Latest / Serious People Podcast / Nisha Balwani on Why AI Adoption Starts with the CEO
Transcript
- Nisha Balwani [00:00:00] As CEO, I was AI-pilled, and that can be really good or bad for your team, because they're like rolling their eyes. They're like, "Oh, you have another thing you've used AI for." Or, like, in a meeting, I'm like, "I'm just gonna quickly do your job with AI and tell you how long I think that should have taken you—" Which I did. I was like, "Okay, I'm asking you for a numbers update. You should be like, 'Nisha, I just Slacked it to you—'" Noah Levin [00:00:22] Welcome to the Serious People Podcast. Today, I'm talking to my friend Nisha Balwani, CEO of RCI Technologies. Nisha is a second-generation CEO taking her parents' established services business into the AI era. We talked about what it takes to make AI work for her government clients, for her team, and for herself as the leader of a growing business. I love comparing notes with Nisha, and we would have had this conversation whether or not we were recording it, so I'm happy to get to share it with you. Today on Serious People, Nisha Balwani. So, Nisha, welcome to Serious People." Nisha Balwani [00:00:59] Noah, thanks for having me. I never thought our paths would collide in this way, but I'm happy they did. Noah Levin [00:01:04] So, okay, so—so, there's so much to talk about. I wanna talk about your life as a second-generation CEO, and what it's like to take over a technology company that your parents founded from the 1980s and try and bring it to its full potential in 2026. We gotta talk about AI things you are building inside the business for your own ops, and then I think we're gonna talk also about the work you've started doing to help other companies with AI as well. So there's lots to get to. Should we start with you and your story? Nisha Balwani [00:01:31] Yeah, definitely. So, I'm a second-generation CEO, and so I didn't always know I was gonna take over the business. The AI stuff gets interesting because, as a second-generation CEO, my biggest challenge is scaling the firm and not building it zero to one, which takes a whole lot of chops that, you know, I'm glad my parents had and did that side of it. And I think to take their platform and scale it has really been... It's been—it's been an honor, but it's also just been a really big operations challenge. And so, like you mentioned, we're building a lot of AI tooling for our firm, and I'm happy to go into that in more detail too. Noah Levin [00:02:02] So, RCI is a technology company— Nisha Balwani [00:02:06] Yes. Noah Levin [00:02:07] —and is building software for government agencies, with a couple different models of how to do that. But not all tech companies are the same kind of tech company. And so I think one of the interesting things about your job right now is you're leading this company that has been around since the '80s, has a certain set of skills, a certain set of talent, and a certain type of customer, right? And you're in this brave new world of AI, trying to map that onto this, like, business that's, as I understand it, like, going pretty great already, but might be disrupted by AI. Nisha Balwani [00:02:40] Totally, yeah. Like, I have this thought all the time, which is: we largely staff developers and engineers, and what is the future of developers and engineers, right? Like, I'm thinking about that. And then, what is the future of tech builds and IT services? So there's just a lot there. But the government is also slower to adopt some of these technologies—rightfully, right? Like, they can't just OpenClaw their way to stuff. They have so many secure needs, and they're a little slower. So I don't see this as like an overnight dilemma, but it is something I'm thinking about. And I think also something that was interesting was, like, right when all this AI stuff was getting hot— I was obviously just, like, you, super, super into it. And my clients were not. They're not as into it, because they don't want to be delivered AI slop. They're more wary of it. So it was interesting to toe the line between, like, okay, we're an established firm with a brand who needs to be looking towards AI, but we can't be so AI-pilled that our clients are like, "All the work you're doing is AI, so why should we pay you?" So I think most B2B services firms that I'm talking to are feeling kind of that same idea. They're getting, like, addendums to their contracts from their clients about, "What's your AI policy?" So there's this interesting middle ground, I think, that we have to strike as well with our clients. Noah Levin [00:03:53] So, maybe we can get into, like, a specific client example in a minute, but talk through the life cycle of a typical engagement. Does it start typically with, like, an RFP? Nisha Balwani [00:04:03] That's right. Yeah. And they're, like, multi-, multi-year. So, like, I think we just started a project that we won in 2023, and we started that this week. So, a lot of this is highly regulated—to no fault of, you know, anyone, it's, they have a different department for procurement, so we're in touch with them. I will say they start with RFP, and then usually it's a relationship that you have, so you're not blindly kind of doing RFP work, because there's pretty a lot of work that goes into them. Noah Levin [00:04:30] Is being good at writing RFPs, is that, like, the strategic competency for a company like yours? Nisha Balwani [00:04:39] I love that you asked that question, by the way, because the second you wanna look into AI for government contractors, it's like, "Oh, we could do proposal writing, and we could make an agent do this and do that." Okay, but, like, that's not the job. The job is building the relationships first. So it's kind of interesting—like, yes, do I wanna accelerate the timeline of compliance in the proposal, or, like, how many things does this proposal need, and project manage it? Yes, of course. I don't want my people spending more time on that. But is it that I wanna do more proposals? I don't think so. It's that I wanna find the right proposals, and that depends on the relationship. So it's interesting where I think about, like, the AI and the human side of things—what you see on the news, which is we're actually seeing people hiring more. I think that's where you see there's still a human need to a lot of this side of the business. Noah Levin [00:05:24] Let's, like, zoom in on whoever is your best salesperson—we don't have to name them, but you know who they are. Is that person getting any leverage from AI, or, if they're not, should they be? Or is it all just, like, charm and three-martini lunches? Nisha Balwani [00:05:40] It is three-martini lunches. Um, but I also really like that question, because my best salesperson is so AI-pilled, and—I don't want to give away his age, but he's definitely over 60. So he's kind of the greatest example of where I think age is not a factor in this. He jumped on it right away to do, like, all the things you should be doing: researching, creating strategies for areas that you're not tapping, because maybe you didn't have the bandwidth before, maybe it was something you would assign to a junior; tracking conversations—like, everything you should use AI to help accelerate your job, instead of taking away from, you know, doing the relationship building. And I keep telling my team, "If you can use some tools to accelerate your job, then you could do more relationship building." I actually think everyone on my team, not even just sales, needs to be in some sort of relationship. Like, if you're the recruiter, 100%, you need a relationship with every person you talk to, every developer you talk to, right? 