Meeting Title: Brainforge x Meg Bevilacqua Intro Call Date: 2025-08-21 Meeting participants: Meg Bevilacqua, Uttam Kumaran
WEBVTT
1 00:04:11.810 ⇒ 00:04:12.880 Uttam Kumaran: Hi, Meg.
2 00:04:14.470 ⇒ 00:04:20.039 Uttam Kumaran: Sorry for the delay, I was just on with a… with a candidate, and it was just going on a little bit longer, so….
3 00:04:20.279 ⇒ 00:04:22.869 Meg Bevilacqua: No worries. We love a good conversation.
4 00:04:22.870 ⇒ 00:04:24.679 Uttam Kumaran: How’s everything?
5 00:04:25.080 ⇒ 00:04:27.679 Meg Bevilacqua: It’s going well, …
6 00:04:28.000 ⇒ 00:04:32.939 Meg Bevilacqua: How about you? It sounds like… it looks like you’re trying to grow like gangbusters, so good for you.
7 00:04:32.940 ⇒ 00:04:42.340 Uttam Kumaran: Yeah, it’s busy. We, we are trying to grow, things are growing, but things are also, like, stretching in the process.
8 00:04:42.530 ⇒ 00:04:48.390 Uttam Kumaran: You know, this is a business I started alone, 2 years ago, and we’re about 15 people now.
9 00:04:48.440 ⇒ 00:05:02.750 Uttam Kumaran: Doing a lot of work in data, and then over the last 8 months, doing a lot of work in AI as well. So we’re getting a lot of demand, things are growing. I think we’re consistently evolving our process and also evolving our talent.
10 00:05:02.810 ⇒ 00:05:06.519 Uttam Kumaran: And I think that process of evolving our talent is…
11 00:05:06.700 ⇒ 00:05:09.460 Uttam Kumaran: Kind of the most difficult piece right now.
12 00:05:09.460 ⇒ 00:05:10.050 Meg Bevilacqua: Yep.
13 00:05:10.570 ⇒ 00:05:13.640 Uttam Kumaran: Yeah, so… Yeah, go ahead.
14 00:05:14.290 ⇒ 00:05:21.080 Meg Bevilacqua: Yeah, that’s always one of the big challenges with consulting, is your product is your people, so you have to make sure that they’re good, and it’s a good fit all around.
15 00:05:21.260 ⇒ 00:05:29.409 Uttam Kumaran: And it’s also, like, people for different stages, you know? I think when we were really scrappy, we could deal with people that were okay when there was no…
16 00:05:29.640 ⇒ 00:05:42.490 Uttam Kumaran: tickets or anything, and as you grow, all I want is structure and process. I’m a big fan of that, and so I’m really, really happy. I’m not the sort of person that pushes for
17 00:05:42.600 ⇒ 00:05:52.969 Uttam Kumaran: Chaos and, you know, not having process and procedure. But as you graduate, some people aren’t used to that, or they’ve only worked in startups, and so, ….
18 00:05:52.970 ⇒ 00:05:53.410 Meg Bevilacqua: Yep.
19 00:05:53.410 ⇒ 00:05:57.899 Uttam Kumaran: Yeah, it’s something that we’re… We’re constantly working on, but… yeah.
20 00:06:00.370 ⇒ 00:06:02.849 Meg Bevilacqua: Alright, well…
21 00:06:03.210 ⇒ 00:06:14.839 Meg Bevilacqua: I’ve read a… I’ve looked at your website, I’ve read your stuff, but why don’t you kind of tell me a little bit about what Brainforge is, and I’ll talk a little bit about my experiences, and maybe we can chat about where that might fit together.
22 00:06:14.840 ⇒ 00:06:29.600 Uttam Kumaran: Yeah, sure. So, Brainforge, again, is a data analytics and AI company, started about 2 years ago. We primarily serve mid-market, businesses implementing data strategy, AI strategy, as well as, like, full-stack data and AI teams.
