Event Replay: How AI Expands What Small Businesses Can Do

[00:00:00] Mark Murray: I’m Mark Murray, Editorial Director at OpenAI. Today’s Forum conversation will look at how AI is expanding what people can do at work, with a special focus on small businesses. First, we’re going to hear from Ted Taylor and Larissa Guetter. Ted and his wife run three restaurants in New Mexico, and Larissa runs an ATV business in Minnesota. They’re going to tell us a little bit about their businesses and show us how they’re using AI. After their demonstrations, we’re going to hear from Alex Martin Richmond and Caroline Chin from OpenAI’s Economic Research team about their new paper, “Work at the Frontier: How AI Is Expanding What People Do at Work.” Their research looks at how people are using AI to take on work outside their usual roles, including work a small business might otherwise need to hire or outsource. So let’s get started with Ted and Larissa. Welcome, both of you.

[00:01:03] Ted Taylor: Hello.

[00:01:05] Larissa Guetter: Morning.

[00:01:06] Mark Murray: Hello. Can you please both briefly introduce yourselves and tell us what you’re going to be demonstrating today? And I’m going to start with Ted.

[00:01:17] Ted Taylor: Yeah, thanks, Mark. So I’m a former tech guy who semi-retired into restaurant ownership. I own three Schlotzsky’s restaurants in New Mexico across Albuquerque, Rio Rancho, and Santa Fe. And today, I’m going to show a fairly simple example of how I’ve been using AI to work on pricing strategies and annual pricing updates for these restaurants.

[00:01:42] Mark Murray: And Larissa, go ahead.

[00:01:43] Larissa Guetter: Hi, I’m Larissa. I run a company called the ATV Big Air Tour. We are live motorsports entertainment, and this year we are traveling to 26 different events throughout the United States, going from coast to coast, where we perform and we launch 400-pound ATVs in the air during a live performance. Today, I’m going to be showing you how I use the new ChatGPT Work to schedule my morning briefings to give me all the information that I need for my business every morning, and how I use another automation to basically scour the internet and tell me all about my AEO and my SEO, and how I can improve those so I can start being found by artificial intelligence engines.

[00:02:26] Mark Murray: And before we get to the demonstrations, is there some advice that you guys end up having for small-business owners who want to get started on using AI, incorporating it into their workflows? What kind of advice would you end up giving someone?

[00:02:40] Ted Taylor: Well, the first is just get your hands on it and start playing with it. You know, I find every day that I spend in it, I’m finding new applications and new capabilities. So not only are there tremendously powerful tools out there right now, but more get released all the time. So get in there, play with it, see what it can do, and check out all the add-ins. It’s amazing.

[00:03:06] Mark Murray: Larissa, what’s your advice?

[00:03:09] Larissa Guetter: My best advice would be to start using voice mode. I don’t know anyone that can type as fast as they can talk, and a lot of the times by the time you get out your phone to start typing, you’ve already lost that thought. So I use Work mode and voice mode for pretty much everything. After I tell ChatGPT what I want to do, I usually tell it my goals before I have it start working for me, and then I always give it the prompt to ask me any more information that you need before you go ahead and start working. A lot of the times, ChatGPT needs just a little bit more context, and as soon as it has that, it can completely change its output and give you something that is completely tailored to exactly, not just what you want, but to get your end goal.

[00:03:52] Mark Murray: Thanks for that. Okay, let’s move into the demos. Ted, we’re going to start with you and how you’re using AI to help run three restaurants in New Mexico. Take it away.

[00:04:02] Ted Taylor: Okay, so here I’m going to start with just a chat entry around the objective, which is to update our annual pricing. I give the chat specifics on location. I did some work in advance just to make this simpler, but in advance, you can do things like identifying your competitors, looking at their year-over-year pricing changes, doing comparisons on menu items. You can also do quality comparisons to properly position yourself. A lot of detailed work you can do, and the more of that you do up front, the better the result will be. You can also iterate, so don’t stress about it. In this example, I’m just asking it to do an update on my menu pricing based on local competitor pricing and portion sizes, giving it some specific targets for the annual price increase for the in-store and then also for delivery items, like DoorDash or Uber Eats.

[00:04:58] Ted Taylor: There’s always a certain markup. You can also do all of that benchmarking using ChatGPT, which I did in advance to come up with these targets. Really straightforward in this example. But actually, you can make this more and more powerful if you’re willing to put the time into it. So here, this is the complete prompt. And I, again, add a little bit to the strategy about things that I want to see have more price increase or less price increase. You can ask it to avoid price shocks. Here I give it a table of the menu items categorized, and then the current pricing in store and delivery. You just drop that table right into the system to pick up as part of the prompt. There’s really a lot more you can do, too, as a restaurant owner.