'Cause we have a pipeline of people. If you're admin and you're helping people onboard, I think you should also be on the phone. So I'm like, "If you're on your computer hacking away at an Excel sheet today, I'm upset." Noah Levin [00:06:48] I think this is—you know, when you and I talk to companies, we're oftentimes talking to the executive at the company about their AI strategy. And one of the things that I think is sort of a paradox of this is that if you're asking them where they need operational leverage—the bottleneck in your company—you end up looking for, like, the sales force that's out in the field, where there's tens or hundreds of people, or the customer service team, where there's an army of those, where, like, the variable cost is high 'cause there's a large number of people doing the job. But then it kind of leaves you with this problem of, like, how do you value the CEO who you're talking to having more time back in their day? Which is also, like, what you and I are trying to figure out for ourselves. A lot of our day is, you know, am I spending too much time futzing around over making myself more productive, or is that actually, like, giving me leverage to go and talk to people, which is what I need to be doing more of my day with? Nisha Balwani [00:07:38] Right. I also think a good way, as a CEO, to think about your operational bottlenecks is, like, what was the last fire you were pulled into—to get your time back? It's so telling when you're pulled into a fire and you're like, "What?" Like, peel that onion back and you'll figure out where you could make a difference. 'Cause sometimes saying, like, "What's your bottleneck?"—I found that, even myself as CEO, I usually answer like what I want. "I wanna double my revenue." Okay, it's not really diagnosing the problem. But if you're like, "Hey, what were the last three client fires you had?" Then, like, now let's talk. I don't know if I can, like, share it without, you know, ragging on my team right now. But, um, we are seeing—funny enough, with AI—like, I love it and I hate it, 'cause we're seeing a lot of candidates come through, and when we interview them on video, they're using some form of AI tool. Um, and we're pushing our clients to just, you know—if you're trying to find a good fit for your team from our bench of engineers, interview them in person, 'cause they're coming on in person anyway. Noah Levin [00:08:39] We're all looking for these people, right? These are the people who are embedded in a function, who are good at doing a thing and are also good at, or seeking to be good at, AI, can be the change maker on the team. And obviously one way to do that is just by example, right? Building things for yourself that make you more powerful in your job, and then everyone kind of looks at you and might try and copy it. But are you seeing any successes with this salesperson who's super AI-pilled and has figured out ways to leverage it for his job, propagating that to the rest of the team? Nisha Balwani [00:09:10] I definitely am. And let me also comment on the first part of what you said. I think it's such an interesting thought about how to bring this to your team. That was the first half of this year for me, which was, like, how to strike the right tone with it as well. Like, I think what people forget on podcasts, like what me and you are doing—it's like, people feel a lot of fear, who maybe don't know about AI as much, or maybe they feel like their job is gonna be taken away. So I think it's, like, very important to strike that balance. And what I find interesting is: As CEO, I was AI-pilled, and that can be really good or bad for your team, because they're like rolling their eyes. They're like, "Oh, you have another thing you've used AI for." Or, like, in a meeting, I'm like, "I'm just gonna quickly do your job with AI and tell you how long I think that should have taken you." Which I did. I was like, "Okay, I'm asking you for a numbers update. You should be like, 'Nisha, I just Slacked it to you.'" Like, that's what I want. So I think they were kind of getting tired of that. So we held, like, an in-person meeting—which I think everyone did, you know—and said, "This is gonna be about AI." Which I started by saying, "Every elephant in the room." I made a slide with every quote I could hear, which was like, "Eye-roll, my CEO's AI-pilled," or, "I'm gonna lose my job," or, "This company won't exist in five years." Like, what did you hear about AI? And I made everyone go around the room and not talk about the business side of it, but, like, "Tell me what you, at home, think about AI." And then at the end I was like, "I don't want to hear your political thoughts on it. I don't want to hear, like, what you think about AI compliance and security, but I just know that your job will be replaced by someone who uses AI. So I'm not gonna lower head count, but I will probably change the dynamic of the team." So there was some seriousness around that, I think that helped. And then I think our key salesperson—I had him demo at that live event, and I was like, "Show us something that matters to the team, not just to you," and that helped a ton. And I think his age also really helped people, 'cause they were like, "Oh, shit, if he can do it and he's learning it..." Fun fact, I also had Saleem come and demo stuff. He was like, "Use Gemini." It was, like, "Show us Claude." No. Anyway, I think that meeting went really well for the team, and I think, instead of me showing something I built, which they were really tired of, it was fun to show someone else having built something. Noah Levin [00:11:22] It's a real change-management case study. It's a new wave of technology that's gonna disrupt everything, and everyone has their own personal adoption curve to climb, right? Along with the companies. There's an emotional component to it. Nisha Balwani [00:11:35] Yeah. And we're still struggling after that. Like, I think we have people who are, like, superstar level, and then just people who are lagging, of course. And then the problem is that the superstars kind of cover for the people who are lagging, 'cause they're building stuff for them, and they're not building that muscle themselves. So I'm truly at a conundrum on how—now, what my next thing is. I think we did the beginning really well, and now it's, like, okay, how do we get to, like, phase two, to bring everyone along? But do you see that as well? Like, when you're building some of these tools for people, it's like—I could build the best tool, but if you don't have people who are excited by it or want to use it, that's also gonna be a problem. Noah Levin [00:12:14] Yeah. Oh, yeah. There needs to be an inside champion for every one of these projects, and I'll give you two examples from some recent clients. One client I'm working with right now is a health-tech company, but I think a 20- or 30-year-old health-tech company, right? So the technology is not