GPS Global Peace and Security Peace. Justice. Prosperous Societies.

GPS Interview

Becca Frank

VP of Technology, GiveDirectly

Becca Frank, smiling, with long brown hair, standing in front of green foliage.

Becca is currently the VP of Technology at GiveDirectly, a nonprofit organization focused on using technology and direct cash assistance to help people living in poverty. She has built an impressive career at the intersection of technology, engineering, product development, and social impact. Before GiveDirectly, Becca spent several years at Ketra, where she held engineering leadership roles and helped guide the development of innovative lighting technology. She also worked with Denver Human Services, where she helped launch technology initiatives and played a leadership role in opening isolation and quarantine sites for people experiencing homelessness during COVID-19. Becca holds a degree in Computer Engineering from the University of Michigan and has experience leading teams, developing new products, and using technology to address complex social challenges.

Frank spoke with MPA students at Brown University’s Watson School on October 26, 2026. The conversation has been edited and condensed for clarity.

Q1 US government funding of non-governmental organizations and non-profits has declined significantly in the last couple of years. In terms of technology, do you think the shortage of aid will change the way in which you deliver it?

We were fortunate to have multiple funding streams: significant funding from USAID before the changes, but also from other bilateral funders and individual donors. From people giving small amounts on our website to very wealthy donors giving millions of dollars. So we were … resilient to the USAID changes, which impacted organizations with a narrower focus or funding base more directly than us. What we’re seeing now is a pretty big shift in institutional fundraising. It’s still highly uncertain, but there’s significantly increased interest in individual philanthropy associated with potential AI windfalls and a lot of individual wealth potentially generated by AI. For us as an implementer, it’s a mixed bag. USAID was more stable until it was no longer … a multi-year funding source, but with more strings attached: compliance and specific monitoring and evaluation expectations. That made it much more costly to administer than programs funded by individual donors, who have fewer expectations about the specific implementation. At the same time, individual-donor funding is more capricious. It can grow and scale very quickly, but it’s hard to predict and depends on specific individuals and the market. We try to balance our funding across channels with different natures. Even now, while we’re potentially having a lot of success with individual donors, we’re thinking about which counter-cyclical channels to still invest in for a sustainable path to supporting cash to people experiencing extreme poverty, since traditional aid funding, small-scale individual donors, and large-scale individual donors each have their own cycles and markets.

Q2Beyond helping individual households, how have you seen direct cash transfers stimulate the wider local economy?

The General Equilibrium Study in Kenya looked at cash transfers’ impact on other households, and the multiplier was about 2.4. We do deep research with randomized controlled trials, the gold standard, to answer key learning questions we don’t understand well. But RCTs are very expensive, so once there’s a well-understood evidence base, we focus on scaling delivery in line with it … what we know about spillovers, where that applies in our programs, and where we can scale up. There, we do lighter-weight internal monitoring and evaluation to confirm early-stage indicators align with the larger studies. So we’ve answered this question in several contexts, including Kenya: giving cash to people in a village can positively impact surrounding villages that weren’t paid. The next question is: as we scale from paying a village to paying an entire district, what happens to spillover effects across villages? We’re aiming to answer that in the year ahead with ongoing research.

Q3You have experience both in private companies and nonprofit organizations. What’s something that’s harder about building technology at a nonprofit than at a company, and is there anything that’s actually easier?

In a private-sector tech company, revenue is one very quantifiable, top-line metric of your whole organization’s health, a closed feedback loop. We all looked at that same chart every day as an indicator of our product’s quality. You have clear, data-driven signals about how customers behave and what they’ll pay for. In the nonprofit space, you have to define for yourself how you’ll measure impact. As a nonprofit, you don’t have that single closed-loop metric. What donors fund isn’t necessarily representative of the impact you’re having. Donors certainly care about impact, but it’s decoupled loops. For us in particular, it’s pretty challenging to go to recipients we’ve just paid $1,000 and ask for critical feedback. They’re like, "You just paid me $1,000, what do you want?" Sometimes we get that feedback, but there’s a complicated power dynamic when we’re trying to get candid feedback. Impact is relatively easy to measure, with a very large evidence base. But we still make day-to-day trade-offs between dimensions of impact. We ask if we should spend more on engineering this year to reach more recipients in the future, or give that money to recipients now? There’s no closed-loop system to answer that from the recipient’s perspective, so we have to do that work ourselves and build alignment as an org. When you work on something mission-oriented, everyone on my team is there because they care about the impact for recipients. That rises above a lot of office politics, and even decision-making based on ego or optics. Even when we disagree, we know we’re grappling with real trade-offs about what recipients would prefer, and we all deeply care about the impact.

