A conversation with Pacific Community Ventures on building AI for fair lending

Pacific Community Ventures (PCV) is a community development financial institution (CDFI) based in the Bay Area, a mission-driven lender that backs small business owners overlooked by traditional banks to create good jobs, with research and data work that reaches nationally. PCV uses Claude across that work, from a voice survey tool that gathers candid feedback from workers to an underwriting co-pilot in early development. Chief Data Officer Sachi Shenoy and CEO and President Bulbul Gupta sat down with Anthropic in June 2026 to talk about why a community lender is building its own AI rather than adopting the credit risk models built for traditional finance, the national data commons they are starting, and the mission that drives the work. 

Anthropic: As a community lender, how is AI changing the way you approach your work?

Sachi Shenoy, Pacific Community Ventures: It is accelerating our work in almost every way, shape, and form. Lending is at the core of what we do, and we've been developing a predictive financial model with AI to augment our underwriters' decisions, so we can get capital to the small business owners who need it faster. It’s also supporting our impact measurement and management work.

A lot of the work you’ve built with Claude relies on human judgment. How do you think about what stays with your team? 

Shenoy: With lending, Claude comes in on the qualitative side. A lot of applicants have sparse financials and no credit history. The thing underwriters rely on most is what they learn in conversation. We use Claude to pull all of the qualitative signals out of an application and quantify them so it can feed the model. The financial model augments our underwriters' decisions. It never replaces the human. These are decisions about a person, not a credit score, and that judgment has to stay with our underwriters.

Gupta: On the research side, Claude helps us reach many more people, but the analysis is still done by humans on our team. Reaching more people is one thing. Deciding what their feedback actually means is the part that stays with us.

You've been deliberate about not simply adopting the financial models built for traditional finance. Why?

Bulbul Gupta, Pacific Community Ventures: AI is coming into more and more financial decision-making. It already exists in traditional finance in investment banking and banking. CDFIs were intentionally created as an industry about 35 years ago to meet low- and moderate-income entrepreneurs and communities where banks and traditional financial institutions have historically excluded, or don't do small enough checks to reach them. So we do not want to bring in a financial model based on what is considered successful in traditional finance. It would disrupt the whole point of the existence of our field. Our whole journey these last few years has been figuring out how to train a financial model on our own inclusive lending data, intentionally designed to serve low- and moderate-income, underserved communities and entrepreneurs. It is designed for us, by us, with us. Our values, our mission.

That financial model is trained only on your own historical lending data. You're now extending the idea beyond PCV with a data commons, a shared pool of lending data from CDFIs across the country. How does it work?

Shenoy: Our financial model uses a machine learning classifier trained only on data points from PCV's lending history. We're also helping a cohort of other CDFIs build customized financial models, each trained only on their own historical data. The larger effort is the data commons, something we launched in June 2026. We invited CDFIs across the country to contribute de-identified historical data, bring it all into the commons so we get adequate data volume, and then we'll train a much larger predictive financial model on the amalgamation of all those data sets.

Underneath the lending and the research is the idea of a good job. What does that mean, and why small businesses?

Gupta: Our affordable capital and our mentorship are for small business owners who are largely lower and moderate income, so that they and their workers have economic mobility through the power of a good quality job. Half of working Americans are employed at small businesses, and they are the hardest to reach. They don't have big teams or HR resources, and a lot of economic mobility work is focused on medium and larger companies. How do we complement what they don't have access to and give them a feedback loop, with their consent and their workers' consent? Then, knowing what's working and not working, we shape the ecosystem that shows up for small business owners: capital providers, technical assistance providers, policy makers, other funders. We want that amplifier effect. It is meant to shape an ecosystem with research-driven insights.

Shenoy: For over a decade, PCV has been a leader in our field on worker voice and on what it means when we say good jobs—really getting at the definitions of a good job and the policy implications. Those are the audiences we have in mind when we put out this kind of work.

You take this work beyond PCV, too. Who else are you building for?

Shenoy: The Radiant Data Hub is one of the programs I oversee at PCV, and its mission is to augment the data and AI capabilities of other mission-driven investors. We focus on peer CDFIs, but if others are doing similar, mission-aligned work, we are open to working with them as well. A recent example is an organization called Sheconomy. The founder, Janine, wanted to model what today's world would look like if, back in 1925, women had earned just as much as men and been just as represented in the economy and politics. We used Claude to help tie historical economics research together into the data-backed headlines behind their national campaign.

What's the larger goal behind a campaign like that?

Gupta: The activation is getting people who have never seen themselves as an investor to reimagine what the world could look like if they used the economic power they already have. The work Janine does, and that several of us do in our own ways, is to activate women to see themselves as investors. In 2017, for the first time in U.S. history, women owned 51% of the assets in this country. With the $53 trillion wealth transfer happening from boomers to Gen X and millennials, women control more and more. Janine runs a national network called Invest for Better, and there are so many other circles of women investing in women. 

What's next on your list?

Shenoy: One thing I'd love to build with Claude is our own internal knowledge base. Like any nonprofit, we're sitting on so much information, and much of it in our staff's heads in different shapes and forms. I'd love to wrap that all up into something rich enough that everyone at PCV can self-serve when they need a historical data point or a stat for a report. I know we can get there.

Gupta: The bigger thing we are building toward is a national financial model designed for underserved entrepreneurs and communities, built from the data of CDFIs across the country rather than borrowed from anyone else.