Small Business and AI: Understanding Adoption and Building Support Systems

Kevin Ortiz:

All right, good morning and welcome to today’s Emerging Tech Economic Research Event, or “EERN” for short. Our event today is on small business and AI, understanding adoption and building support systems. My name is Kevin Ortiz and I’m a co-lead for EERN here at the San Francisco Fed where we study the economic implications of emerging technologies such as artificial intelligence. And we’re so thrilled to welcome you all here today. We at the San Francisco Fed engage regularly with businesses and communities to gather real time insights on local economic conditions. These perspectives complement the data and analysis that inform the Federal Reserve’s monetary policy decisions. Today’s conversations will explore the implications of AI for small businesses. We’ll start off with a presentation from Natalie Holmes, senior researcher in our Community Engagement and Analysis group at the San Francisco Fed. She’ll be sharing her findings from the 2025 Small Business Credit Survey.

Then we’ll hear a panel discussion with small business lending and technical assistance practitioners at the forefront of AI driven changes in the small business ecosystem. We hope you gain valuable insights into how small businesses are looking at AI and what this signifies for the broader small business community. Of course, if you’ve ever been to a Fed event before, I do need to give the standard disclaimer that these views that you will hear today are those of our speakers and do not necessarily represent the views of the Federal Reserve Bank of San Francisco or the Federal Reserve System. And so with that out of the way, let’s begin. Please join me in welcoming Natalie Holmes.

Natalie Holmes:

Thanks so much, Kevin. Thanks everyone for joining us today. As Kevin said, my name’s Natalie. I’m a researcher on our Community Engagement and Analysis team here at the San Francisco Fed. So I’m really excited today to share with some forthcoming research with one of my colleagues, Rocio Sanchez-Moyano, who you’ll hear from in about 20 minutes. And I’ve just got two quick acknowledgements before we start. First, there’s a whole production team here that makes these events happen, wouldn’t be here without them. So to thank everyone here offscreen involved in making this happen. And then by way of segue, I have to thank our colleagues at the Federal Reserve Bank of Cleveland where the Small Business Credit Survey (SBCS) is housed and operates.

Okay. So for some context, about 99% of US businesses are small businesses and small businesses employ 40 something percent of the US workforce. So this is a really substantial part of the US economy. For today, with the Small Business Credit Survey, we’re thinking about small businesses as firms with fewer than 500 employees. Importantly, this is going to include two segments, employer firms, so those that have paid employees and non-employer firms, or those with no employees other than the owner, I’ll use that term interchangeably with sole proprietor.

So over the past several years, as GenAI has come onto the scene and changed very quickly, there’s been research trying to kind of keep pace and it seems like small businesses are less likely to have adopted AI than larger businesses. We’re going to explore some of that today. That’s what motivated these kind of supplemental modules in the SBCS over the past couple years. So in 2024, there was a single text field question asking about AI adoption. And then based on responses to that in part in 2025, there was a full survey module asking about AI adoption. So I’ll be sharing some results from both of those survey years today.

So to start, are small businesses using AI? Yes. We find that about 60% of firms, including both those groups, employers and non-employers, are either using actively or planning to use AI. It is true. We find that employer firms are more likely to be using AI currently than our non-employer firms. And I’ll show some cross tabulations by firm characteristics, by owner characteristics, but basically that’s going to hold is that whatever categories we’re talking about, employer firms are more likely to be using AI than our non-employer firms. Note that that unsure category at the right is pretty small and it doesn’t vary meaningfully across a lot of the indicators we’ll be looking at, so I’m going to mostly leave it aside for now.

Okay. So within small businesses, larger firms are more likely to be using AI. So from left to right, you can see those employer, number of employee size categories, or at the right we’ve got the largest groups at 50 to 499 and they’re more likely to be using AI relative to those smallest groups. The groups planning to use AI in the next year, but not currently using, it’s pretty flat across all those size categories, about 15%. And then finally, the firms that have said affirmatively, no, I’m not going to use AI. That’s pretty flat until we get to that largest category, the 50 to 499, where we see a substantially smaller group that are in that category. So putting it all together, we see that the largest firms are the most likely to be using or planning to use AI even within that small business group that we’re talking about.

As we might expect, industry is an important dimension of variation here and AI adoption is a lot higher in some industries. I’m presenting results here just for employer firms, but the rank order of these is consistent across non-employer firms. It’s just simpler to talk about employer firms only. There’s some of what we expect here, right? So finance and insurance, professional services: The types of industries where a lot of the work is done with the computer have pretty high AI adoption already. The only group here way at the bottom where you see fewer than 50% either using or planning to use are leisure and hospitality, so think restaurants, hotels, those kinds of things.