[00:05:51] Ted Taylor: Some of the key targets are around cost of goods. You can make this analysis more sophisticated by feeding in the sales per menu item over the last six months or year so it can take that into account. You can also give it a direction to do local or national competitor analysis. And you can even go in and give it your supplier pricing by item and have it help build out estimates of your cost of goods per menu item to refine the targets even further. There’s almost no end to what you can do. And as a franchisee, I get pretty solid input from our franchise, but it’s always helpful to really take and do a deep dive locally and see what’s happening in your specific market and how to make your pricing reasonable, taking into account the inflation we’re experiencing, but still competitive with the other companies out there selling somewhat similar products. Really useful to do this through ChatGPT. It’s just so fast. So doing this myself manually, I wouldn’t be able to do anywhere near this kind of competitive analysis. It would take me days, whereas it just tears through it in a matter of moments. In fact, here we’re done.

[00:07:08] Ted Taylor: It gives you a summary. It can give you a spreadsheet summary that has the overall pricing strategy and then line-item-level descriptions. For this, I’ll just ask it to give a quick example output on the screen here without having to go into any other files, but it’ll really give you a thorough explanation of its logic and justifications for every item. Then you can go through it. In my experience, I’ve done this for a couple of stores and it’s really pretty close. Sometimes it’s missing a little bit of context on the first try, so you can either manually adjust it or you can give it the context so it does it even better next time. Here’s a quick example of some of the outputs. It looks roughly like you’d think. It’s a list of menu items with the current pricing, the recommended pricing in store and delivery. So a tremendous time saver. It did a remarkably good job. And like I said, I continue to explore more and more tools I can use in these businesses. I think that wraps it up. Thanks, Mark.

[00:08:05] Larissa Guetter: Okay. And so what I wanted to show you today is what I do in the new ChatGPT Work. So you can see right now I’m just going into my scheduled. Every morning I have two different briefings that I schedule. The first one is my morning briefing that’s basically giving me a summary of everything that is going on with that tour that is coming up. So what events do we have this week? What is listed out there with our media? A lot of the times we aren’t the only people that show our event information. We have multiple different places sharing it for us. This example is showing a Facebook post. When I ran this report, it showed me that the client that we’re working with today actually advertised our show at the wrong date. And then this one that is showing you, they advertised it at, this one says it starts at 8 p.m. when it really starts at 7 p.m. But I had actually no idea that this Chamber of Commerce had even listed our event.

[00:09:03] Larissa Guetter: So ChatGPT runs this report for me every morning. It finds all the inconsistencies all the way across the internet and tries to find me clickable links so that I can go through and make sure that all of this information is correct. Prior to ChatGPT doing this for me every morning, I would spend hours going through and finding our Facebook, our website, our client’s website, media websites, everywhere that had our event listing on. And I would have to try and correct all the information that sometimes got wrong. A lot of the times we’re dealing with 26 different events every year. We’re dealing with five or six different people that do media just for that one event. A lot of times things can get lost in communications, and so ChatGPT really just helps me figure out what I need to focus on and what is the most important. And then it also keeps a running list of all the different things that it’s changed for me so that I have a record of what happened with our marketing.

[00:09:56] Larissa Guetter: And then the second thing that it does for me is it finds me all of the recent ChatGPT and OpenAI developments that could pertain to my business. So I love to try and stay on top of everything that’s coming with all this new technology and how I can apply it to my business because I’m a team of two. Me and my husband run this event, but we actually are using ChatGPT to compete with other businesses that have the biggest budgets in the world. And so this is just helping me do the work of multiple team members. And then the second check that I run every morning is called the AI visibility check. So this is helping me make sure that my SEO and my AEO, the answer engine optimization, is correct everywhere online. So not only does this help me keep my current customers happy to make sure the information that’s being relayed is correct, but it’s also helping me get found and stay on top of it with a small business, because my goal is to be searchable from AI. The way that people are searching for information these days is changing, and as a small business, it’s very important, I believe, for us to be on top of this so that we can be seen on the same level as our larger competitors. So with ChatGPT, this is helping me find the clickable links to go through everything that I need to every single day to make sure that everything is being displayed the way it was supposed to. So, for example, last night when I ran this prompt, ChatGPT actually alerted me that my Frequently Asked Questions section of my ATV Big Air Tour website was actually not working correctly and was only able to pull three questions out of a list of 30. So it actually couldn’t give any more information about my business because that error kept showing up. So this is just what I use every single morning to try and figure out what I need to pay attention to so that obviously my customers that are coming to events have the correct information. And moving forward, my business is represented not by just SEO, but the AEO and the way that AI shows my search information.