SaaS in the modern way—it's a services company that has a very powerful technology backend. And so change management for AI at that company is really about getting these people who have been doing a thing and have a lot of pride in the way the operation works to be excited about a new mode of working. And the big win that happened last night is I got an email from one of them with a document that she wrote with Claude. By the way, this is a senior engineer who's wildly smart at this company but is cautious about AI and is trying to figure out the use cases for it. And she said, "You know, I think a lot of the knowledge that we are capturing in our business is happening on Zoom. So Claude and I wrote this document about how to capture our Zoom meetings and get it into a knowledge base, and Python scripts." And she's like, "You know, what do you think?" Like—awesome. At that point, it doesn't even matter if the plan is right. I'm just so happy you wrote it, right? Nisha Balwani [00:13:19] Yeah. I was excited enough to share with you. That's awesome. Noah Levin [00:13:23] But I do think it's, like, you learn by doing. So you can always see where it's going, and then be like, "Okay, at least we're moving in a direction, and now we can, like, change the direction." And you don't know, 'cause everyone's company is different and the culture is different. But I definitely agree that you need an internal champion who's willing to try and fail and try again. So, when the government says, "One AI, please"—what's the current appetite for AI in the products you're building for them? Nisha Balwani [00:13:50] For what we're doing with the government right now, there has been, I would say, maybe, like, 10% of it is AI-related—maybe some level of AI acceleration within, like, project management or business analyst work. So they're not gonna be that, you know, AI-forward. I am seeing them bid out some, like, strategic AI work, more on, like, a strategy side—so working with, like, the McKinseys and the Accentures of the world. And then we're working with those firms to build out, like, more AI-forward teams. So that I'm seeing, but it wasn't like a transformation the way it was on the private side. Noah Levin [00:14:26] What do you think you do if you're personally AI-pilled, but you happen to work at an entity—like a government entity—that just isn't climbing the adoption curve at the same rate that you are? Are there any levers you can pull, from where you sit—as, like, one of your clients, for example—to get some leverage out of AI within the confines of, you know, the rules that you're playing by? Nisha Balwani [00:14:46] Yeah, that's a good question. I definitely, like, go to dinner with some clients who I'm very friendly with and who are super AI-pilled, and we spend the whole dinner, like, kind of like me and you, just riffing on all AI stuff. But I feel like that's a question I should go back and ask them. Does it kill you to wake up in the morning and not be able to do that at work? I think it's interesting. I think maybe they see more of the complications of it, because they're so close to the problem every day. So maybe they're seeing a different lens than what I'm seeing. But it's a good question—like, if you're AI-pilled in a company that's not super AI-forward, I feel like that would be hard. And then, the other way, I've talked to business owners who are, like, they're the only AI-pilled one, and I'm like, that's also hard, but on both sides—on the team side and on their side. I, I personally think it's asymmetrical. Well, it's way better to have the CEO be AI-pilled and not the team than the other way around. It c—it can only, it can only cascade in one direction. It has to start—like, you have to be playing with it if you're, like, a CEO and a business owner. Like, I feel like I could not go to my team's in-person meeting and be like, "All you guys should be using AI," if I did not have an understanding of, like, where they would fail, or where AI was not that great. I don't think they would have had the same level of respect for the ask. It's so hard, though, 'cause I talk to so many business owners and leaders who want to bring AI into their company, and they're like, "I don't have the time to do this." And I get that, too. So I think it's a tough place. Noah Levin [00:16:09] Let's talk about your personal use of AI, then, for a second. At RCI, or just Nisha on your own—like, what's the thing you're doing with AI right now that you're most excited about? Nisha Balwani [00:16:19] Well, A—just super excited that we have a Claude Enterprise account, so, like, or a Claude Teams account, whatever. So just happy that I can see everyone's usage, and everyone's very much using it. Like, I feel like it's a reflection on the work that we're doing, that everyone else is also using it. And— Noah Levin [00:16:34] Are you token-maxing? Do you have a leaderboard up? Nisha Balwani [00:16:36] We're definitely not token-maxing, but we're definitely hitting usage credits left and right. I think people have built a lot of MCP connectors. So we have a huge database, as you could imagine, of talent, and we need to track, like, okay, this person was on a project, it lasted five years, now their project is ending—what's the next project we're gonna put them on? And, like, we might not get that project for two years. So there's a lot of, like, okay, who's in our database? So that has a connector to Claude, and that's been amazing, but it hits usage so quickly, 'cause it's a massive database. So, anyway, we're thinking about how to take that outside of Claude. What am I most excited about? Honestly, it's just the fact that I don't do any bullshit work anymore. As CEO, I did—I still did a lot of, like, just looking at numbers and doing analysis of, like, okay, we want to bring on an additional recruiter, which I said earlier. It's like I would've done an analysis on how much can my business sustain, and what do we think we can do with this? Literally, I did it in three seconds this morning, like, while I was drinking coffee. I was like, pull my P&L, put it in Claude. There's just a lot of time that I think I get back. And the connectors—like, being able to scan my Granola, my Slack, and my Microsoft 365. I think one of the best use cases I have is my one-on-ones with team members. I had a doc before on each person, and, like, what they like and what they wanna grow into, whatever. And I just have a folder on them now, and it's so easy to never forget anything when I'm in a one-on-one. I'm like, "What did we talk about last week?" And, like, "What's important to this person?" Oh, my God, so many use cases. When Claude is down, I don't know how to do my job. Noah Levin [00:18:00] Yeah, it is. It's sort of like the new bank holiday is when Claude has an outage and the whole country just gets to stop. Nisha Balwani [00:18:07] So— Noah Levin [00:18:07] Exactly. Nisha Balwani [00:18:09] There you go. And then, I think also, like, my team not having to log things—so, logging a CRM, I don't need to see you doing that. Like, every night, Claude just goes through your emails, goes through your messages, and logs the CRM with activities. Noah Levin [00:18:21] So the CRM, I imagine, is like a pretty important source of truth? Nisha Balwani [00:18:25] Yeah, except, funny