New Hampshire’s benefits program embraces AI amid modernization push

NEW YORK CITY — As numerous states struggled in the face of massive unemployment insurance fraud at the start of the COVID-19 pandemic, New Hampshire had something of a head start.

New Hampshire Employment Security had modernized many of its IT systems in 2011, allowing residents to file claims electronically and also upgrading its infrastructure to cope with fraudulent claims. So while other states lost billions of dollars in fraud during the pandemic, New Hampshire estimated it lost just $500,000 on just 169 claims that were paid out.

That time was still a struggle for the agency, however. More than 200,000 people out of a state of 1.4 million were receiving unemployment benefits during the pandemic. That office also runs job training programs and job fairs to connect Granite Staters to employment opportunities.

Now, the onset of artificial intelligence has state leaders thinking about the future, and the possibilities that the technology offers in expediting benefit payouts, streamlining employee workflows and enabling greater efficiency. The prospect has Deputy Commissioner of Employment Security Richard Lavers excited for the future in his state.

Route Fifty caught up with Lavers last week at the Google Public Sector, GenAI Live & Labs event in New York City to hear more.

This interview has been edited for brevity and clarity.

Route Fifty: Let's talk about the past before we talk about the present. What was it like determining benefits? Are we talking paper and all sorts of old coding languages that no one knows anymore? What did it used to look like for you guys?

Rich Lavers: New Hampshire was fortunate. Prior to the pandemic, we had modernized our system, so folks weren't on paper. Since 2011, folks have been able to file from wherever they want on a web-based system, so you're using your mobile device, tablet or PC to file that claim, or you're able to come into our offices and utilize free computers that we have for folks to come in and file that unemployment claim. It's all electronic, with some paper options for those people that don't want to be electronically filing, but when they come into the office, we try to hold their hand and get them to file electronically rather than on paper. It's a modernized system that we've been able to continuously enhance in the years prior to the pandemic, and it's a system that we're able to leverage without having to do a wholesale change for all the programs that got rolled out during the pandemic.

What people not involved in the unemployment program don't realize is how much of a sea change the unemployment programs had during the pandemic. Eligibility got broadened way out beyond what it traditionally is, the amount of benefits got increased significantly, and the utilization was like nothing we have ever seen in the history of the unemployment program, back to the 1930s when it was created. We're fortunate we have a modernized system that has a good set of bones and that we're able to continue to enhance. Looking at some of the ways AI can drive performance, we're able to do that on the backbone of the application that we are currently using.

Route Fifty: It's interesting that you guys modernized a long time ago compared to a lot of other places. What was behind that?

Lavers: I can take only so much credit for it, because those decisions were really made in that 2007 time before the Great Recession. But there was a recognition of the need to transform the system. I think a lot of that was being driven by demands from the public who were not happy with the way in which they were currently filing, as it felt like an antiquated process. We launched a new system in 2009, right when unemployment was spiking during the Great Recession, so it was a very challenging time for the agency to try to roll that out, and so it had its fits and starts. At the beginning, it was a system that was rolled out through a partnership with Deloitte that they had developed for the state of Ohio. They were able to customize it and make it workable and usable for New Hampshire.

What we've done since then is we've taken advantage of every opportunity we can, with both funding opportunities through the federal government and the state, to enhance that system. Now we have something that continues to be improved. We continue to improve the experience for customers and for our own staff and how they go about their daily activities. That's the challenge now: constantly improving it so we don't sit back and just become complacent with what we've done.

Route Fifty: What have some of those enhancements looked like?

Lavers: This is all through some simple conversations that we had initially with the Google Rapid Innovation Team. We didn't know much about Google, but they knew a lot about the unemployment program, and that's what impressed us the most. Within their public sector group, they had resources from the public sector, from actual programs, from people that had run unemployment programs during the pandemic. They came in, and it was a simple conversation. Nothing was being pushed on us.

Those conversations evolved from looking at some pain points, and that conversation was being driven by my own subject matter people but equally being driven by the subject matter people that Google had within the public sector group. We were initially able to identify a great project, which is our Adjudication Assistant Project, which is under development right now. That should be coming up to production within the next three to four months. That was looking at that pain point of gathering information at the front end of someone's unemployment claim. Despite our best efforts to try to have the most efficient set of questions to ask you about your recent job loss, it always feels that it's not about you. It always feels like we're asking the wrong questions. It unfortunately enhances the feeling that people have at that stage where they've gotten that professional punch in the face, they've lost their job, they feel like they're being abandoned by so many people, and now you've got a government program which is critical for you as you look for that next job, and it seems like you're not very important to them.

What we're doing with Gemini AI is, rather than asking all those people to make those dropdown box selections and all these predetermined questions, we're just going to start with a very simple question, and say, “What happened when you lost your job? You can tell us as much as you want or as little as you want.” Then Gemini AI, through the training that we've done and continue to do with our subject matter people — we're training it to know our laws, rules and policies, so it can then ask in real time the correct questions, and hopefully as few of those questions as need to be asked to make sure to pull the information that our seasoned human adjudicators need in order to determine whether or not that job loss was one that qualifies you for unemployment benefits.