So what do the firms say? Here’s a pretty typical example from one of our professional services firms at the top. “Yes, of course we use AI, design, engineering, HR, anything that helps us be better, faster, more competitive.” Here’s a group. It’s actually a construction firm, so lower levels of adoption that’s nonetheless using AI. “As a forward-thinking, eco-friendly plumbing and HVAC company, we’re always looking for innovative ways to enhance our services and improve our customer’s experiences.” And then this is a pretty typical response we got at the bottom that we’ve included because it always makes us chuckle: “No, we make cheeseburgers.” I’d just like to point out though that, so this restaurant from the leisure and hospitality group, there’s substantial variation there, right? There are a lot of firms that we heard that are also using AI for menu design and graphic design, et cetera.

Okay. We know that industry is an important dimension, a variation here. There are some differences in adoption between employer and non-employer firms. So for one, we found this to be really kind of interesting and exciting. Among non-employers, sole proprietors, older firms are less likely to be using AI. I’ve included that green bar, which is using AI currently, as well as that gray bar that shows these are the firms that have said, no, I am not going to use AI. So sort of not surprising that there’s that inverse relationship there, but this is a relationship that we don’t see among employer firms. It’s just among the sole proprietors.

So part of that has to do with owner age among non-proprietors, right? So your firm’s been around longer, you may be older. So from left to right here, you can see that among the sole proprietor businesses, older owners are less likely to be using AI in their firm, perhaps slightly greater adoption among women owned firms than men owned firms, but that’s marginal, pretty flat across race and ethnicity of the owner.

And then one trend that we see that persists across employer and non-employer firms is that firms in rural areas are substantially less likely to be using AI. So same thing here for employer firms, pretty similar, right? There are substantially fewer firms in the sample, employer firms where the owner’s under 35, which is why that confidence bar is so big on that category, but generally that trend holds where the older the owner is, the less likely they are to be using AI. Same thing with gender. Interestingly here, we do see some differences in AI adoption by owner race where non-Hispanic white owner firms are more likely to be using AI. And then finally, we see that same geographic trend hold. So rural firms are less likely to be using AI.

Okay. How are small businesses using AI? “We use it for everything we can pretty much.” That was something that we heard. And again, this is fall 2024. “We use it to simplify business processes, synthesize information, save time and increase productivity.” And then we had some more kind of interesting and creative applications here. This is one that stood out to us: “AI is a powerful tool for training and hair braiding using different techniques to provide personalized interactive learning.” And then finally, another category that I think we’re going to start seeing more of is this integration of physical sensors with AI processes. “We use predictive maintenance models and machine learning algorithms to monitor equipment performance and provide early warnings of potential failures, thereby reducing downtime and maintenance costs.”

So we asked firms what tasks they’re using AI for and we see some differences here between firms that are actively using AI and those that are planning to use AI. This again is just for employer firms. The patterns are pretty similar for non-employer firms. The bigger source of variation here was between those actively using AI and those just planning to use AI. The technology is changing so quickly. We think part of this could reflect the functionality of what’s available. So if it’s working and you have a task in mind, you’re already going to be using it, versus maybe it’s not quite there yet so you’re still planning to. And then just one final note is that most firms are using AI in their firms for more than two tasks. It’s just under a quarter only doing one thing.

So we’ve asked firms to describe how integrated AI is into their operations and this is where we do start to see some interesting things, differences emerge between employer and non-employer firms. So there’s a marginally significant difference here in the fully integrated column. So the non-employers are more likely to describe AI as being fully integrated into their operations. That jumps out even more when we’re looking at non-employer firms versus the 50 to 499 employee category, and that’s a pretty substantial difference.

And then again, we might expect the depth of integration does vary by industry. So again, finance and insurance at the top, overwhelmingly firms are saying that AI is at least partially integrated into their operations. One thing that’s a little surprising, that second industry category, business support and consumer services, which includes things like barbers and travel agents, they were second to the bottom among actual users of AI, and a slide that I showed earlier, and yet those who are using it describe it as at least partially integrated into their operations.

So how important is AI to small businesses’ work? We asked specifically how important is the AI that you have adopted in your firm to the production of your core goods and services? So not ancillary things, the actual meat of what your business is doing. And more than 60% of firms say that it’s at least somewhat important. And then we really do see things pull apart by non-employer versus employer firms. So substantially more non-employer firms are more likely to say that AI is very important to the production of their core goods and services and then they’re substantially less likely to say it’s not at all important.

