In this episode, host Stephen Goldsmith talks with Oliver Wise, Acting Under Secretary for Economic Affairs and Chief Data Officer at the US Department of Commerce. Drawing on his time in New Orleans City Hall and with the federal government, Wise shares practical insights on using data to drive results — and how cities can harness the power of generative AI without waiting for perfection. From "use case truffle pigs" to building AI-ready data systems, this episode explores how public leaders can unlock smarter governance through better data practices.

Listen here, or wherever you get your podcasts. The following is a transcript of their conversation.

This is Stephen Goldsmith, professor of Urban Policy at the Bloomberg Center for Cities at Harvard University with another episode of our podcast. I'm delighted to have Oliver Wise as our guest today. I guess, Oliver, I could say I've watched you grow up in the data movement, maybe not quite that literally, but beginning as the insightful Chief Data Officer and more for Mitch Landrieu in New Orleans and now Acting Under Secretary for Economic Affairs, Chief Data Officer at Commerce, where you spent several years in addition to your seven at New York. I know you as one of the country's most creative and interesting data officers: welcome to you, Oliver Wise.

Thank you, Mayor Goldsmith, the admiration is mutual.

Give us a minute or two about, how did you end up in New Orleans? I can figure out how you got to Washington after New Orleans. But how'd you get there and what started you?

It was a plane. It was the Southwest plane. [laughter] I was actually just in New Orleans this past weekend. I had a great time there, but I came to New Orleans in the post-Katrina years in ‘08. I worked in New York City before that as really, kind of my first professional job working with a guy, Chuck Brecher at the Citizens' Budget Commission. Maybe you know him from your New York days?

But he was part of a generation of policy makers who helped New York City steer the ship during the seventies fiscal crisis. And I was very inspired by that public service and that comradery that came with that. And when Katrina hit New Orleans, I was in graduate school. I was working at the same time, and it was like, “I really want to be part of that.” So I moved to New Orleans as soon as I graduated from grad school and was able to work initially for the Rand Corporation and then Mayor Landrieu was elected, and I joined his transition team and then went on to a permanent role in his administration. And what an exciting time to do public service!

Indeed. Well, I have a lot of questions for you about your evolution in data and analytics and generative AI, you know, where you have been active, but the most important question I have for you -- before we begin -- is what is a truffle pig and why do you care?

[laughter] So in New Orleans, here's how it came about. We had a really great team of data scientists. We had four, six at our high watermark, I think an average of about four of them were figuring out how to use R and Python and machine learning to help solve problems, especially help address operational problems like getting to the nearest house that needed a smoke alarm first or optimizing our ambulance program. But the use cases, like how to apply that intel, is really the secret sauce to a successful use case. You can do all the cool data science in your office, but in order for it to have impact, you really got to get the use case from the people who are in operations. And we found that those use cases don't just come from the mayor or the senior most person. Indeed, really where the magic is, is those people who know where the problems are but don't yet know the medicine of how to address that problem.

So we call those people who could come up with these gold use cases that address problems that were high leverage, that were doable, and that also help the enterprise grow their data analytics maturity over time: we called them “use case truffle pigs” because they were able to snort out this kind of difficulty finding tuber that's underground. And if you can get a truffle pig to snort those out, you can make really good money. So that was our analogy to those use case truffle pigs. And then we designed a training program and a workshop to help those truffle pigs identify those use cases.

It's one of the most interesting phrases I've heard in our many years of talking about data analytics. So let's stick with your truffle pig just for a second. So, the truffle pig depends on the readiness of data, and particularly now when we think about generative AI. So, I want to ask you a number of questions about generative AI, but before we get there, how do you think about data preparation either at the city or the federal level? What steps need to be taken to unleash the power of these insights?