enough, we have it in Airtable. I think for the last, like, few years, we weren't sure. We were about to go to, like, a HubSpot, or a more conventional CRM. And then, right when the MCP came out, I was like, "Oh, this is the best decision we ever made," 'cause now we're already built for automations in Airtable. Noah Levin [00:18:41] Let's just dwell on this for a minute, 'cause I think this is great, like—everyone's got a CRM, and there's all kinds of legacy ways to manage a CRM. Even—I think a lot of people will say Airtable sounds fancy relative to their Google Sheet. Nisha Balwani [00:18:53] Yeah, that's funny that you say that, 'cause I'm like, "Oh, I feel like Airtable's basic compared to some of the larger CRMs." But yes, continue. Noah Levin [00:19:00] So who set up your Airtable for CRM? Was that you? Nisha Balwani [00:19:04] Yeah, that was my whole—like, my job is to scale this company. When I first came in, I was like, "Guys, where—like, where are people's names and email addresses?" And each salesperson kind of had their own Rolodex. And so I was like, okay—and Airtable, I think, was relatively new at the time, but I already understood that we needed some sort of relationship. So, for us—like you mentioned in the beginning of this episode—we're not a traditional SaaS company. Like, I think a lot of the CRMs on the market are built for, like, okay, moving someone in a pipeline very quickly. Well, everything I described to you on this call, like, for so many reasons I won't go into again, it just would not work for us. But every person we staff—every team member we embed in a client's team—they have, like, a program manager they're working for, a client that signs off on the time sheet. Like, there's three people. And, honestly, we just go sell to the same people over and over again. So my CRM doesn't have to be this, like, new-leads big thing. So I already understood that I needed the relationship of, like, who has already worked for this program manager, and which project were they on—connect that to the contract. So I started it there. People did not like my Airtable love. They still wanted their spreadsheets. I would say this was, like, a big success now, five years later, but a huge fail in the moment. And getting people to track things they've never tracked before was hard. But, yeah— We're now using that same database that I built the bones for. And, again, funny—as, like, CEO, you're like, "I'm just gonna build the bones to this, to show you what it could look like." Noah Levin [00:20:32] So what's happening now—it sounds like Claude—to make it so that people don't have to track things manually? What's the actual routine running behind the scenes, or what are the connectors? Nisha Balwani [00:20:44] So, Airtable first—if you're starting this from scratch and it sounds super overwhelming, Airtable has a connector. You add your Airtable account there, and you just ask Claude to build a different set of information. We put, like, five or six spreadsheets into Claude, and we asked it for a schema: "If you were to design this database from scratch, and we wanna eliminate all spreadsheets—or we think we do—you tell me what you would design it as." And it gave a full doc of, like, "This is why I would do this, and this connects to this," and it—Claude visualizes. So it was really helpful for people who weren't as used to working with a relational database. I would definitely start there, and then, once it has it, it can build out each section. So you actually don't have to be doing any of the manual work. For us, what that looks like is we also have the Microsoft 365 connection for our email. So anytime there's an email logged or something, it can automatically be put into the Airtable. So let's say we had a new project kick off this week, which we did—Claude is automatically scanning it and logging it into Airtable, saying, like, "This person started. This looks like the person who signed this time sheet. This looks like the program manager, and this looks like the length of the engagement." Noah Levin [00:21:54] Is that automatic scanning happening on your, like, your team's computers, on, like, a schedule? Nisha Balwani [00:22:00] Yeah, it's happening—like, the salesperson built one, and then shared that scheduling skill with everyone else. Noah Levin [00:22:07] So everyone's required to have that running on their computer, to grab their emails and munge it into Airtable? Nisha Balwani [00:22:13] Yeah. And I think, because it's such an obvious use case of, like, no one wants to log stuff, I think everyone felt really strongly that they would do it that way. Noah Levin [00:22:21] AI is such an interesting—it's such an interesting foil to, like, the Beautiful Mind spreadsheet that every company seems to have. I'm working with a company right now—this is a CPG company, and I'm pitching them on a forecasting-and-replenishment system. And they're in an interesting spot because their head of ops, who does all of this today, is going on mat leave, and everyone on her team is also going on mat leave. And so—by the way, and this is—do you want to hear my hot take of the week? My hot take is that one of the greatest forces for AI-transformation initiatives is mat leave. I think mat leave is— Nisha Balwani [00:22:58] Wait—that is so valid. Like, PTOs and mat leave. Noah Levin [00:23:02] And it's this moment where someone who's been holding the business together by sheer force of will, with a white-knuckle grip, is suddenly like, "I have to share my spreadsheet with people? I have to explain this to them now?" And suddenly these bus-factor-of-one people at companies are having to find a way to structure their work. But what's amazing is that these Beautiful Mind spreadsheets—especially the ones that are in Google Sheets, where you can see not just the current state but the entire history of the spreadsheet, the change log of it—it's a decision trace, right? It's like a whole map of everything that's ever been done on that process, in completely unstructured, chaotic form—which is exactly what Claude wants: to just grok where it is right now, and then to spit out, "This is the process that you have been meaning to run all along." Nisha Balwani [00:23:50] Totally. And, honestly, I think people would be surprised at the size and scale of companies that are still using spreadsheets. Like, we still have a lot of spreadsheets in our firm as well, even though we have, like, custom AI software in our firm too. So it's hard to get off the spreadsheets. And exactly what I shared with Airtable, by the way—it was hard to get people off their spreadsheets. So then people had their own spreadsheets for the information that was supposed to be a single point of truth in our database. So some of it goes back to being like change-management issues as well. But also, spreadsheets are just a great source of information—like you're saying—dump them into Claude and see what Claude has to say about what you're trying to do. Noah Levin [00:24:30] Okay, we've learned that Nisha likes Airtable for unstructured data, and spreadsheets for unstructured data. Nisha Balwani [00:24:35] My Airtable, or—the other, I guess, like—now I'm gonna talk about Airtable. I'm into it, and I think we're gonna get into it, but the product we built for our invoices