23 00:06:29.840 ⇒ 00:06:49.609 Uttam Kumaran: You know, mid-market businesses, these are typically 20 to 100 million in revenue. We’ve done a lot of work in e-commerce and in B2B SaaS, just as the nature of my background and my business partner’s background. But I would say we’ve also worked at several other domains, like health, legal, you know, CPG, things like that.
24 00:06:49.930 ⇒ 00:07:00.000 Uttam Kumaran: Yeah, I would say my background is in data engineering. I worked as a data engineer, I led data teams, led product in a data startup, but I’ve worked in startups my whole career.
25 00:07:00.260 ⇒ 00:07:16.400 Uttam Kumaran: increasingly smaller ones, until I sort of decided to leave that and try to start this business. This is not, like, a venture-backed endeavor. This is a completely bootstrap business. I would say it’s hopefully closer to, like.
26 00:07:16.400 ⇒ 00:07:27.330 Uttam Kumaran: mom-and-pop shop business than it is something crazy and venture-backed. I… I just felt a lot really burned out of that model from, working in those businesses, and, I think
27 00:07:27.330 ⇒ 00:07:38.139 Uttam Kumaran: You have to compromise a lot on the quality and the way you treat people and that, and so we have a really, you know, we strive to have a great delivery, but also, as you mentioned, like, the people…
28 00:07:38.290 ⇒ 00:07:53.650 Uttam Kumaran: our product, so we try to hire great engineering talent and really treat them really well. And yeah, so from, you know, we… in the beginning, it was sort of sales just through word of mouth, and now we actually have a lot of processes around how we’re doing go-to-market.
29 00:07:53.650 ⇒ 00:08:09.239 Uttam Kumaran: So we… we have, like, outbound channels through LinkedIn, we do events, we still have a lot of word of mouth and referrals from existing clients, you know, past partners. We have a big, sort of, partnerships motion right now as well, partnering with vendors and other agencies.
30 00:08:09.270 ⇒ 00:08:22.539 Uttam Kumaran: And so, over the past, you know, few months, we’ve actually been able to grow business very, very steadily, you know, beyond anything that we’ve done in the past, to the point where, you know, we’re signing about one new client every one to two weeks at this point.
31 00:08:22.540 ⇒ 00:08:33.909 Uttam Kumaran: Which is really great. I think it leads to a lot of other challenges, where we are building our PMO, sort of as we fly. We’re sort of building out our different class of
32 00:08:34.010 ⇒ 00:08:39.420 Uttam Kumaran: Talent in each sort of service area, and trying to move, you know, as much of…
33 00:08:39.650 ⇒ 00:08:48.529 Uttam Kumaran: our executive time into working on the business versus continuing to work in the business. So that’s kind of, like, the inflection point that we’re at now.
34 00:08:48.800 ⇒ 00:08:49.770 Meg Bevilacqua: ….
35 00:08:49.900 ⇒ 00:09:03.019 Uttam Kumaran: So yeah, I mean, I think we… we… the way… the reason… I think the big driver for a lot of our growth has been the AI piece. We… in fact, we didn’t start off as, like, an AI consultancy. I was using AI a lot to build this business.
36 00:09:03.110 ⇒ 00:09:22.639 Uttam Kumaran: And building not only tons of workflows, but agents and agentic workflows, and actually brought on our first AI engineer just to work with me to continue to automate our business and give us leverage. So we’ve been able to build it super lean and cost-effective, and in that process, we actually learned what it’s like to go into a company and
37 00:09:22.730 ⇒ 00:09:25.279 Uttam Kumaran: and build AI systems that work.
38 00:09:25.450 ⇒ 00:09:28.589 Uttam Kumaran: And so that naturally led us to build, sort of, what our
39 00:09:28.730 ⇒ 00:09:44.159 Uttam Kumaran: what our services arm could look like when we go do that for clients. And so we continue to do a lot of that. I would say a lot of the stuff that I ask our AI team to do is quite a bit beyond what our clients even ask us for, so we’re always able to kind of stay on the edge there, and we’re really
40 00:09:44.210 ⇒ 00:09:52.159 Uttam Kumaran: passionate about finding leverage that way, and building, you know, a consultancy that is very AI-native. However, that doesn’t mean, like.