Again, importance meaningfully varies by industry as we might expect. So again, think about those kind of computer-based jobs, professional services, finance. For that top category, over a third of those firms say that AI is very important to the production of their core goods and services, versus down at the bottom, manufacturing retail where a majority of firms using AI say that it’s not important to their core work.

So we’ve talked about adoption levels, how integrated AI is into firms’ work. In what way specifically is it changing? So there are going to be a series of five questions that I’ll go through. One is we asked how AI adoption changed spending on outside services, spending on payroll costs, productivity, the quality of the goods and services you provide, and then sales. So we’ll go through each of those in turn. So first, about a third of firms reported changes in how much they spent on outside services because of AI. And again, we see differences between non-employer and employer firms here. So in particular, non-employer firms were more likely to say that they decreased their spending on outside services than employer firms. One quick note before we go on to is we thought it interesting how large a portion of firms actually didn’t know. They weren’t sure how AI had changed their spending on outside services.

So stick with me here. We’re going to talk about changes in payroll spending based on AI adoption. So the question that we asked was, how has your business’s use of AI changed spending on labor costs, (payroll)? Most reported no change, well over 50% by every firm size category. Again, many aren’t sure how their payroll costs have changed as a result of AI adoption. What starts to get interesting is in the groups where we saw changes in spending on payroll. So for this, we’re going to think about each firm’s size category in turn. So for the smallest ones, the one to four, and five to nine employees, we did see increases in spending on payroll, but to a much greater degree we saw decreases in spending on payroll. So among those smallest businesses, they were more likely to say that as a result of AI adoption, they decreased spending on labor costs.

For the larger groups there, we can’t really distinguish. Maybe there’s some marginal difference there in the same direction in that largest size category, but that to us was a notable finding that among the smallest firms, they’re reporting decreases in spending on labor costs.

And then among all the firms that reported any changes, we asked them to try to quantify change in headcounts. We’re talking about costs before, now it’s head count. Among all firms that report any change, the median change was a decrease of one employee.

So next we’re going to talk about changes in productivity, quality of goods and services, and sales. Everyone thinks that AI makes them more productive. So I just want to point that out. Really large green bars there. There’s some gray, which is no change, but for the most part, people using AI think that it makes them more productive. There’s no quantity measure there. It’s just we’re literally asking people, “Are you more productive now?”

When we think about changes in the quality of goods and services, there’s kind of an interesting difference between employer and non-employer firms. A substantial portion thought that using AI improved the quality of the goods and services they offer, but that was more true for the non-employer firms. And when we think about employer firms, they were still positive, but to a greater degree they said there’s really no change. And then finally looking at sales, for the most part there was no change, although some indicated that they were improving sales by adopting AI. No one said decreases anywhere. That’s just another thing that stood out. So AI, they don’t think is making them worse off in any of these measures.

So finally, what have small businesses found challenging about adopting AI? Again, I’m showing employer firm results because these were actually pretty similar across employer non-employer firms. And the biggest difference here was between people currently using AI versus those planning to use AI. So some of these make sense for where you are in your AI adoption journey. If you’re struggling to find or adapt the right tools, if you are trying to figure out how to train your employees, it kind of makes sense that that’s something that people planning to use AI might be actively struggling with to a greater extent. One that really stood out to me as somebody who has been starting to use AI over the past couple of years is that the group that has adopted AI was substantially more worried about accuracy. So bias or misinformation, which could be a result of saying, “Oh, this tool is not what I expected. I have to figure out how to make it work.” It could be now that it’s in my business, it matters a lot more that I’m getting it right. But that was one really interesting data point that I will say is consistent with my experience of starting to use AI more seriously. And I’m curious to hear from you all about that as well.

So what did the firms say? “As a precision machine shop, we might use AI, but we still have to be sure that our parts are perfect and AI is still too unpredictable and still gets a lot of questions wrong, that it may be a while before it can be trusted 100%. For now, it’s helpful with some of the more remedial aspects of our business.”

“We’re not using AI for any customer focused activities because we feel AI does not allow for a genuine customer experience, nor does it allow us to keep our pulse on customer needs, interests, or impressions.”