process—I realized it's easy, once it's connected to Airtable, for that information to keep linking back into Airtable and being the single point of truth. So that was another reason why I feel like Airtable is a good step up from a spreadsheet if you're, like, "Oh, I don't wanna build a full database somewhere else, that feels super techy." It's, like, an easy UI for someone who maybe was on a spreadsheet before, to get comfortable with Airtable. Noah Levin [00:25:05] Today's episode is sponsored by Valet. Nisha and I talk about a critical invoicing process that she had to build, that almost entirely lived in one person's head. The real work was getting those rules into a form that the rest of the team could use, while still keeping a human in the loop. That same last-mile problem shows up again and again with the things we build with AI. An internal dashboard or tool might work for the person who made it, but the rest of the team still needs a secure place to open it and share it. Valet is built for that. It turns AI-created dashboards, artifacts, and tools into secure, shareable company resources in seconds. Valet is the founding sponsor of Serious People. This team took a bet on the show before it existed, and I'm grateful to have them along for the ride. Go to valet.dev to learn more. The idea that systems like Airtable and Linear and Workday and Salesforce are the s—source-of-truth systems for data, right? The company data lives in the system. It's the one true source of that data. And the agents that we're building to work with this data are becoming the source of truth for the process—for the business process. It's like the SOP now needs to be encoded in something. And so, as you build agents in Claude, like the skill that you're de—describing that takes their emails and puts it into the CRM, or, as with the cloud-hosted agents that Valet runs, the skills and connectors that are in those agents become the source of truth for the process that you run on the data. Nisha Balwani [00:26:38] Yep. And I think what's interesting—or maybe hard to grasp if you're new to it, it definitely was hard for me to grasp in the beginning—is the real unlock is the intelligence layer. Can that agent be self-learning, and, like, bring up trends, and what else can it do beyond just doing the process the way someone else was doing it? So that's what we're working on now with even just, like, logging. Like, logging is great, but what can you actually—what can you surface, intelligence-wise, from being able to log everyone's activities? Noah Levin [00:27:03] Have you had any successes at that? Nisha Balwani [00:27:05] Not yet. Not yet. It'll come up with stuff, but it just—it doesn't feel like it's struck gold yet. Noah Levin [00:27:11] So you built some amazing stuff for invoicing at RCI. Tell the story of what you set out to build and how it went, 'cause I think you had some dark alleys you went down at various points in this process. Nisha Balwani [00:27:24] Yeah. I think this is a really good story in general, but I think what happened was everyone was using their own, you know, AI tool. At that time, we were actually using ChatGPT—so we all had our own ChatGPTs, and I was like, "Okay, but this needs to extend beyond just each of us hacking away and writing emails and documents." So I kind of saw that we needed to upgrade a workflow pretty quickly. I had this one process on my mind for a long time—for, like, the past three years. How are we gonna upgrade this process? Let me describe it. So, the reason it's difficult to modernize this one process in my business is because, again, we embed talent into our government clients. So that means they're submitting time sheets to the client, and then the client is sending them to us. So there's no portal, there's no fancy time tracker—not using Harvest, there's none of that. Yes, could we ask our consultants to double up and do something for us too? But then we have no idea if they're properly approved, and that becomes a compliance issue. Every government agency, every department, everyone has different rules for, like, how many hours you have to work before you take a lunch, how many overtime approvals you have—all this stuff. And it also all needs to map back to what we scoped for the client. So, are we above—you know, is that matching the contract, and is that matching the run rate of this contract? 'Cause then we also need to let the client know, like, "Hey, you're kinda tapping out all your milestones really quickly." So, anyway, all to say—this is something that one person in our company had very much the institutional knowledge on. This is your bus factor of one, right here. And I knew it. I've known it since I joined the firm. But it was also easy to kind of ignore, 'cause everything—ironically, if you ask me what the bottleneck in my company is, it's not that. Like, she makes zero errors. But it's just that she's the only one who knows the process. And she's a very valuable employee. So if this is becoming a very manual process, which it was, and the technology is moving quicker than her abilities, I want her to not have to be hacking away at 12 different spreadsheets. And then also, like, we started scaling, right? So now you're like, okay, maybe I was tracking this many time sheets, now I'm tracking 10x that. It's not okay for one person. I'm seeing all this, and I'm like, "This is not great." The way we first thought we would—I don't even remember now what was my first take. It was totally wrong. It was, like, we're gonna build—like, we're gonna hard-code some sort of system. I forget, but whatever first take I had at the process of how we were gonna fix this invoicing process was not AI-forward enough. Noah Levin [00:29:54] All to say, don't feel discouraged if your first version of something is totally not the right answer. Like, it was a learning for us as well. So whatever first version we had—I think it was vibe-coding a web app—it wasn't really fully taking advantage of everything that AI could do. Nisha Balwani [00:30:07] So what we ended up building is a really cool product that can scan the emails where we get the time sheets, and it can log them into a platform, and it's smart enough to know the business rules. We've taught it all the business rules—so, for this client, this is how many hours this person can work, and all that stuff. And it flags confidence levels if it's unsure, which is something we really wanted. We still wanted a human in the loop, so she's still reviewing the time sheets. And it has everything. It has an audit trail—anything that was of concern, we built into it quite easily. What's cool is AI can parse really well, it can read really well, it can reason really well. So when it's unsure about something, it's good at reasoning through it. So that part is great. I think the best part about this is, once she clicks approve on the time sheet—she used to update, like, five different spreadsheets. And now—how many times can I say Airtable on this podcast?