41 00:09:52.260 ⇒ 00:10:04.910 Uttam Kumaran: there’s still not room for people, it doesn’t mean that, like, everything about consulting is thrown out the door. In fact, it’s just more about having more documentation, more understanding of, like, what’s the 80-20 of
42 00:10:05.270 ⇒ 00:10:16.790 Uttam Kumaran: where our time should go, and how does that affect client NPS, versus, like, a lot of the pencil pushing and moving information around that can totally be augmented by AI.
43 00:10:17.080 ⇒ 00:10:18.040 Meg Bevilacqua: … Yep.
44 00:10:18.170 ⇒ 00:10:33.980 Uttam Kumaran: So that’s kind of, like, yeah, where we are. I’m here in Austin, by the way. I lived in New York for a while and grew up in the Bay Area, but our team is completely remote, so we have folks all across the U.S, and then also in Asia and in Europe as well.
45 00:10:35.850 ⇒ 00:10:53.579 Meg Bevilacqua: All right, awesome. So, I’m Meg Bevelocqua, I’m gonna say my name just so you know how to pronounce it if it comes up. Sure. And I’ve spent my whole career in professional services. So, straight out, I majored in biology, realized I didn’t want to be a biologist when I grew up, and went around looking for what I did want to do.
46 00:10:53.620 ⇒ 00:10:57.270 Meg Bevilacqua: And I landed at a very small
47 00:10:57.410 ⇒ 00:11:13.320 Meg Bevilacqua: consultancy that was helping clients get certified to ISO 9001, ISO 27001, all those professional standards, which is how I met Troy. He was the operations manager for one of our… one of the clients that I worked with, actually one of my favorite clients.
48 00:11:13.320 ⇒ 00:11:14.330 Uttam Kumaran: Oh, awesome.
49 00:11:14.550 ⇒ 00:11:19.060 Meg Bevilacqua: And that was actually a great job, because it was my job all day, every day, to learn.
50 00:11:19.110 ⇒ 00:11:22.950 Uttam Kumaran: To talk to people about what they did, how they did it, and why they did it that way.
51 00:11:23.100 ⇒ 00:11:24.180 Meg Bevilacqua: …
52 00:11:24.580 ⇒ 00:11:31.720 Meg Bevilacqua: after about 18 months of that, and that was at a very small startup, right? It was 5 people when I was there. So…
53 00:11:31.720 ⇒ 00:11:47.029 Meg Bevilacqua: After that, at that point in my life, I was ready for a little bit more structure, so I jumped to a slightly larger company, about 200, 300 people, where I was doing… externally, I was working with Fortune 500 companies on their immigration and mobility,
54 00:11:47.100 ⇒ 00:11:48.310 Meg Bevilacqua: programs.
55 00:11:48.520 ⇒ 00:11:59.740 Meg Bevilacqua: And the challenge that we kept running into there was our company had been extremely good at getting people to where they need to be, and extremely bad at managing populations afterwards.
56 00:11:59.740 ⇒ 00:12:09.749 Meg Bevilacqua: Right? So there was a whole demand from the clients for ongoing data management, ongoing population management, that we weren’t able to provide because our data backend was so bad.
57 00:12:09.750 ⇒ 00:12:12.050 Uttam Kumaran: Right. So internally.
58 00:12:12.050 ⇒ 00:12:16.680 Meg Bevilacqua: my boss said, hey, you did a bunch of process stuff, right? Figure out how to make this work.
59 00:12:16.680 ⇒ 00:12:18.430 Uttam Kumaran: Yeah. And that eventually….
60 00:12:18.460 ⇒ 00:12:33.709 Meg Bevilacqua: ballooned into redesigning our, internal case management system to go from sort of, like, a very flat information warehouse with not a lot of guidance or structure, to something that was a lot more process-driven. So you could take a…
61 00:12:33.710 ⇒ 00:12:50.700 Meg Bevilacqua: immigration consultant, which is basically sort of a junior-level paralegal, and guide them through the most common immigration processes so that they had complete data capture that enabled us to do that level of population management after the case closed that our clients were increasingly demanding.