So now we asked among firms that responded they weren’t using AI and didn’t plan to use AI. Why was that? The majority of firms say that it’s not applicable to their business. Think back to, “No, we make cheeseburgers.” So, far and away, that was the greatest response, followed by prefer not to use. So we can kind of think of those in a similar vein. The remainder of these do kind of follow in rank order, the challenges of people already using AI, so finding the right tools, data privacy concerns, accuracy again, ethical and social concerns came up a fair amount. And then a lot simply found it too difficult or confusing or weren’t sure where to start.

So here’s a variety of responses we received here: “We’re going old school. We had internet operations in the past. Now customers come through our front door, buy something and leave. I think there’s a future in the past.”

“We’d love to have some help with developing a way to use it. Small businesses can’t compete with the big boys to undertake the cost and time or money of development. We just can’t stretch ourselves any further.”

And finally, this one really resonated: “No, I don’t use or plan to use AI. I don’t know how and there’s nobody to teach me.”

So wrapping up here, we know that small businesses are using AI. Even though adoption is higher among larger firms, it’s the smallest firms in this group that say it’s more integrated into their operations and it’s more important to their work. And for all the opportunity that AI can provide, we can see there are clearly still some challenges and barriers, so lower adoption in rural areas for example, or figuring out how to find the right tools.

So just to bring us back to the scale of this, we’re talking about 99% of all US businesses and almost half of the US workforce. We’re thrilled to have this really amazing qualitative and quantitative data from the small business credit survey to hear directly from businesses, but there’s also a massive ecosystem and infrastructure that exists to support small businesses as such a huge part of the US economy.

So that’s why I’m going to hand off to Rocio shortly to introduce our guests up here and we’ll hear directly from them. Thank you.

Rocio Sanchez-Moyano:

All right. Thank you, Natalie, for grounding us really in the data behind AI adoption among small businesses. But research is just part of what we do here in the community engagement and analysis function at the Federal Reserve Bank of San Francisco. We also engage with leaders from the community development ecosystem. Throughout these conversations, we’ve been hearing increasingly how small businesses are navigating the opportunities and the challenges that come with AI adoption. What we’re learning is that the story isn’t just about the technology, it’s about the entire ecosystem that supports small business success from the technical assistance providers helping entrepreneurs understand new tools, to lenders adapting their products and services, to the business owners themselves finding the practical ways to integrate AI into their operations. We’re really seeing change at every level. And we’re fortunate today to have two practitioners with us who are on the front lines of this transformation. They bring firsthand experience about the small business ecosystem and are going to share with us the AI adoption, what it looks like in practice.

Well, to my left, you’re right. Bulbul Gupta is the president and CEO of Pacific Community Ventures (PCV), one of the US’s first impact investing funds and a nonprofit community development financial institution or CDFI. There, she is leading a restorative capital strategy focused on quality jobs and ethical AI to build economic mobility in underserved communities. Prior to PCV, Bulbul helped co-found an ethical AI think tank in Berkeley and spent her career nurturing entrepreneurial ecosystems in the US and abroad.

Hope Hartman is the executive director of the Larimer Small Business Development Center (SBDC) in Larimer County, Colorado. Hope is an executive leader with background spanning software, startups, education, and nonprofit leadership. She has worked in business development and training across 28 countries and currently leads initiatives supporting entrepreneurs, small businesses and AI innovation throughout Colorado and the national SBDC network.

Thanks to all of you who submitted questions online, there are so many incredible conversations to be had on this topic. I’ve really tried to weave in as many as possible today. So we started with the 30,000 foot view from Natalie about what we’re seeing in terms of AI adoption among the small businesses. Hope, can you tell us a little bit more about what does an SBDC do? Who are the small businesses you engage with and then what are you seeing in terms of how they’re using AI?

Hope Hartman:

Sure. I’m so excited to be here today. Thank you. So the SBDCs, for those of you who aren’t familiar, we have three core pillars of service. We offer no cost one-on-one confidential advising, practical business education, and connection to resources. And then in regards to who we serve, it’s really across industries. We literally start with the idea stage to growth to scale and even succession and exit planning. I mentioned it’s all the industries. And then in regards to how they’re using AI, it’s changing rapidly. So about six months ago, the low hanging fruit was marketing and think marketing content. I think that’s where everybody started. Now we’re seeing a lot more experimentation, operations, financial analysis, market research. So it’s really starting to explode and how people are willing to experiment.

Rocio Sanchez-Moyano:

Great. Bulbul, can you tell us a little bit, what does a CDFI do? Who are your small business clients and where are you seeing opportunities in the adoption of AI among your clients?