—it goes back to our Airtable, which has the bill rate, the contract amount, all that. So all those spreadsheets we were doing to make sure we weren't running out of money on our contract— All that stuff goes back to the Airtable. So I think what was really surprising to me, and what I learned, was that originally I just wanted to build the workflow. I just wanted to help her out. Let's make it so that she's not the only one who knows how to do this, and she has some help doing, like, the really easy time sheets—the ones that are just eight hours a day. Cool. Done. What I didn't realize was how intelligent it could actually be, and how it could start surfacing trends, and how it could go back to a database and eliminate so many other spreadsheets for us. Where we haven't gone yet—where we will go with it—is, like, pushing it straight to QuickBooks for invoicing. We're just still making sure that it's at, like, a 98% success rate before we do that, 'cause once you invoice the government wrong, that's it. So, that has been such a cool use case. And, funny enough, I'm working with another client who has a very similar use case. They're also government-contracting, but they're in construction. And when I learned about their invoicing process, I felt so much better about our company, 'cause I was like, "Oh, we're not alone." They have some sort of weird government-compliance thing where they need to take 10 photos per invoice. It's not as easy to use one of the off-the-shelf tools for project management or time tracking for them. So they came up with something like we did, which was kind of a hacked-together process of spreadsheets, and we're kind of building out something similar for them, but obviously super custom, 'cause their workflow is very different than ours. Noah Levin [00:32:31] When you say that you actually built this, what does it look like to build it? Was it you doing the building? Nisha Balwani [00:32:36] It was my husband—like you mentioned, it was Saleem at first. We were trying to get him to build a prototype of what it could be, essentially—so it didn't have to work perfectly, but I wanted to show it at that live company meeting we had. So I wanted something to show the team, like, "This is where we wanna go, and we wanna build these tools internally." So it was a vibe-coded app. But then we used one of the developers on our bench, who was not staffed to a project, who is quite AI-pilled. I kind of talked to our engineers, was like, "Who wants to work on this project?" And I passed it off to them, just from a security and development standpoint. They can move a lot quicker in debugging than us vibe-coding it. But that has been the gateway to everyone else wanting something in their department. And, luckily, we do have a bench of engineers and developers who are, some level of AI-pilled but cannot do this AI work for government clients. So that's been really good. Noah Levin [00:33:34] It was a helpful tip to have, like, a husband who works in AI sitting around the house, just looking for projects. Nisha Balwani [00:33:40] Totally. He had, like, two weeks off between his high-demand jobs, and I was like, "Instead of relaxing, here's my idea for you." Noah Levin [00:33:48] I think he f**king loved it, so— Nisha Balwani [00:33:50] Loved it. He won't admit it. He won't admit it, but he was sad when I was like, "I need to take it away. I need to give it to the engineer now." And he's like, "No, no, just—just, just one more night. Just one more night." I was like, "Okay." Noah Levin [00:34:02] So I want to demystify how you actually get started building something like this, for someone who's got their own super-wacky invoicing needs. So how does someone like Saleem get the requirements for this into Claude or Codex to start building it? Were you just dictating into the microphone, "Here's how I think it works"? Nisha Balwani [00:34:19] This is a good question, because there's the way we did it, and then there's probably the way you should've done it. But the hardest part about this is, again, I don't do the workflow, so I had the person who owned the workflow—she was really the one who needed to give the information. But, again, she'd never built a technical product before, so she wasn't sure of the exact framing to give. So it does take a lot of time working with the person who is in charge of the workflow, and the more work you can do in the beginning, just understanding your current workflow—funny enough, I was on another podcast a few weeks ago, and that person's business is all about SOPs, and I'm like, "If you have your SOPs written, or if you had an agent watch your work and write it for you, that is a super helpful first step." Something that we do is a screen share of the entire workflow—every single thing you open, every single thing—and understand the source of where things are coming from. That's super helpful. And then take that screen share and have—I think ChatGPT—look screen by screen by screen by screen and write out the workflow. Even with that backbone, you're still gonna need the person who designed the workflow to make some system-design decisions. But, anyway, to answer your question, I think you really start—most of your time is understanding what the workflow is, and where do you want the human in the loop, and where do you still wanna be automated? I think those design questions are really important. And, Noah, I think you know this way better than I do, 'cause of your background in product. But I think we started there. All the logic that a human applies—so, for us, every time she looked at it, she just knew, like, "This client, these hours," or whatever. So, have that written somewhere. The system did a lot better when we gave it written documentation, like, "This client, this client, these rules." That's where I would start. We went a little bit more of a roundabout way. And, I think, also keeping in mind, you may wanna start with one client. Let's say we were like, "Hey, our biggest client, where we get the most time sheets, we're gonna start there." That's fine, but think about the other clients and the requirements they have, 'cause we built kind of the system for that one client, and then we went back and were like, "Oh, wait, there's actually other areas that the other clients have compliance for." So not just lunch, but holidays, let's say. So give that some thought when you're building that first iteration, even if you're not building it in that first iteration. Noah Levin [00:36:32] I'm hearing two things in the answer you just gave: one is, like, "Here's the right way to do it," and then, "We didn't do it that way," as the other. Which—I think that's pretty much how I feel about my entire life right now, which is, like, when I get to the end of a project and I go, "Oh, that's how I should have done it." Nisha Balwani [00:36:47] Yep. But I feel like we wouldn't have learned it unless we did it that way. And now everything we've built since then has been so much better. And the fact that we're building it for other clients is even better, 'cause I'm like, "Oh, we already made all the mistakes." And we understood what they felt like. Noah Levin [00:37:02] The other amazing thing about what you're describing is that the pain points that you're feeling trying to figure this out—everyone is going through a version of this, right? And so you and I, and a lot of other people listening, are stumbling through this acute problem, which, in this case, is: how do I take a business process that's mission-critical, that's in one person's head, and put it into software—first by getting it out of that person's