62 00:12:50.940 ⇒ 00:12:51.950 Meg Bevilacqua: …
63 00:12:52.080 ⇒ 00:13:02.209 Meg Bevilacqua: So after that kind of wrapped, I realized that that was what I wanted to do. That was where my passion lay. I wanted to do that kind of big work for clients all the time.
64 00:13:02.240 ⇒ 00:13:20.660 Meg Bevilacqua: So I went back to business school, got my MBA, and went to go work at Deloitte. And since then, I’ve done a lot of different types of work, but the big through line has been helping clients figure out how to capture, manage, and distribute information to make the best possible decisions and run their business the most… in the most effective way possible.
65 00:13:20.970 ⇒ 00:13:23.770 Meg Bevilacqua: … as…
66 00:13:23.880 ⇒ 00:13:38.929 Meg Bevilacqua: you may have heard, nowhere in there comes in data engineering. I’m very much coming at this from the business side perspective. I’ve worked with enough data engineers, and I’ve done it, like, I’ve had enough insight to that to be able to translate very, very effectively.
67 00:13:39.100 ⇒ 00:13:54.070 Meg Bevilacqua: But my core competency is in sitting down with people and talking to them about what they do, how they do it, and why they do it that way, and then helping them, like, asking good questions to try to guide them to how to do it more… how to help them do it more effectively.
68 00:13:54.230 ⇒ 00:13:55.849 Uttam Kumaran: Yeah. And, like.
69 00:13:56.060 ⇒ 00:14:07.000 Uttam Kumaran: I guess, talk to me about those systems, like, are you… are you guys prescribing strategy, or is there also an implementation piece? Or, like, what are the typical, like, what’s the roadmap when you get engaged with the client?
70 00:14:07.380 ⇒ 00:14:21.950 Meg Bevilacqua: So, I tended to come in on the strategy end of things. So, that was where we’re talking and doing the initial type of setup. Most of the time, we do a handoff to a separate team to do implementation, but I have worked on some longer-term projects where I’ve been in
71 00:14:21.950 ⇒ 00:14:30.709 Meg Bevilacqua: running, sort of what we call the operations end of things, so taking it from that implementation and then doing post-implementation refinement.
72 00:14:30.710 ⇒ 00:14:44.709 Meg Bevilacqua: So I’m familiar with, like, how do you take an in-flight system and make adjustments? How do you look at, okay, we set this up in one way, how do we need to adjust it on the fly to make it work more effectively? And that…
73 00:14:44.780 ⇒ 00:14:55.340 Meg Bevilacqua: came in at a marketing organization, right? So we were brought in to do a redesign of their data warehousing system. Essentially, they had
74 00:14:55.720 ⇒ 00:15:06.850 Meg Bevilacqua: 6 different teams, each of which was spinning their own analytics from 7 different sources, they were all pulling in different ways, they were all getting different answers for what should be the exact same data.
75 00:15:06.960 ⇒ 00:15:11.170 Meg Bevilacqua: So we came in and helped them put together one unified data warehouse.
76 00:15:11.410 ⇒ 00:15:13.940 Meg Bevilacqua: And then…
77 00:15:14.250 ⇒ 00:15:22.030 Meg Bevilacqua: Also took all of the disparate reporting that they had and started the long process, with lots of screaming and pushback.
78 00:15:22.030 ⇒ 00:15:27.359 Uttam Kumaran: Of condensing it down to a core set of reports that everyone could agree on.
79 00:15:27.360 ⇒ 00:15:27.970 Meg Bevilacqua: Right.
80 00:15:27.970 ⇒ 00:15:36.250 Uttam Kumaran: Yeah, very similar to what we do for a lot of our clients, where on the data side, we come in, and usually we’re getting called because there’s a problem. It’s not like all…
81 00:15:36.460 ⇒ 00:15:50.399 Uttam Kumaran: figured out yet, and… but it is sort of outcome-driven. Like, we… we sell to the… to the C-suite, typically, and so the data engineering work and the data modeling work are implementation details, but we’re after outcomes.