Bulbul Gupta:

Yeah, thanks so much for having this conversation. So Pacific Community Ventures, much like many other CDFIs, our mission is really to make sure that we’re investing the bulk of our capital into low and moderate income communities and low and moderate income entrepreneurs. And so we meet that mission with, and we’re a small business CDFI, so we’re really intentionally investing low cost capital to make sure in our case that we’re not just doing working capital lending, but matching that with mentorship and sometimes working with SBDCs to do that as well. And in our case, as a social impact investor, wanting to measure that it’s not just financial return, but that we’re also improving job quality and economic mobility in the entrepreneurs and their workers. So that’s what we measure for ourselves and our investments are statewide. So we’re investing in small businesses throughout the state of California, which is the world’s fourth largest economy.

And this past year is actually the first year we’ve surveyed our small businesses on how they’re using AI, where they’re using it, comfort levels. So I’m so excited to dig deeper into Natalie’s slides. But similarly, we see that about 80% of our clients are reporting using AI in some way. About 40% would say that they’re using it for emails, scheduling, pretty base level functions and about 20% so far report using it in core functions. So finance, bookkeeping, like much deeper. And about 20% would say that they’re very early exploring and another 20% is just like not using it. So a lot of overlaps. We’re actually seeing, I think the biggest demand is in like, how would we use this smarter for marketing? Because I think again, if we think about the fact that our average small business owner for PCV is about seven to 10 workers in that sweet spot where we’re really looking at business and job trade off potentially, which we’ll talk about more. We want to make sure that we’re helping them drive customer traffic. And so that’s definitely a curiosity we’re seeing more and more out of our initial results so far.

Rocio Sanchez-Moyano:

Great. Thank you. And Bulbul, I want to stick with you because the flip, we’re talking about these are the opportunities among the small business owners, but I know that there’s growing pains in the adoption. Where do you see risks? Where do you see the challenges that your small businesses are having?

Bulbul Gupta:

Yeah. And again, not super surprisingly, Natalie’s Slides picked up on some of this, but I think the two biggest things we’re seeing from our clients and community partners we work with to reach underserved clients better and better. They’re sort of two major themes. One is access and feasibility. So can I access tools that I trust through trusted vendors, trusted partners? Because much like we hear with any one of our small business owners, like I’m in the business during the day, I’m on the business at night, I don’t know how to vet five different bookkeepers, five different financial softwares. I also don’t have the time to vet five different AI tools, right? So how do I know what tools are going to be best for my size of business, my industry, my sector? So I would say access and like trusted, vetted tools is one major bucket.

And the other major bucket is really around trust and safety, data privacy. And we see this, and similar, I think Natalie showed this in her slides. We see a bigger comfort and one of the latest reports out of Intuit and ICIC that also did a national survey of small business owners. I think one of their findings was also that businesses owned by non-people of color and males tend to have a higher comfort level using AI in their small business. And if our investments are also going into low and moderate income and underserved communities, how do we build data privacy, consent, safety that their data is going to be used in ethical ways? Who owns that data? How is it going to be used to make decisions about them that they don’t have clarity or transparency on?

And so that trust and safety barrier is also preventing I think many small businesses who could potentially be testing where could I use this with myself or my coworkers to optimize, which is the ideal way we want to see people using it, right? Not necessarily displacing staff, hopefully in jobs, but we want to see adoption happen where it’s optimization of human and machine together. So I think that’s a challenge and adoption we’re starting to see more and more.

Rocio Sanchez-Moyano:

Great. And Hope, how about you? Where are you seeing some challenges in adoption? And I know you’ve also been thinking about how to help small businesses prepare in advance for some of those challenges.

Hope Hartman:

Yeah. Thanks for that. I also wanted to mention just in regards to the data we saw, a lot of the clients that we help, it’s more, I would say microsize, 25 employees or less, but even the sweet spot is 10 employees or less, just so we know what group we’re talking about. So I think there’s a few pieces to this. One, I am concerned about people establishing policy and if policy is too strong of a word, best practices or guidelines, I think we need to start there. That would also involve the ethics around AI usage. I think small businesses really want to get in front of this, but then the other aspects to this are letting the business strategy lead the why behind the AI. I think it’s really easy to get caught up with the tech trend. It’s very exciting. There’s new products being released every week, but I think businesses really need to say, why do we need to use this or consider adopting this tool and sort of assessing people’s readiness?

