4 Actionable Lessons on Artificial Intelligence for Nonprofits

If there is just one thing we can all be certain of when it comes to artificial intelligence (AI), it’s that it is constantly evolving.

For nonprofits overwhelmed by all the tasks on their to-do lists, the amount of information on AI can be intimidating. But the technology also has the potential to alleviate some of that stress.

Here are four insights for nonprofit professionals looking into how AI can support their missions, drawn from sessions at BridgeTECH, an event co-presented by NonProfit PRO, the Direct Marketing Association of Washington (DMAW), and the Association of Fundraising Professionals Washington DC Metro Area Chapter (AFP DC).

1. Use Nonprofit-Specific Tools

With how many AI systems are out there, it can be challenging to find what works for your nonprofit. While it might be tempting to lean on the generic, well-known options out there — such as ChatGPT — these tools require finetuning to work precisely how your organization needs them to.

According to Tareq Alani, co-founder of Chorus AI, AI tools are increasingly moving away from prompt engineering and toward context engineering. That means with generic tools, nonprofits would have to set up that context manually, which is a more time-consuming process than crafting a prompt.

“We're bringing in your own documents and data, and then we can also integrate with your marketing platforms or CRMs, right?” Alani said during the BridgeTECH session, “AI Real Talk: What Your Donors and Competitors Are Saying (and What They’re Actually Doing).” “So being able to have all that context in one place, and being able to find it with AI search without having to have an expert CTO and engineer and data scientist on your team — which most of us don't have — is really critical. It's all about the context. Sure, you can technically do that with ChatGPT, but then you're ending up spending tens or hundreds of thousands of dollars to build those integrations.”

By using an industry-specific AI tool, you can not only ensure the output more accurately meets your goals, but you’ll be able to make time for what matters most.

One example of this in action is that, with an industry-specific AI tool in its toolbox, the Progressive Multiplier Fund has been able to streamline data analysis to create in-depth case studies.

“This understands what we're doing, and I was able to see how it took the 20-plus documents that I had added in and summarized it in a way that makes sense and is very cohesive,” Valeria Sosa Garnica, senior director of learning and impact at Progressive Multiplier Fund, said.

2. Leave Human Expression to the Humans

Of course, the ethical decisions behind using AI are not lost on those in the nonprofit sector. Questions of when it’s OK to use AI and at what level are common.

Marc Ruben, partner at M+R, explained that his company is exploring the capabilities of AI, but insists that keeping humans in the loop is an indispensable part of the process to help prevent bias, keep data secure and more.

“Generally speaking, because there's so many gray areas in terms of where is the human in the loop, we feel like the closer we are to human expression — human expression of emotion, of individual story, an archetype, an example from a program — the more we're dealing with that kind of stuff, the more we need human involvement,” Ruben said during the session “AI Real Talk: What Your Donors and Competitors Are Saying (and What They’re Actually Doing).”

3. Prioritize Data Quality

Nonprofits often struggle to maintain clean data. Maybe it’s because they don’t have consistent formatting standards, or perhaps there’s been turnover in key roles — with each staffer tracking donors differently. But without clean, structured data, AI can’t analyze donor behavior accurately or produce reliable predictions.

“Data is definitely never the sexy part of modeling,” Shane Marple, vice president of audiences and data analytics at SimioCloud, said during the session, “Powering Your Audiences With Extensive Data and AI Modeling.”

Most nonprofits store their data across multiple platforms — email, CRM, donation systems — with no consistent donor ID or unified field structure. And if they want to get the most out of any AI tool, they will need to unify that data.

“A part of AI usage is doing that process as well,” Marple said. “It's not just building the model, but it's combining the data so that you attribute all the right activity and traits to a person at the individual level, and that you have a complete, whole and accurate picture of the program.”

He warns that even the best AI model can’t overcome bad inputs.

“The model could be as great as humanly possible, but if it’s being framed on dirty or incorrect data, then your outcome … is not going to be accurate.”

4. Start Small and Test Incrementally

Adopting AI doesn't have to be a massive, high-risk leap. In fact, Tim Maxton, senior director of annual giving at Dana-Farber Cancer Institute, recommends a methodical, evidence-driven approach.

Dana-Farber's team began with its most familiar tools — recency, frequency, monetary value and internally developed regression models — before bringing in machine learning–based predictive modeling, Maxton shared in “Powering Your Audiences With Extensive Data and AI Modeling.” When introducing externally built models, they followed a phased testing strategy.

“For a full fiscal quarter — three months — we did half of our selects based on the traditional RFM and half [using] our renewal and sustainer models,” Maxton said. “And it was conclusive and consistent and significant: being modeled performed better for us.”

Importantly, Maxton noted the evaluation didn’t rely solely on assumptions or enthusiasm for new technology. Instead, his team sought hard evidence that the new approach would improve outcomes before scaling up.

“We have a 27% increase in revenue per active donor,” Maxton said. “That's by virtue of increased average gift and an increase in the average number of gifts per year per donor.”

His takeaway for other nonprofits? Don’t wait to have a massive AI infrastructure. Start with a contained, measurable test. Let the results speak for themselves before you scale.

“Caution is not a bad thing,” Maxton said.