82 00:15:50.400 ⇒ 00:15:58.300 Uttam Kumaran: Which is, are they able to get accurate data? Are we seeing them use data in meetings more? Are they requesting for…
83 00:15:58.300 ⇒ 00:16:09.209 Uttam Kumaran: for next-level analyses, right? And we… I will say, though, we nailed implementation. Like, that’s what my background is, and that’s what I built the business on, is, like.
84 00:16:09.610 ⇒ 00:16:26.280 Uttam Kumaran: truly, like, we’re confident in what we’re able to do, our understanding of the best-in-class tools, methodologies, and the pace at which we execute, but a lot of engineering teams just do that, and they never achieve an outcome, or they can never translate it, and so we have to do both.
85 00:16:26.920 ⇒ 00:16:40.670 Uttam Kumaran: I would say, and sometimes, I don’t know, again, I’m not a consultant by trade, so I try to think about, like, the angles at which people start these businesses, and I think people either come at it from one angle or another. They come at it from the strategy side, and they figure it out, how to, like.
86 00:16:40.670 ⇒ 00:16:53.569 Uttam Kumaran: do the work, or they come from my side, and then, like, try to figure out the business side of it. Luckily, in my career, I’ve worked with executives and executive reporting, and, you know, with finance and, you know.
87 00:16:53.620 ⇒ 00:17:02.719 Uttam Kumaran: done a lot of stuff on the… on the capital side, so I’ve worked with executives my whole career, so I’ve been able to do that similar translation work that you have just probably from
88 00:17:02.760 ⇒ 00:17:20.770 Uttam Kumaran: this side of the fence. That’s what, you know, life has been like, you know, in going to increasingly larger companies. We come in as usually their entire data team, so not only the implementation details, but also head of data, so, like, strategy work, KPI definitions.
89 00:17:20.800 ⇒ 00:17:42.669 Uttam Kumaran: They come to us for, hey, we want to launch a new product, like, what on the data side do we need figured out to do that? Okay, we want to implement advanced targeting. Do we need a CDP? How is it all going to work? Who do we need staff? So, that’s the sort of areas we’re getting to. We started the business just coming in and like, hey, I’m going to come do one piece of that puzzle. As we’ve grown, we’re now actually coming in and
90 00:17:42.730 ⇒ 00:18:01.489 Uttam Kumaran: being the entire data team, and similarly, I think on the AI side, we will start to come in as kind of like your automation team for hire. I think the AI side is actually probably the least defined, just because a lot of stuff is getting decided on in the past, you know, few years, and basically every few months now.
91 00:18:01.490 ⇒ 00:18:10.339 Uttam Kumaran: But a lot of… but the demand is incredibly high. So we’re talking to a lot of large businesses that maybe they’ve established an AI council, or they…
92 00:18:10.700 ⇒ 00:18:26.710 Uttam Kumaran: have tried to build some agents internally, but have not seen adoption, and we’ve developed sort of a framework for how we build, measure, and, like, build feedback loops into the way we develop AI, but also having a very modest set of goals, right? Yeah.
93 00:18:26.750 ⇒ 00:18:36.469 Uttam Kumaran: I feel like a lot of clients have never… they tend to talk to automation agencies or AI people that sell, like, this really grandiose future, and that’s not what we do, like, we…
94 00:18:36.820 ⇒ 00:18:48.730 Uttam Kumaran: I have to build systems that work, and so we come in and we really identify systems and processes that we can augment. And a lot of that is because I’ve tried to do it in my business, so I kind of share the grief.
95 00:18:48.730 ⇒ 00:19:00.099 Uttam Kumaran: But also, I know as a business owner what I needed to see to be convinced, and how AI is affecting our work, and so when we go to clients, it’s a… there’s some camaraderie there, you know, in the way we pitch.
96 00:19:00.570 ⇒ 00:19:01.700 Meg Bevilacqua: Yeah, and that’s…
97 00:19:02.400 ⇒ 00:19:07.060 Meg Bevilacqua: Honestly, it’s reassuring to hear, because when you see data and AI consultancy, that can mean a whole range of things.
98 00:19:07.060 ⇒ 00:19:07.450 Uttam Kumaran: Yes.