So one, if you’re talking about the sole proprietor or the owner, they might be innovative and say, “Hey, I want to do this,” but then you have a team of people. So you have to assess all the people in the business and then do you have clear workflows? Because again, how do you know what tool to use for what if you’re not really clear on the process of your business. And then there’s the assessment of the tools, but then just running maybe a little experiment of how could we implement this and what’s the return on investment? Why does this matter? And then I think if people can see the gains, a lot of people will immediately say, “Oh, there’s time savings.” And I think that that’s huge for all of us, who doesn’t want to save time, but I think there’s other productivity gains and so I think there’s just a risk if people get really caught up in the latest tech trend. And so again, just assessing why they’re using it, what they’re hoping to gain from it and being aware of how it changes people’s roles when you introduce this incredibly fast technology.

Rocio Sanchez-Moyano:

Thank you. So I know a lot of the folks in the room today and a lot of our guests online are all part of this broader small business support ecosystem. Given what we’re seeing both in terms of opportunities and challenges, what type of support is needed to help small businesses during this adoption process? Hope I’ll start with you.

Hope Hartman:

I really don’t think it’s any one organization’s responsibility. I don’t see how anybody could do it because the needs are so vast and wide. So for me, it’s a lot about the collaboration and finding out who’s doing what work in this space and how can we disseminate it, making sure it’s really accessible on every level. I do look at, I don’t know why in my mind, I think public libraries would be a really great place to help because so many people feel safe in a library. So I’m sort of envisioning AI hubs where there’s all kinds of organizations and collaborators giving to that effort.

Bulbul Gupta:

That is where we started the first computer labs in the 1990s. Well, for public use, we used to … All right, I’m old. I remember that.

Rocio Sanchez-Moyano:

Great segue into Bulbul, how are you thinking about the ecosystem of support?

Bulbul Gupta:

Yeah. So it’s funny we were just talking before this about how fast this conversation has happened in like all of a sudden, all small businesses and ecosystem providers have to figure out an AI strategy or feel like they have to. We’ve been slow cooking our data and AI work behind the scenes for a few years. I think what we’re really trying to look for and ensure in the ecosystem is sort of two major things. So similar to what I was saying earlier, our small business owners are regularly looking for like, who can I trust that is going to … I get pitched tools all the time. I don’t know how to vet different folks for different functions that I could use support on. I’m open to admitting I could use support on, I mean all of us working parents unite. So if we can build both curated, vetted tools, I think that is a huge time savings and you were talking about that as a way to optimize any one of us, time wise.

And in our case, we’re also, our small business owners regularly say, “I want to get mentorship and advice, but I need that person to understand my lived experience as an entrepreneur.” Their advice has to be applicable to someone who does not necessarily have friends or family money or does not have an easy access to getting a bank loan, which is why they’re probably coming to CDFIs and works for what I have access to and what I can make happen. So I think culturally competent and trusted, vetted ecosystem providers. So how do people like us work together to get trusted, vetted resources and competent advisors?

The other bucket we see is integration and ease of use of the tools. So trusted resources and vendors or partners is one thing. How does any one of my tools speak to the other to make, so I don’t have to go into five different tools to use five different software programs. And that is actually part of why we announced a partnership last week with Anthropic that launched Claude for Small Business where they’re intentionally integrating Anthropic Workday, PayPal, Intuit, and a handful of other companies whose products and services are intentionally designed for micro and small business owners to help small business owners integrate, whether it is payroll, invoicing, bookkeeping.

How does this tool intentionally designed with this many partners bring down the time challenge and the ease of integration of all of these tools speaking to each other rolled out with AI fluency workshops with partners like us. And the same way where we use Claude in our technology stack for like our voice, AI or other work, feedback loops to their team to meet their public benefit mission of how are their tools being used to advance economic mobility and social impact for them to continue to learn how to design products and services better. So I think for us that feedback loop was part of what we agreed to join this partnership, but those are the kinds of things we hear small business owners expressing the most.

Rocio Sanchez-Moyano:

Great. And I want to stay on this ecosystem level and dive a little bit deeper into the work that both of your organizations are doing. And I think you’ve both teed this up. Hope, I know that both you at your local SBDC, but as also as part of the national SBDC network have been thinking really deeply about AI and that the America’s SBDC actually has a program called AIU that you’ve been involved on in the curriculum committee. Can you tell us a little bit more about that effort?

Hope Hartman:

I can, yes. So America’s SBDC, we received a $10 million grant from the Google Foundation for a three year term to increase AI education. And so it has a few prongs to it. One is how do we educate the small businesses that we serve? Well, we have to train the trainers. So a big part was train the trainers and that was through establishing a learning management system, an LMS, developing a full day certification program, having ongoing workshops. And then the final part of that was a student engagement project where we could work with university or even high school students to give them real world experience, real problems and using AI. So yes, that has been a huge endeavor. I was on the curriculum team. There were 10 of us nationwide from Alaska, Hawaii, Georgia, Colorado, and we were tasked with training 300 of our advisors last September. It’s the largest group of people I’ve ever trained live, but there’s close to a thousand SBDCs, so it’s all states and US territory. So to me, it’s still not enough, but you have to start somewhere.

