Q&A: How a Political Data Giant is Approaching AI

Political professionals and the organizations that they rely on to do their jobs have been scrambling in recent years to integrate artificial intelligence into their work.

For most, that means asking large language models to help analyze data, refine scripts and talking points or, in some cases, generate ad content. But there’s still an ongoing conversation about how far to go when it comes to AI integration. Campaigns & Elections sat down with Sunil  Sadasivan, the vice president of engineering at NGP VAN, to talk about how the Democratic data giant is approaching AI. 

The following interview has been edited for clarity and length. 

C&E: When youjoined NGP VAN last year, your hire was seen by a lot of folks in the ecosystem as kind of this commitment – or recommitment – to building out and improving the political tech stack. Tell me a little bit about how you’re approaching this role, and what are some of your plans?

Sadasivan: It’s been 10 months now. I think I started in June, so it’s both short and long. I started in political tech on Sen. Booker’s presidential campaign, and sort of stayed and did some consultancy in the world. Everywhere I went, VAN ended up being sort of in the lap of every organization. And so I started to see the distribution, and where VAN sat was pretty critical for the progressive and Democratic ecosystem. 

Previously, I was a technology builder myself, and so I always felt comfortable with both good engineering, but then also sort of the transformation of organizations, as well, to meet the moment of innovation. I previously worked at the Department of Veteran Affairs, modernizing some of their technology and everything. I even built a product on top of VAN and I got to see, actually, quite a bit of how users used NGP VAN and the resiliency and trust that a lot of users have with it. That was one of the reasons why I decided to really take a look at the organization. 

The product is huge. There are so many different users. There are so many different ways organizations see it. But at the same time, I got a sense of where AI was going ahead of some of this, and I saw NGP VAN as a core pillar of technology adoption. Organizers and campaign staff come in, they log into NGP VAN, and whatever you can do there shapes a lot of campaign decisions. 

All of this kind of sat with me a little bit. But, you know, talking with [NGP VAN General Manager Chelsea Peterson Thompson], [COO Alex Stanton] and [Senior Vice President Lou Levine], I saw a lot of autonomy and the ability to shape some of my own vision for the organization, so it seemed like the right place in this moment. 

Ten months into this, I’m pretty excited, because I’ve been able to make quite a bit of impact and also help adjust and set the culture for really good, innovative thinking. At the moment, we’ve got a lot in the works, and I love the customers as well.

C&E: We’re seeing more and more campaigns and organizations use AI in some way, shape or form. They’re either experimenting with it or they’re trying to integrate it into their daily work. As far as NGP VAN goes, how are you thinking about the role of AI and what it can do?

Sadasivan: We’re using AI heavily internally. As an engineer, AI has just opened up quite a lot of what we can do. We’re a small engineering team, but the demands on us are heavy, and so we have to find ways to scale our work and be able to innovate fast, especially in a storied organization where the code base is 20 years old. It can be pretty rigid to operate in, but we’re doing a lot here. 

That said, our approach to AI and the products that we’re building – because we are building quite a few AI products – is to, at the end of the day, be deliberate and grounded. We’re not chasing trends for the sake of it. It’s really oriented around: What are our users telling us? What do they need? What are they trying to do and accomplish? At the end of the day, there’s a job to be done. AI is just a tool like any other tool. We’re talking with users, doing research, talking with campaigns and organizers about what they are trying to achieve, and orienting around that anchor point, and then evaluating whether an AI solution is the thing that we should be building. 

From an engineering standpoint, I think there are core principles that we’ve established. Definitely privacy and security is a must, given the type of data that we operate with. We look at governance and our partners under the hood in building some of these tools. And then environmental sustainability, as well. All those things matter to us as an organization. 

But like you mentioned, campaigns are trying to innovate, and that’s why I fell in love with campaigns. They are out-of-the-box thinkers. You have to be, by the nature of it, so we want to empower a lot of that too. We’re also looking at ways to connect your data and NGP VAN to your own AI tools, as well. We’re really excited for you to just build products and workflows that make sense for your organization. It’s really exciting. We’re seeing a really unique thing. 

One thing I want to be clear about is that our AI features that we build internally are opt-in by design, so you won’t be force fed an AI tool. That’s because our organizations and our customers are all across that spectrum of how they’re going to use these tools. 

C&E: I think this is kind of a good time to bring up the Commitment on AI that NGP VAN put out. What prompted NGP VAN to do that, and can you walk me through what it means?

Sadasivan: Kind of going back to that original answer, we think it’s important for our customers to have trust that we’re building AI with rigor and thoughtfulness. We have a pretty strict AI steering committee internally within the organization. So let’s say we’re building an AI product –  take the disclosure inspector that analyzes FEC compliance reports under the hood – there are a couple partners that we work with. That means using cloud models built by Anthropic, an American company that cares a lot about a variety of commitments to training these models ethically, and then hosting on AWS Bedrock, which is a cloud provider that we fully own. We can guarantee that that data is not being sent back to the frontier models for further training and that sort of thing. 

C&E: Are you hearing any apprehension from clients about the direction that the AI is headed here? What has the feedback been like so far?

Sadasivan: It’s varied. We’re hearing from a number of clients, often coming from campaigns that want us to go even further along with AI. And they’ve got many different ideas, which is really exciting. Then we do have customers who are very clear that they don’t want AI features or they have standing commitments within their organization. For us, as tool-builders navigating this, that’s why we are saying all of our AI tools are entirely opt-in. At the end of the day, we will build some of these out for those organizations to choose. 

We do think that there’s a pretty clear place for AI and innovation, and we’re also trying to figure out that line. So we’re going to continue to build some of these tools, but at the end of the day, for the organizations who understandably are not there, we get that and we’re not going to push that on them.

C&E: NGP VAN is, I think, in a really unique position compared to a lot of other entities insofar as you all deal with a tremendous amount of data. How are you thinking about privacy and ethics?

Sadasivan: The first thing to note is that data has been a core tenant of VAN from the beginning. That data is your data and your organization’s. There’s no selling it. There’s no training of that across organizations. So if we ever do offer insights of what’s happening, it’s an entirely independent sort of model that sits just for your organization, and isn’t used to train, broadly writ large. This is why that tenet of privacy is the first thing that goes into how we build these tools. 

The disclosure inspector is a good example. You file, as a compliance team, a report with the FEC every quarter. We’re looking at that report in the context of your own data, taking a look at that against FEC rules…The output of that tool is saying, ‘Hey, you might be flagged with a request for additional information by the FEC on these three new rules that they just put in that you might not be aware of’ – that kind of stuff. But as we build that out, that data is entirely within the context of your organization, and it sits on and uses models that are on AWS Bedrock, which is a way for us to host and know that that data is not being sent to or shared by any of those frontier models. So privacy, absolutely, has been at the core from day one. We’ve got competing campaigns and competing organizations using us, so that trust is really important for us.