99 00:19:07.450 ⇒ 00:19:12.669 Meg Bevilacqua: And it’s… when you go in promising the moon with AI, you wind up with very unhappy clients.
100 00:19:12.670 ⇒ 00:19:13.170 Uttam Kumaran: Yes.
101 00:19:13.170 ⇒ 00:19:22.680 Meg Bevilacqua: It’s a hard conversation to have. It’s hard to go in there and say, look, you’re not going to be able to replace your entire customer support team in the next year.
102 00:19:22.680 ⇒ 00:19:33.440 Meg Bevilacqua: That’s not gonna happen. What you can do is do, like, help it guide the initial screening so you can reduce it, and you can make the team that you have more effective, but you can’t get rid of everybody.
103 00:19:33.440 ⇒ 00:19:49.669 Uttam Kumaran: Yeah, and I’m also… I’m here to build things that work, and that build a long-standing client engagement, so we don’t engage with clients who bring us on where they’re like, we want to fire everybody. That’s not, like, a healthy thing I want to be involved in, and also, expectations are really wide.
104 00:19:49.670 ⇒ 00:19:55.469 Uttam Kumaran: And just because it’s a growing industry, and there’s a lot of executives that are reading what’s in the media versus, like.
105 00:19:55.470 ⇒ 00:20:01.880 Uttam Kumaran: what’s on the ground, and so there’s a huge education component. There’s also just us being able to
106 00:20:01.880 ⇒ 00:20:16.099 Uttam Kumaran: to look at our business and find opportunities and go to market, because a lot of clients will call us and say, like, we need AI, but I don’t know where to implement this. So there also is a huge strategy component. So one of the things that we did recently is we’re going to market with a workshop.
107 00:20:16.110 ⇒ 00:20:35.729 Uttam Kumaran: You know, I know for big consultants, it’s typically, like, a loss leader service, which just gets people in the room, agrees on a couple of things, and they’re like, well, come do it with us. Very similarly, I think we’re trying to build a workshop offering, but actually do it where it does lead to really great conversations and outcomes.
108 00:20:35.730 ⇒ 00:20:43.169 Uttam Kumaran: And entire time in this business, I’ve never sort of said, like, you have to do this with us, or it’s never gonna work.
109 00:20:43.220 ⇒ 00:21:02.379 Uttam Kumaran: I’m like, we want to guide you this. You feel free to shop it around, and I want to show that we are the best choice every time. And so, for us, like, having great people and having great experiences, to facilitate those conversations, get companies to a couple of proof of concepts that they want to work on, and then us just being in the room and can…
110 00:21:02.380 ⇒ 00:21:21.659 Uttam Kumaran: execute that is something that we’re seeing a lot of success in, which is all kind of on the strategy side. And I think a lot of our work starts off as strategy in one way or another, either audit or a strategy, where we come in. Typically, for some clients, we come in for two to four weeks. We meet all the characters, understand all the systems, look at all the costs.
111 00:21:21.780 ⇒ 00:21:39.320 Uttam Kumaran: we either… we have some, like, proof of delivery, where we generate a report, or we fix a couple things, but in addition, we generate, like, a 3-6 month roadmap, diagrams, so we give them a slew of information, and at a pace at which I don’t think
112 00:21:39.320 ⇒ 00:21:50.309 Uttam Kumaran: they’ve ever seen from a consultancy, and then make it really obvious what the next steps are, and then say, hey, here’s our price that we would come do for you. And that’s been a great ability for us
113 00:21:50.310 ⇒ 00:21:55.949 Uttam Kumaran: To come in and avoid, like, sticker shock, but also show that, like, we’re the right partners for them to
114 00:21:56.020 ⇒ 00:21:58.770 Uttam Kumaran: you know, build a solution or improve something. Yeah.
115 00:21:58.770 ⇒ 00:22:11.439 Meg Bevilacqua: Yeah, that initial sort of strategy engagement is very aligned with the work that I did, the type of work I was doing at Deloitte. Basically, you come in, you figure out what the lay of the land is, and then you come up with recommendations, and…
116 00:22:11.540 ⇒ 00:22:19.659 Meg Bevilacqua: Typically, you kind of, like, leave that and let them go and think for a little bit, and then you come back, like, 6 months later and actually do whatever they decided to fit in their budget.