[00:11:58] Mark Murray: Thanks so much, Ted and Larissa. And now we’re going to hear from two economists at OpenAI who are going to share more from their recent paper on how AI is expanding what people do at work. Those economists are Alex Martin Richmond and Caroline Chin. Welcome, both of you. It’s a pleasure to have you.

[00:12:16] Alex Martin Richmond: Thanks so much.

[00:12:18] Mark Murray: All right, Alex, I’m going to throw you the first question. Start with the big picture. What did you learn about how AI is changing who does what at work?

[00:12:28] Alex Martin Richmond: Awesome. Happy to answer this. So broadly what we do in this report is we take individuals, we link them to their occupations, and we link them to their work-related ChatGPT activity. And what we see is we see people do, of course, lots of tasks that we typically associate with their occupation or group of occupations, but we’re seeing them start to expand more and more to tasks outside their occupations. We’re calling this task crossover. So the biggest takeaway from this report is that we’re seeing people do a lot of tasks that are traditionally not ascribed to their occupation or job function.

[00:13:05] Mark Murray: Got it. And Caroline, what does AI allow people to do that they might not have been able to do on their own before?

[00:13:12] Caroline Chin: Thanks. So maybe one example of this is that people who may not have a technical background can now use AI to write a simple script or an analysis that they might have otherwise needed to outsource or ask another team member for. And so this really enables people to expand the kinds of work that they’re able to do on their own.

[00:13:29] Mark Murray: Got it. And, Caroline, another question for you is which kinds of work tend to travel most across roles, and which workers are taking on the widest range of tasks?

[00:13:48] Caroline Chin: Yeah, so we really see the kinds of work that travels really centered on two different categories of tasks. And so the first, which you might imagine, is engineering-related tasks. And so we see that this may include writing scripts, but it also may include debugging technical issues or troubleshooting IT problems. And so these are settings in which someone may need to otherwise call up an IT department or file a ticket. But using AI, they can really take that into their own hands and troubleshoot it themselves. The second category that we see is in marketing. And so I think these are the types of tasks where people are really trying to promote the kinds of work that they’re doing and would have required a specialist, someone with more expertise in this field, and they’re able to maybe do the first cut themselves.

[00:14:42] Mark Murray: Alex, what did the data show about the relationship between workplace size and cross-functional AI use?

[00:14:49] Alex Martin Richmond: Yeah, I think this is a super interesting part of the work. So we link these users to the workspace they’re a part of.

[00:14:54] Alex Martin Richmond: Users that sit in a workspace that has two to five seats do about 19% of this cross-occupation work. It’s about 19% of messages. At the largest tier of workspace seat that we measure, it’s down to 16%, which is this roughly three percentage point difference. It’s almost 20%, a little less than 20%, of this base of 16%. So this is a pretty big gradient for this kind of effect for this workspace size. It was really interesting to us to see this clear trend, a clear declining trend from more cross- or outside-occupation work at the smallest, in that sort of end users from the smallest workplaces, to much, much less at users from larger workspaces.

[00:15:44] Mark Murray: All right, Alex, what might explain why this pattern is stronger at small firms rather than larger ones?

[00:15:50] Alex Martin Richmond: Yeah, I mean, I think we all probably have some idea of how this might work. If you’re looking around and you’re at a place with two or three or four other people, there just aren’t that many other people that can help you with whatever you’re doing. And so I think naturally at these smaller workspaces, people are used to being scrappy or they’re used to being more creative. And so when they don’t have someone else to turn to that can do this particular task, they’re more likely to try using AI to sort of expand the range of things they can do.

[00:16:20] Mark Murray: Yeah, and Caroline, you ended up hearing the demos from Larissa and Ted. We’re talking about small businesses. How could AI empower a small business to take on things that might otherwise require another hire or outside support?

[00:16:35] Caroline Chin: And so I think a small business requires people to do tasks across a whole range of functions, and you know that this is typically very difficult for a single person to do. And so imagine a 10-person company who’s trying to launch a new product. It could use AI to organize customer feedback, to turn those findings into a marketing plan, to help them draft email copy or social posts, to run website operations. So we think that this really allows people to take that first step towards implementing these sorts of tasks that otherwise would have required outsourcing or hiring somebody else.