head, and then by writing code around it, right? But we're not the only ones thinking about this—so are the fine folks at OpenAI, who just wrote a feature where—and I don't know if you've seen this yet—the desktop app for Codex will watch you do a thing and then will encode it. And, obviously, it may not be that simple with some processes that have edge cases and things, but you may not know this—the process you went through, of screen-recording and then using that as the basis for prototyping, is exactly what the Ramp team does. So Ramp, when they were building automations for their finance team—their own internal finance team—they basically just had them record a Loom, right, like a training video, of "Here's my screen share, here's me explaining to you, almost like you're a new hire, how I do the end-of-month closing of the books," and then that becomes the basis for, you know, building the requirements. Nisha Balwani [00:38:15] I think you're totally right. Everyone's thinking about, "How do I automate this workflow? What do I know about this workflow?" That's key, too. There's a lot that, sometimes, I think people know so innately that they can't encode it—it doesn't come out on a screen share, exactly like you mentioned, edge cases. But sometimes it's also not the right workflow to tackle first, and I think that's a bigger question. Like, it took us a while to land on the invoicing one, for many reasons, and we knew it was gonna be a little bit hairy, but we went for it anyway. I think it's also different when you have engineers and developers, 'cause you're not so worried about changing the scope a little bit. But it might not be the first one I would recommend a client do, right? 'Cause it required a lot of business logic. Noah Levin [00:38:55] How does your invoicing person feel about her job with this tool versus without it? Is she enjoying life more? Is she worried about being replaced? Nisha Balwani [00:39:04] I love that you asked that. I actually talked to another founder who is in a similar boat as me, where he's very AI-pilled, and he told me he built something over the weekend, this is months and months ago. And it was something really cool—him and I were super excited about it. And then, on Monday morning, I'm like, "How did your team like it?" And he's like, "It was silent on the team call." Like, so it's like—just for your question—it's like I was like, "What? I thought it was the coolest thing ever." He's like, "They were like, 'Yeah, cool.'" So, I think in the beginning—I had been talking to her for, like, a year about how we were gonna upgrade this process, by the way. It wasn't like all of a sudden everyone's talking AI, I sign onto a LinkedIn post and I'm like, "Guess what? We're changing our whole company." No. I had been laying the groundwork. We actually already had Looms recorded, we actually already had docs on it, so it was not new to her. I think she was excited. I think, in the midpoint, when it was failing, she was less excited. She was like, "I no longer trust it. My process is so much better." Now, when we work with people, I'm always like, "Let's be really careful about how we iterate." And that champion can easily turn, 'cause they also want that work to be 100%—like, they're in charge of that outcome, right? But now she's very excited. Now that it's in play and it's making her life a lot easier, she's able to focus more on the edge cases and the negotiating back with—again, remember, I was like, "I want you to pick up the phone and call people." She can call the contracts person and figure things out that maybe were a gray area before. So now it's good, but it's been a journey, and it'll probably continue to be one. Noah Levin [00:40:29] I want to talk about your new line of business. So, this is how I understand it—I want you to tell the origin story, but you are a person who is super AI-pilled, and is curious, and is spending your nights and weekends—you know, in addition to taking care of a child, and a husband, let's be honest, needs the care—you are learning all these things and finding ways to apply them. And also, you are the CEO of a technology-staffing firm that has been around for decades and has an army of very talented people behind you. So— Nisha Balwani [00:41:00] I need a vacation. Noah Levin [00:41:01] Yeah, you know—well, that too, you know—they have Wi-Fi in paradise. So, you've started doing your own line of AI work, and I'm curious to learn more about that. Nisha Balwani [00:41:11] Yeah, definitely. I think, because I shared the whole invoicing story, that sort of prompted some of the work that we're doing now, which is—we had this experience with AI, and it was very different from what you see online, what you see on LinkedIn. It was true, and it was raw and real, and I was like, "Oh, this is probably—other companies are feeling this pain." I'm in a few women-CEO groups, and I remember presenting this AI case to them and kind of showing them the journey, and a bunch of them were like, "Sorry, can you do this for my company?" And I was like, "Sure, happy to have a developer look into what your workflow is and what we can do for you." So, essentially, what I'm doing now is talking to small and mid-sized companies about how to bring AI into their company, like, for real. "Where should you start" is always the first question people ask. There are workflows that you may wanna automate, but they might not be good cases for automation, so we usually start there—like, where should we actually begin thinking about automation in your company? We come in and we decide what type of system to build. It could be something that just exists as a layer on top of their existing systems, if they're already using a bunch of off-the-shelf tools. In most cases, it's something custom that we're building to match their exact needs. So, a lot of times, we're kind of coming in as a partner and doing that sort of work. A company we worked with—she's an agency, has three different systems that track where her people work, how long they work at each place, and there wasn't really an off-the-shelf tool that she wanted to pay for to give her a profitability-by-client view. So that was a very small project where we were like, "Let's just build you a dashboard and a reporting tool that can surface profitability trends, also, right when it's time for contract renegotiation." So that was a no-brainer for her, 'cause she was able to renegotiate three contracts that year by saying, "Actually, we have the data, and we spend X amount of time on your contract." That was really good. I think that hit her bottom line immediately. Noah Levin [00:43:01] What's the most common fallacy in the workflow that people pick first, and where do you try and reorient them? Nisha Balwani [00:43:07] What's funny is I have a diagnostic I send to most people I work with, and there's all these questions, and the last one is just, "What didn't I ask you about?" And most of the engagements we have are truly from that "what didn't I ask you about" box, because you've already answered a bunch of questions about your business, and then something's burning inside, and then you put it in there. If I'm thinking of— Noah Levin [00:43:26] Like, talking to a consultant to do a thing in my company that I don't feel equipped to do with AI—what