Q4GiveDirectly uses an AI and machine learning framework to target cash relief. What guardrails prevent bias in the training data? And how is recipients’ data protected when they use AI chatbots to, in your words, ‘navigate family conflicts and ask questions they cannot take to anyone else’?

I want to be clear that we are at a very early stage in developing our chatbots. We use AI and machine learning in a couple of ways, at different stages of maturity, and the chatbots are in early prototyping. … With a prototype, it’s relatively contained: appropriate data retention and secure channels for the data every step of the way. What it means for recipients to ask a chatbot sensitive and hard questions is … what we’re learning in this early prototype and pilot phase. … It’s relatively nascent. We approach a lot of that early-stage work through focus groups with recipients. For example, we tested a bunch of visual ways of explaining our consent language to see what worked best for recipient comprehension. … Focus groups are one of the biggest tools in our toolbox. On using AI and ML to deliver programs efficiently: we usually work with telecom companies that already have cell data about their subscribers. We use it in two ways to identify people to pay. First, geography. In a program in Kinshasa, DRC, a cell tower was in a very, very poor area … if you lived near it, you were very likely experiencing extreme poverty, so we paid everyone who lived near that tower. Second, for areas with very heterogeneous poverty, we use cell phone metadata; … what towers you’re near, how much data you have left, how much you top up, what plan you pay for; … to figure out whether you’re likely experiencing extreme poverty. This isn’t a way subscribers would normally expect their data to be used, so … the key is meaningful, informed consent up front. Recipients opt in - I understand this program and want to participate - and only then do we access their cell data for this purpose. We’re also exploring, though it’s very complicated in practice, doing this fully on the telecom companies’ servers. They’d expose an interface: an API, or application programming interface and keep all individual-level data in their own systems. We’d send only the consent and information from a recipient in the program, and they’d tell us whether that person meets the criteria. That’s how we aim to keep using those datasets, which are tremendous for efficiency, how many people we can reach, and at what cost. … Otherwise, we typically send staff door-to-door to enroll recipients. This lets a whole set of people text us to enroll, get verified, and get paid basically immediately. It unlocks huge scale and efficiency, but has very high upfront costs to do with consent and security, basically all on the data provider’s side … We’re continually working on those relationships and on making it secure.

Q5In an authoritarian context, how do you ensure that the information you’re getting is from real subscribers you’re transferring money to? Or are you only doing this in contexts where there are democratic states and non-state-controlled media and cell companies? Out of curiosity.

We’re largely only doing this in contexts where we’re confident the telecom operator is an independent entity. But even our in-person models get complicated … some of our biggest instances of fraud have been collusion among mobile money agents registering recipients for phones, so at a small scale that’s also challenging. … At the macro scale, we need some basis of trust in the telecom partnership to do this.

Q6What are you most excited about in terms of technology for the future, in the development of AI?

We’re at an inflection point where, with AI, engineers have much more leverage to build things tailored to an organization’s specific needs. At GiveDirectly, we’re rebuilding a bunch of our tech on in-house platforms rather than commercial off-the-shelf solutions. We made that decision before some of the AI advancements; partly, it reflects our scale, but AI has accelerated it. I’m excited because it enables efficiency and scale: we can design for our programs’ unique needs. It takes a lot of work and upfront cost … but, for example, today a field manager spends a lot of time assigning field officers to houses and villages each day and tracking their metrics. At scale, we can make that super streamlined. That may be more specific to our implementation than you were hoping for … but these are the challenges I find super fun about putting this into practice. Like, field officers just rode motorcycles for 2 hours to a remote village in the hills of Uganda, and now they need to sync their phone data to our central server and are running into issues. Those in-the-weeds scale challenges are part of what makes my job so fun, and I’m excited for our road ahead in building the technology that will serve them.

← All interviews