[00:17:19] Mark Murray: Thanks, Caroline. And Alex, as a big consumer of the reports and papers that you end up producing, kind of going on to some other things that you’ve done in the past about AI trainings, obviously for this kind of crossover, what kind of trainings do people need to be able to have to actually have this kind of crossover support and see these kinds of crossover skills?

[00:17:43] Alex Martin Richmond: So I think this is a really big and important question, and some of our work addresses this, and I think there are still lots of open research questions here. I’d put the kind of skills you need to develop into two buckets. Sort of one is asking or doing or trying to figure out how to do these tasks. And I think, as Ted and Larissa both sort of alluded to, a lot of that is about trying. It’s about going to AI, being creative about what things you’re doing that you could try doing with AI. Capabilities are improving rapidly, so if you tried something six months or a year ago and it didn’t work very well, you should try that again. So people should be seeking and actively engaging with these kinds of tasks to figure out what things they could be doing and how they could be most helpful. And I really think learning by doing is the best way to go there. The other trickier part of this is when you start to expand to tasks outside your normal expertise or experience area, sometimes you don’t know what you don’t know. And I think this is the sort of brand-new world we’re living in, where sometimes you can do things and you can create things that you have never evaluated before. I can create a website. I’ve never professionally launched a website as part of my professional career. And so I’m not used to looking at a website and saying, is this a good website? Is this going to be annoying for someone that’s not me to use? Is this, for example, going to be ADA accessible if someone who is blind needs to access my website? There are all these things that you might not know that you don’t know when it comes to evaluating and putting into practice these outputs that are outside your traditional expertise. And this is where I think we have more work to do to understand exactly how people can develop the expertise they need over time to make these tasks part of their new workflow and not just experiments and, less ideally, failed experiments.

[00:19:52] Mark Murray: And so I think we are still learning how to teach people how to develop those kinds of skills, but certainly the first step is asking and the second step is figuring out not only how to refine your own outputs, but also, if you’re a business owner or you run a small business, how are you making sure that your employees, for example, are verifying these new kinds of outputs they’re creating as well. Caroline, your thoughts on training and how people can end up getting this kind of crossover ability?

[00:20:23] Caroline Chin: Yeah, I think I broadly echo what Alex is saying. I think we’ve talked a lot about how people can try different tasks using AI, things that are typically outside of their expertise. But then I think many of the examples that we mentioned are intentionally kind of earlier-stage ideas. How do we think about these kinds of marketing plans? How can we write a simple script? What can we do to troubleshoot some IT bugs? But I think to go from that step, from running these one-off analyses to test a bunch of different things, to explore the space of what AI could do and what you could do with AI, to really turning that into a whole production that can scale to a lot of users, that can scale for a bigger team, is a transition that I think we’re all learning. We’re all learning how to develop what could be allocated to AI and what maybe we would need to consult somebody more specialized for. And so I think that skill is something that we’re all trying to develop, and that’s rapidly changing over time.

[00:21:37] Mark Murray: Since I have two labor economists that I’m speaking with, I’d love for you to kind of end up wrapping on a big picture, using almost history. Are we at a moment where this kind of crossover work and doing crossover roles is almost unprecedented, that people are moving into things that they have not been able to do before? Talk about our moment in history with this. And Alex, I’ll start with you first.

[00:22:05] Alex Martin Richmond: So Mark, on the grand scale of human history, people were generalists, right? I guess you can even think really about a pre-agricultural-revolution time, right? Or I imagine there was a little bit of specialization. I mean, I guess you had hunters and gatherers, for example. But in general, I think the really broad sweep of the history of work is people becoming more specialized over time. And so what I think is interesting about this inflection point is that maybe we’re at a point, actually a rare point, where some people are being rewarded for being more generalist again. And so people are taking on a wider variety of work. I think that’s one of the sort of earliest takeaways from this line of research. We have a lot more work to do here, but it seems like this is the kind of technology that’s enabling people to, broadly speaking, add to their workflows. And we know that there are real coordination and communication costs in terms of handing off work to other people. I mean, I think that’s one of the most complex parts of my job, even though it can often seem simple, is communicating effectively to someone else how you’re going to hand off a task to them and what you need to do, what they need to do next, and what your original vision was. And I think it’s really interesting that AI sort of means that that communication cost, relative to the cost of bringing in someone else to do it, might be larger. And so we are rewarding, in some sense, in cases where you can more seamlessly work with AI to produce those outputs, this kind of generalist attitude.