We'll use Gemini AI for fact gathering from the individual, we'll use it for fact gathering from the employer, because a lot of times the version of events for a contested job loss — not so much a regular layoff, but a contested job loss — those versions of events are quite different, depending on where you sit. We'll be utilizing that trained Gemini AI to gather that information, and then Gemini will be packaging it up for our human adjudicators, providing them that summary. If they're a skeptic and they don't trust it, they'll be able to look at the source material. They'll be able to look at exactly how the question was asked, the sequence of the questions, the exact answer as it was provided, so that they can make sure that they are in the same swimming lane as where the AI is going with its summary.

We'll give them all the links to law, rule and policy, so taking that most seasoned and best adjudicator and elevating those that aren't quite there yet and bringing them to that same level and be able to package that all up so the adjudicator can then make a decision. It will hopefully improve the amount of time that it takes to issue that decision, because I have never met an individual who feels that they receive their unemployment benefit too quickly. The more time we can shave off there but getting the right information right out of the gate, rather than having to go back to you and ask additional questions, that's the goal.

Once we put this into production, we have a whole list of other projects that we want to do that touch upon all that whole lifecycle of the customer. The one that gets me the most excited is not only the additional AI infusion that we can do into the claim filing process to improve performance there, but also in connecting you with training and job opportunities. There is so much data that we have accessible, but it's hard for staff to be able to have the level of knowledge that they need about all the different training programs that they have available, and the ones that you might be eligible for and might not be eligible for. We’ll use AI for that and use AI to better match you up with those job opportunities, for which we have all the data. Now it's just a matter of utilizing the power of AI to personalize it, connect you with the programs and the employers and job opportunities that are the right fit based upon where you want to go and where you have been in those experiences that you have. I think that's the most exciting piece, because that's the end that we want to get to, is getting folks that unfortunately need the unemployment program, we want to get them back to work as quickly as possible.

Route Fifty: Has training your own employees to use some of these techniques and this technology been tricky? Has there been skepticism, excitement, somewhere in the middle?

Lavers: New Hampshire is very much a privacy hawk state, so when you get into different technologies, in particular AI, you have a lot of skeptics. We've brought a large team of people, both skeptics and non-skeptics, into the very beginning of development with the adjudication assistant. We've brought them in so we can demonstrate to them and show them how much we value them and their expertise, and that I'm not looking to replace them. I'm looking to harness the strengths and the experience that they bring so that we can improve the process. No one wants to be part of an organization known for a bad process or bad product. You want to be proud of the organization that you're a part of, and I want my employees to be proud of that organization. When they're out there and they say they work for the state's unemployment program, what you want to get is someone be like, “Oh yeah, my cousin just had to file a claim, and you guys have really improved that process. They said that they were interacting with AI, the questions were personalized, and they got paid faster than their friend who had been filing in Massachusetts or filing in another state.” That's the type of experience you want, and helping skeptics that work for you understand that that's the end goal, that it's not just replacing those human beings but allowing them to be more efficient in what they're doing and taking away some of those mundane tasks that they don't enjoy doing, and really leveraging the brain power and their experience and trying to gain their trust. I've found that the best way to do that is bring them in right out of the gate.

Route Fifty: There's a lot of talk right now about this idea of government efficiency, and a lot of attention on it at the federal level. What does it mean to you to be an efficient government?

Lavers: I'm passionate about public service, and a line that I use quite a bit is, a lot of the programs that we run, whether it be the unemployment program or Medicaid or [Temporary Assistance for Needy Families] or [Supplemental Nutrition Assistance Program], they're not profitable, but they're critical, and when you need them, they’d better perform well. Bringing efficiency to the public sector is not a negative thing.

What efficiency means to me is accessibility and performance. I need to make sure that the product, the service that I'm delivering from the government side is accessible to everyone where they don't need to know the secret handshake to utilize the program that they're trying to leverage, because they absolutely need it. And I'm trying to improve the performance so that the time they need to spend on getting set up with that program, we're reducing that. We're asking the questions we need to ask, and we're getting them in and out the door as quickly as we possibly can.

Transparency is key as well, particularly on the public sector side. Here’s a real quick story. My boss just decided he was going to get a new car. He drives a hard bargain when it comes to negotiating. He thought he drove a great bargain, got what he wanted, all the features, closing the deal. He's at the dealership, talking with the gentleman who's finishing things up, finalizing the paperwork, and he says to him, “Hey, before I get driving, I really wanted to say thank you to Aaron and Veronica.” And the guy's like, “Well, Aaron's upstairs, but Veronica doesn't exist.” He had a great experience, but because he didn't know he was interacting with AI, the way he tells the story, you could see there was a lack of trust that he had with that organization. From a public sector perspective, the level of transparency can't be high enough, and making sure the public knows when they're interacting with AI so they can help better identify and see the improvement in the process that is derived from AI is key.