Rocio Sanchez-Moyano:

That’s great. And I know part of your work on the curriculum committee was actually having to redesign curriculum. Can you tell us about what’s changing, what necessitated, like having to implement change and what did you learn from that change?

Hope Hartman:

So as we all know, it is just evolving so quickly and I likened AI, I said, “You can walk and then run, maybe get on your bike, get on the bus,” but this is like being on a speeding bullet train. This is how quickly it’s evolving. So by the time the first version was published, it already seemed outdated. So that’s one. So that’s going to be tricky for any of us developing recordings. And I would dare say there was an over-reliance on AI for that first version. In my strong opinion, it was a little generic. It was repetitive. Some of what was on the screen didn’t really match what was the auditory part, which was all AI generated and it sounded incredibly robotic. And so when we looked at it, we said, “Well, first of all, we need real humans.” It’s okay. We can use AI, but we want real people doing the real work and we don’t just want generic content that then there’s an assessment.

We want people to be exposed to some concepts, but then as we’re empowering people, they need to go and try it. And part of the assessment is what they learned when they were trying it. Did they have to iterate? Was it one-shot prompting, few-shot prompting? All these little techniques. And so we think it’s better. Also, Agentic AI came out right when we released the first version and there was nothing about AI agents. So we included that. But six months from now, where will it be?

Rocio Sanchez-Moyano:

Well, and Bulbul, I know that PCV has had its own adoption journey and thinking about how you’re integrating AI into your organization and how do you serve small business clients with them. Can you tell us what you’ve learned so far?

Bulbul Gupta:

Yeah. So our AI journey started, again, sort of really slow behind the scenes, four or five years ago. When I was coming into PCV, I think part of where we were starting is we’ve been a CDFI for many years. We’ve invested in entrepreneurs for many years. What do we know and understand about our own data? What works, what doesn’t work, for what type of client does it work best? How do we then better design the product, the service from scratch to be super entrepreneur centric, meeting them where they are and then making sure we hang onto them in order to have an economic mobility outcome over multiple years. And so the better we understood, and we started working with this data and AI startup that I had known from the impact investing world from many years ago on segmenting our data, helping us design better and better data analytics like feedback loops that then go back to the lending team, the advising team, the research team.

That segmentation work, we then brought in a voice AI tool that they had pioneered to be able to actually listen to the experiences of workers in our small businesses in their own natural language. So in any major language spoken in the world, you can do voice prompts into this AI tool and the synthesis provided back, the natural language sentiment analysis provided back, helps us bring that feedback, kind of like Natalie did with us, to entrepreneurs to say, “This is what your representative sample of workers is saying about their experience of job quality in your business.” The bulk of these entrepreneurs don’t have HR teams. They don’t have someone to do this for them. So how can I, as a decision maker and a business owner, and I’ve been an entrepreneur before this role, use that feedback to then better design my decisions of if I have margins in my business, how do I invest longer term for what is better for my overall staff, right?

Is it health insurance this quarter? Is it living wage? What is that next thing that they need and want and can keep me on this journey? And then we really went through a big data governance and setting in our AI policy over really two years, redesigning our whole technology stack, making sure that we were practicing our AI policy guardrails so not to do disproportionate harm. Do no harm is a really, really hard bar to meet using any technology tools, but how do we make sure we’re not doing any disproportionate harm to any of the communities we serve, making sure we have opt-in, opt-out functions, data privacy protocols, and as a CDFI, no personally identifying information is put into AI tools. And then also just making sure we’re being transparent and holding ourselves accountable in what our clients can expect from us.

And then lastly, actually that worker voice study I mentioned earlier launches today. So, to be able to share the experience of how we’re using AI transparently with our community and then most recently we have for the first time trained our own large language model on our own inclusive lending data. So when we think about increasing efficiencies to optimize use of these tools with human decision makers inside our lending team or our team, how do we help them get to yes even faster through all the paperwork? So this is really trained on our lending data, our underwriting values and matrix to help them as a coworker make decisions more efficiently so they can actually spend their time, we call it humaning harder to get more of the maybes to yes.