[00:23:51] Mark Murray: Caroline, the same question to you, and moving away from the hunters and gatherers, but at least in the last 50 to 100 years, how do you see this moment in AI?

[00:24:03] Caroline Chin: Yeah, so I think thinking back to the previous waves of technological innovation. If you were to ask somebody in 1940, where the agricultural sector was one of the largest employers of workers in the U.S., what they would envision if we had mechanized labor, they might say, who knows what kind of work we might be doing? Maybe we’ll all have a lot of free time. And I don’t think one thing that they would have suggested is that now we’d have a whole bunch of mobile app developers or web developers or e-commerce workers, or the way in which technology can really enable different ways of working has rapidly changed over time.

[00:24:50] Caroline Chin: And I see that AI, this is an example of potentially another shift in the way in which we work, in that the platform under which we conduct work has evolved a lot. People are now using AI to do all kinds of things that previously would have required a lot of specialized skills or a lot of time. And so I see this as potentially, this is an early indication of potentially another shift in the kinds of work that people are doing and the medium in which people are doing it.

[00:25:28] Mark Murray: Thanks, Caroline. All right, now we’re going to get to our Q&A section. Thank you for all the people who have been paying attention and listening in on today’s conversation. And we’re going to take this question from Svetlana Romanova. She asks, what evidence would distinguish genuine capability expansion from simply shifting specialist work and its risks onto small-business owners?

[00:25:52] Mark Murray: What evidence would distinguish genuine capability expansion from simply shifting specialist work and its risks onto small-business owners? Caroline, you want to take a crack at that question?

[00:26:04] Caroline Chin: Yeah, sure. And so I think the question is really pointing at maybe one of the findings in the report, which is that, well, what are we going to do if people are really using AI for legal advice and drafting legal documents, or taking on the risk of these sorts of heavily licensed occupations? And I think that one thing that we’re really hoping for here is that people maybe, like we find that people are asking AI for advice, but then not necessarily, like we don’t really look into whether or not they’re conducting these sorts of end-to-end tasks. And so we really see this as an opportunity for people to consult AI for financial information, just broadly for legal information. But then, of course, if you’re actually going to file a legal case, you’re actually going to do something that requires these licensed skills, you would still likely hire a specialist.

[00:27:12] Mark Murray: Alex, what are your thoughts?

[00:27:14] Alex Martin Richmond: Yeah, I largely echo what Caroline said. And just to maybe mention again something I had earlier, is that we are probably going to have to develop mechanisms, if you’re a small-business owner, for deciding where it’s appropriate to delegate and generating some basic review procedures for AI-generated work, especially when you’re trying new tasks and especially when those tasks have traditionally belonged to regulated occupations.

[00:27:44] Mark Murray: Got it. Here’s our final question that we ended up getting from Q&A. The question is, what one idea would you hope small-business owners take away from today? And just talking about your research, Alex, what would your answer be to that?

[00:28:00] Alex Martin Richmond: I think the big takeaway from this report is that lots of people are trying new things that are outside their traditional lane, if you will, with AI. So I hope that if you’re a small-business owner, you can be empowered by that and you can take away from today that you could try something new, that you could expand your views on what AI might be capable of. And this in itself is kind of a skill you need to develop, of sitting down to do something at your computer and saying, oh wow, I wonder if I could try outsourcing that, how that could go. And that’s a skill I’m developing for myself still every day. But I think, yeah, lots of people are trying something new and you should too, I think is the thing. I’m trying that myself.

[00:28:44] Mark Murray: Caroline, what’s your answer to that question?

[00:28:47] Caroline Chin: Yeah, I think this is a really exciting moment for exploration. And I think this hints at a lot of things that we’ve mentioned before, where AI can now enable people to do so many things. And I think it’s a really exciting time to just push your business to expand, push your business to kind of a different level, to sort of see what you can do with AI, to see what’s possible using the tools provided by AI that wasn’t possible before.

[00:29:19] Mark Murray: Awesome. Alex, Caroline, thank you very, very much. Really enjoyed your paper and today’s talk. Before we close, I also want to thank Ted and Larissa for sharing their examples from their workflows. And again, thanks, Alex and Caroline, for sharing your research. If you’d like hands-on practice, check out OpenAI Academy’s Small Business Community for upcoming events and practical resources. You can also register for our next two Forum conversations, the first, Building a Personal Health Operating System on August 20, and K–12 Showcase: Lessons From Teachers and Administrators on August 24. We’ll share all three links in the chat.

[00:30:01] Mark Murray: Thanks again for joining us. Until next time.