Yeah, well, I think actually the lesson from cities is really helpful here in the federal experience. I'd say you’ve got to start with the problem. You can't boil the ocean, right? You go to any CDO conference now, public or private sector oriented, and AI readiness is all the rage. But I think Gartner has a good definition for a good way of thinking about AI readiness: your data is ready if you can use AI effectively to address your use case. So what you’ve got to do is start with the problem you're trying to solve and then ensure that your data is in a state that it can be useful. There are some use cases like a chat bot to analyze and make much more ready all of your organization's policies like your administrative policies or your executive orders. If you're in the federal level, you're thinking if you're in the local level, it's like whatever your analog is to that, that's a great use case because actually there's not a huge amount of data preparation to do that data.

Those policies are all text, they're all fully public and they're probably ready to go. And you can point an AI at that website, and as long as that website is well governed and that AI is drawing on the most recent policies, you're probably good to go. And in those cases, you are AI ready and you shouldn't let AI readiness be an excuse to not get going. But on the other side, and the use case that we've been spending a lot of time at the Department of Commerce is, if you're trying to use generative AI to actually analyze data -- and at Commerce we're major purveyors of public data; we have the Census Bureau here, we have NOAA, which has the National Weather Service, we have the Bureau of Economic Analysis, which publishes GDP and other major economic indicators. And so we serve the data analysis public all the time.

And when generative AI really hit the scenes going on two and a half years ago, we really became interested in how do we ensure that we are publishing our data in a way that allows our users to leverage these new technologies effectively and reliably. And there are much more involved standards in preparing your data. It boils down to -- and I can go into depth on this if you like -- but you have to ensure that your data is not just machine readable. That's been the paradigm of the last 10 or 15 years. So it's a necessary but not sufficient condition for that data to be able to be automatically extracted and possible by a machine. But you have to attach the appropriate metadata and you have to follow the relevant data standards to ensure that the machine or the AI can properly understand and interpret that data so that when they're going after the data, when you give the AI a prompt, it can formulate the appropriate query strategy and then go out and find the data, bring it into its own environment, and interpret that data appropriately.

So a lot there. Let's think about in the following way. Years ago I wrote about you and blight stat in New Orleans, I don't know, somewhere between 10 years and a century ago. I can't remember exactly when it was. And we've been looking…now this goes to your last comment at how generative AI actually renders obsolete almost the previous approach to stat and performance measurement. That is to say, we should be able to iterate more quickly. There should be more truffle pigs located in the organization because they're able to communicate with the data more easily. So how do you think about -- we'll get to the second about the accountability of generative AI -- but how do you think about the power of generative AI in the application of data insights more generally throughout an organization?

Yeah, I think it's a very exciting time, and I think we're only beginning to crack the nut on this one. When I look back at our time in City Hall, it took us a while to…what were we doing? And if your neighbor or the person who showed up to your meetings wants to understand what you were all about, it's helpful to have a phrase to convey that. And our tagline, well, “we use data to set goals, track performance, and get results.” And I think that framework is actually helpful in thinking about how GenAI can be useful within a stat shop. One is setting appropriate goals, and that's really important and that's really the executive's responsibility, of course, with the technical assistance and the analysis support of your stat team. But the city can certainly help the executive craft the appropriate goals by analyzing that public leaders policy documents, their campaign speeches by soliciting appropriate public feedback and then translating those aspirations into actually achievable measurable goals.

So I think there's exciting applications there. Then on the ‘track performance’ category, I mean that's where there's huge opportunity to unleash AI to better be able to track the appropriate metrics. I mean, there's just huge potential there. And then the ‘getting results’ part was always the fun part, and that's the secret sauce, the magic, which is what actually happens in the meetings. You have the executive there who's set the ambitions, you have the data people there who are tracking the performance, but those creative ideas and solutioning is what makes those meetings exciting, and that's where you need the insights, the operational know-how of the real on the ground practitioners there and having GenAI help those users come up with new, novel solutions to old problems is a really exciting opportunity.