should I expect to pay for that? Nisha Balwani [00:43:34] Yeah, that's such a good question, 'cause it really depends on if you want to phase it out, and what the process is. I think the only thing I can really add there, to help understand complexity, is understanding where the sources are coming from. So if there's a lot of connecting that needs to be done with, like, many, many, many different APIs, that usually drives the complexity up. It's really hard to answer cost questions like that. I will say, a lot of our clients—we give them a phase one, and then phase two, phase three—like, they're kind of building on top of it, so that phase one is more of an entry point that they feel confident with. Noah Levin [00:44:07] That resonates with me—the idea that phase one is really, like, a—it's a use case, right? It's a vertical use case of, "We're gonna give you this thing that adds value," but it's also giving you an excuse to build out some foundation layers that'll make the next use case easier. So now you have a knowledge base, or you have a place to interact with your agent as a human in the loop, or whatever those commonalities are for different use cases. Nisha Balwani [00:44:29] Exactly. And I think sometimes, while we're building phase one, something comes out that we're like, "Oh, that's a really small project," and we can just nip that in the bud—let's just do that. So that's where the smaller engagements tend to come from, like, from existing clients where you're already in their ecosystem, so you already understand their stuff. Yeah, I had that same challenge—you really do need to take a few passes at discovery before you can understand the scope of work. And so, there's two ways to do it. The way I've been doing it so far is just giving away a lot of my time for free. Noah Levin [00:44:58] Um— Nisha Balwani [00:44:59] Same. Noah Levin [00:45:00] Which is actually—it's actually worth it for me right now, 'cause I'm getting a lot of value out of discovering what's going on inside of companies and starting to spot patterns. But I think there's some version of this where the discovery is its own engagement and produces an artifact that's useful even if you don't go build the thing, right? "We mapped your processes, here's what's actually happening inside your company," or, when you're ready to go build a thing, "Here's the blueprint of what you should consider building." Nisha Balwani [00:45:25] I'm glad to hear you're having the same thought, 'cause it's actually really helpful for me to understand the client pretty deeply anyway, so I don't mind doing that upfront engagement. But I also understand that it is a time commitment on both sides. If they're no longer interested, then you're like, "Oh, okay." Noah Levin [00:45:39] Can I do a quick lightning round with you before we wrap up? Nisha Balwani [00:45:43] Yeah. Noah Levin [00:45:43] All right. What is Claude doing on your computer right now? Nisha Balwani [00:45:49] Building our team's strategic deck for tomorrow morning that I'm presenting. Noah Levin [00:45:53] How's it going? Have you seen it recently? Nisha Balwani [00:45:54] We'll have to see. Honestly, this is the third month it's been doing it, so it's really good at it now. The first month, it was not good. Noah Levin [00:46:01] Did you give it feedback the first month, and build it back into the skill, or? Nisha Balwani [00:46:05] Yeah, and building better context for it was actually the key. It was like—I had it hooked up to the right themes and things like that, where it could find themes based on my conversations. And so it actually got a lot better. Noah Levin [00:46:16] That's great to hear. The iterate-and-improve part, I think, is undervalued in these skills. There's a lot of, like, "Why didn't it work?" Nisha Balwani [00:46:22] Yeah, which is why, when earlier you said, like, "I don't know if I'm wasting time or if it's valuable," I kind of feel like that month I was like, "This is such a waste—this meeting is in three hours." But then this month, I'm not even doing it—like, I'm on this fun podcast with you. Noah Levin [00:46:36] What's the stupidest thing that you've seen AI do recently? Nisha Balwani [00:46:39] Oh, God, it does a lot of stupid things sometimes, right? Why am I blanking? Noah Levin [00:46:43] You can pass if you want. Nisha Balwani [00:46:44] —is smarter than yours, Noah. No, I'm just kidding. Noah Levin [00:46:47] I wouldn't be surprised. Nisha Balwani [00:46:49] I'm just kidding. No, I think—it's not answering your question directly, but I think the biggest issue we're seeing is just trusting AI too much, and not taking that one pass at, like, "Is this real?" Like, we love AI, it's amazing—like, I'm in 10 different, you know, Claude things—but I'm seeing, even on my team, the more we're pushing AI tools, it's, like, you still need to be outcome-based. So that's not really answering your question, but that is the biggest challenge I'm having right now with AI. Noah Levin [00:47:14] I had one of those a week ago. I sent someone an email back that I told Claude to write, but it wrote the email and sent it, and the response was, "Sir, that is the most em dashes I've seen in a one-sentence email." Nisha Balwani [00:47:28] Yeah, I actually can't get it to get rid of the em dash. Like, even though I have a whole tone guide and writing guide—but I actually don't have it send emails on my behalf, 'cause I'm not that— Noah Levin [00:47:38] That's probably not—that's probably not the wrong answer. But I have a trick for you on em dashes. Nisha Balwani [00:47:43] Oh, yeah. Noah Levin [00:47:43] If you're that sure of something, right—like, no em dashes, no em dashes—you can make it a deterministic rule, and have it check for that, using a hook, so that it literally runs a script and searches for the em dashes. So I have... For me, that's the word "genuinely." Nisha Balwani [00:48:01] Oh, yeah. Noah Levin [00:48:02] And— Nisha Balwani [00:48:03] It always writes— Noah Levin [00:48:04] —it always writes "genuinely," which I think is so funny, because it's literally the opposite of genuine, 'cause it's a machine. And so now I—you see it writing the email and proofreading the draft, and it says, "Oops, I wrote 'genuinely.' Rewriting." Nisha Balwani [00:48:17] Yeah, yep, yeah. It does that with my tone guide too, but I—okay, I need to do the—I need to do the hook, because it still ekes out its em dashes sometimes. Noah Levin [00:48:26] Yeah. Once you're sure of something, putting it into a deterministic rule and tying it to code is really useful. Nisha, I'm so glad we got to do this. I would have had this conversation with you without recording it, but I'm glad we recorded it so other people can hear it, too. If people want to be helpful to you, where can they find you, and what can they do for you? Nisha Balwani [00:48:43] They can find me on LinkedIn. You can DM me if you want to talk more about building custom AI into your business, or anything about what I said today. I kind of love chatting about this stuff, as you can see on this podcast. But, yeah, that would be great. Definitely DM me on LinkedIn. I'm happy for anyone to find me. Noah Levin [00:48:59] Awesome. Nisha Balwani, thank you for coming on the Serious People podcast. Talk soon. Nisha Balwani [00:49:04] Thanks, Noah. This was so fun.