organization. Domanico said that CHAI used a dedicated server to scan millions of web pages to understand the giving behavior of potential donors. Domanico said that CHAI “found that this is a better way to do homework on what a donor is focused on.”
Domanico said this process identified and prioritized 200 major donors with which CHAI hadn’t previously worked. The organization then crafted a strategy to reach out to eight donors whose current giving behaviors were aligned to CHAI’s values and interests, beginning a number of new conversations about potential funding.
Reporting and Workflow Tools
Fine and Kanter see AI as increasing efficiency in a number of ways, such as managing a large number of donor-funded projects. High among the benefits they identify in implementing AI tools is its capacity to take on the more rote or routine parts of fundraising, thus freeing fundraisers from those tasks and allowing them to focus on high-value interaction with donors.
Rebecca Widom, director of data science and analytics for DonorsChoose, cited her group’s ability to classify requests from teachers using natural language processing. “An example of how that’s helpful involves requests for music resources. We don’t get a lot of those, so we want to make it easy for donors to find those projects.”
DonorsChoose also uses AI tools such as Amazon Comprehend to find replacements for items that might not be in stock as the group fulfills donor-funded orders. “Natural language processing allows us to do this and makes it possible for us to approve orders involving replacements automatically.”
Future of AI Adoption
While the authors are optimistic about the future of AI in the nonprofit sector, they see a number of obstacles to its adoption, and the report urges caution on a number of these challenges. For one thing, they warn nonprofits not to use AI tools to scale up bad practices, such as what they characterize as the all-to-common transactional nature of many donor interactions. Instead, they recommend using it to shift more interactions from transactional to relational.
In addition, because AI algorithms frequently use proxies, or substitute variables, to uncover patterns that inform their decisions, unintended biases could emerge. As an example of “proxy discrimination,” the Brookings Institution cites the use of data about whether a person uses a Mac or a PC—a factor that is also correlated to race—to determine the likelihood of people paying back loans.
The authors note potential ethical issues along these lines, should AI be adopted more extensively in the sector. For example, they ask whether algorithms used in AI tools contain or exacerbate any existing biases. As noted above, the ability of larger organizations to implement AI, along with a bias favoring better-known causes, could increase the share of giving captured by large, well-known groups. The authors provide a detailed ethics checklist in their report to map out issues and principles to address them.
Despite such concerns, it seems likely that AI tools will become increasingly common in the nonprofit sector, especially in fundraising. The Brookings Institution’s West noted, “COVID is accelerating many tech trends, and the same logic applies to the nonprofit sector. Charities must automate administrative processes to become more efficient.”
The authors of AI4Giving agree. “The virus has been a more powerful catalyst for increasing giving and innovation than anyone could have ever imagined. In a post-COVID world, AI will no doubt help to sustain the spread of giving and generosity.”