You've obviously been at multiple levels of government. Do you see cities using much of the data that is made available by the federal government? How do you think about your Commerce data being available for economic development, or are we utilizing the ability to move data up and down among different levels of government?

Oh, that's a good question. I mean, yes, local governments and cities use federal data all the time, often without even knowing it. If you are locating, if you're doing any geospatial analysis at all, you are applying, you're looking at some phenomenon, like number of permit applications, number of arrests, number of 311 calls, whatever within a boundary. That boundary is generally created and authorized and governed by the federal government, whether that's a census tract or otherwise. So you probably use it all the time if you're doing any sort of normalization by per capita or comparing your city to data in another city where you have to normalize by per capita or some other denominator, you're using federal data. So it's often the foundational piece, but federal data doesn't tend to come out all that frequently. The more low latency data is going to come from a city's administrative sources or from private sector data. And the federal government, especially the census, is already moving quite aggressively in a direction of their data stack, which began a hundred years ago as 100% survey base. It's now much more blended where it's incorporating data from other federal agencies like IRS or Social Security, also state and locals, and also the private sector. And that's allowing for more frequent, more granular data products, which I think will be more useful for state and local governments. I think there's a lot of exciting opportunities here.

Just a couple more questions. What have you done at the federal level in terms of data accountability, algorithmic bias, inter-agency transfer of data? What are two or three things that you've learned from your complex work at the federal level that our listeners, who are disproportionately state and local, would benefit from?

I'm particularly proud and excited of the work we've done over the last 18 months or so on one: kicking off a public conversation and then also doing some real technical work on figuring out how do we, as public data stewards, publish our data in new and novel ways to meet the user expectations in the mid-21st century. And we're doing a lot of experimentation at NOAA and census in particular, and also BEA. So we have a report that we published in January that lays out guidelines in best practices for publishing public data to be more LLM friendly. We're now in the pilot phase for that. NSF has latched on as a really important partner in this and funding better statistical data, AI readiness across the federal government. But I think this is a space where federal and state and locals can really collaborate more on ensuring that all that great work that was done to make data open, that we're continually evolving those practices so that we can make best use of that data and that users can use modern tools to leverage that data.

Just going back to my time in City hall, we spent a lot of time and effort making sure that the blight data that we were analyzing for operational purposes was also made public, but the public just didn't have the capacity -- and probably doesn't still -- to actually download that data and analyze it and set up some Tableau dashboard, they're not going to do that. But the power of, if you make that data available and with just a bit more, it's not a huge leap to get from where you are already to a place where it's truly AI ready. You can then enable the public user to just use natural language to get answers to that question. And I think that presents a sea change opportunity to promote transparency and engagement with the public.

Yeah, we have an exciting project with a Knight Foundation to look at how generative AI will allow better insights to be shared between stat programs and community groups. Think about registered community groups, if they could use natural language to access the open data, it ought to improve the quality of the conversation like causal and identifying other remedies.

I think what local governments are really good at is experimentation and just doing the application work of generative AI and leave it up to the feds for all the standard setting and all that stuff. I’ve got to tell you, it's very gratifying to be in federal government. I'm very thankful for this opportunity, but it's way more fun and you can solve a whole lot more very discreet, satisfying problems in local government. So, read our report, but get started on solving problems! And I think it's really exciting to see leaders, especially in Boston, but also elsewhere, get going with very minimal and very practical generative AI policies and show value, get the flywheel going, and start getting value from these new technologies in new and novel ways. And I think that local governments, especially cities, I'm so bullish on their potential to do that. Cities lead the way in that space.

This is Steve Goldsmith, professor of urban policy at Harvard talking to Oliver Wise, one of the country's most sophisticated folks in terms of data and analytics saying to us today: follow the truffle pig and have fun, and you’ll accomplish something at the local level. Not the final bit of advice I anticipated, but Oliver, as usual, thank you so much for your time.

Thank you so much, Mayor.