So that’s really I think the multi-year journey we’ve been on and we’re now launching the first national cohort with 10 to 12 other CDFIs to actually train the first nationally representative algorithm on inclusive lending data from the CDFI community, not bringing in models that have been trained on what is considered successful in traditional finance. And that I think stands to unlock, I think the Kellogg Foundation estimated like 10 years ago we could unlock $8 trillion in GDP growth if we actually fully invested in underestimated entrepreneurs in this country. That I think stands to unlock that answer in at least one of the most promising developments and use of AI in our organizations.

Rocio Sanchez-Moyano:

Great. This is a multi-year process, there’s a multi-year process going forward. How are you thinking about what comes next and how has your experience really informed that?

Bulbul Gupta:

I think the thing that sort of sits with me the most is twofold. One, given the conversations we’ve had in the last bit, how do we make sure that as we are iterating this on behalf of and with, and with guidance and input from the communities we serve, which again are largely low and moderate income, underserved, have traditionally been left out of good access to capital, that we bring our communities with us on this whole journey and that requires a much higher bar of transparency from us, guardrails and being really upfront about our ethical use policies. So the trust and safety piece is something I think about a lot to make sure we’re not accelerating or exacerbating harms. And the other I think is we need more proof points of what does ethical AI look like in the work that we do in community development because there isn’t a lot of examples of where it’s used for actual economic mobility or upliftment outcomes, and that hurts the trust and safety and community use, right? So we want to see more communities be able to own, govern and have voice and agency over how AI is developed and then used by, for, and with them, not on them.

Rocio Sanchez-Moyano:

Great. This has been such an engaging conversation and I wish I could keep you here all day, but I’m going to limit myself to just one more question, but it’s kind of a big picture one, which is what are we watching? What are you specifically watching in this space? What are the gaps in our knowledge that still need to be filled in order for everyone in this room and online to be thinking about how to support small businesses as they go through this adoption process?

Hope Hartman:

Well, I think everyone needs to know small businesses, they’re completely overwhelmed. I mean, it’s like you cannot turn off the work and so this is yet another thing that they need to learn. I think sometimes because of that overwhelm there is resistance to it, but I do feel that we all need to upskill and part of that is just increased awareness. And so what I’m watching is the transition and adaption and adoption because we are hearing that over 50% of businesses use it. We saw some data earlier. There’s a projection that up to 75% of small businesses will be using it by the end of the year, but still in that set, over 50% said they don’t know what they’re doing. So again, that’s maybe like getting caught up with the trend. And so I’m concerned about the human transition. The tech is moving so fast and a lot of people don’t like change.

So how do we write ethically and carefully, educate, and get people on board because it’s just going to start showing up. It’s showing up in non-AI platforms, QuickBooks. It’s showing up in tools that business owners use all the time, whether they like it or not. So I don’t want to see people get left behind and that’s a concern I have.

Bulbul Gupta:

Yeah. I sort of touched on this earlier, but I think piggybacking off of that, we need to make sure we’re enabling safe spaces where people can test and learn and have more advisors, trusted partners like us doing that. I think in our case, I’m also looking for how are different segments of small businesses using it? Where are they optimizing efficiency? Where are there going to be trade offs? As we were seeing some in the earlier slides, we’re obviously watching for that as a good jobs investor. We want to make sure we continue to see job quality in our small businesses. So it’s too early yet, but those are the trade-offs I think we’re going to be looking for is where solopreneurs, micropreneurs, small and growing businesses, how do each of them use AI tools? Where is there optimization and where are there trade-offs they’re going to be facing?

Because with inflation, tariffs, climate change, I mean so many other things that small businesses have to address, thinning margins in their business, trade-offs are inevitable in any economic downturn, right? And so how does this fast accelerating technology that there’s definitely FOMO about, as my 15 year old would say, how does this then trigger the next level of trade-offs? And that’s something we’re going to be watching really closely.

Rocio Sanchez-Moyano:

Great. Thank you both so much for joining us today and sharing everything that you’re learning. Thank you everyone for attending our event today. There’s a lot more EERN yet to come this year, so please stay tuned. Our next event is July 1st. Larry Schmidt from MIT will present his latest research on impacts of the technological advancements on human capital and the labor market. Also, if you haven’t already, please subscribe to EERN for notifications of new content on AI, including research from our Federal Reserve economists, speeches from FOMC members, business and community insights from here in the 12th District, and upcoming EERN events. I’d also encourage everyone to check out the EERN archive for recording of our past events. Thank you for staying connected with us at EERN and see you next time.