117 00:22:19.660 ⇒ 00:22:32.540 Uttam Kumaran: Yeah, so tell me, like, kind of what you’re thinking about these days, like, what is interesting? Like, I’d love to hear, like, even reflection. Anyone who I… who works… I know works at Deloitte or EY, I always ask, like, what do you think
118 00:22:32.940 ⇒ 00:22:43.189 Uttam Kumaran: after working there and, like, even speaking with a firm like ours that’s really small and probably very different, like, I’m just interested in your thoughts and reflections.
119 00:22:43.530 ⇒ 00:22:45.240 Meg Bevilacqua: So, I…
120 00:22:45.570 ⇒ 00:22:57.440 Meg Bevilacqua: consider my time at Deloitte a success, even though I got laid off. I learned a ton, I made… I had a lot of… built a lot of really good skills, I had a bunch of really good colleagues. I think… and it’s…
121 00:22:57.740 ⇒ 00:23:01.850 Meg Bevilacqua: you know, I walked you through sort of this, like, the step stairs of my,
122 00:23:01.850 ⇒ 00:23:02.490 Uttam Kumaran: Fantastic.
123 00:23:02.490 ⇒ 00:23:08.139 Meg Bevilacqua: career progress, right? I went from a 5-person company to, like, a 200-person company to a 70,000-person company.
124 00:23:08.140 ⇒ 00:23:08.650 Uttam Kumaran: Yes.
125 00:23:08.650 ⇒ 00:23:16.100 Meg Bevilacqua: And there’s a lot of advantages that come with working with a 70,000 person company. There’s also a lot of…
126 00:23:16.860 ⇒ 00:23:18.479 Meg Bevilacqua: In turn… it’s easy to get…
127 00:23:19.320 ⇒ 00:23:22.719 Meg Bevilacqua: Not to get lost exactly, but it’s easy to…
128 00:23:24.270 ⇒ 00:23:37.379 Meg Bevilacqua: You wind up in a much more systematized and much less nimble environment, and there’s a lot of… a lot less ability to sort of chart where you want to go and what you actually are interested in doing.
129 00:23:37.590 ⇒ 00:23:38.560 Meg Bevilacqua: …
130 00:23:38.730 ⇒ 00:23:56.129 Meg Bevilacqua: And that’s one of the reasons why, now when I’m on the job hunt, I’m looking at, again, at sort of the smaller end of the scale. I think that there’s a lot of opportunity there, there’s a lot of interest, and there’s a lot of ability to sort of, frankly, define your own job a little bit better than if you’re working for a big corporation.
131 00:23:56.310 ⇒ 00:23:57.879 Meg Bevilacqua: Which is exciting.
132 00:23:58.010 ⇒ 00:24:08.329 Meg Bevilacqua: In terms of what I’m thinking about in the marketplace, I think that if you work… if you’re interested in data, and you’re not saying that AI is top of mind right now, you’re lying.
133 00:24:08.330 ⇒ 00:24:08.840 Uttam Kumaran: Yeah.
134 00:24:08.840 ⇒ 00:24:09.990 Meg Bevilacqua: So…
135 00:24:10.900 ⇒ 00:24:23.890 Meg Bevilacqua: when I was, … and, like, the last couple of months when I was at Deloitte, I was doing a lot of conversa… having a lot of conversations and taking a lot of, sort of, thought piece looks at how to do… how… basically how to make AI
136 00:24:24.100 ⇒ 00:24:36.459 Meg Bevilacqua: pay, and you mentioned something there where it’s, like, companies keep sending out AI pilots and not seeing adoption, and I think you hit the nail on the head there, right? So AI is a…
137 00:24:37.390 ⇒ 00:24:41.939 Meg Bevilacqua: Hammer that all of the executives are carrying around trying to find the right nails.
138 00:24:42.280 ⇒ 00:24:42.670 Uttam Kumaran: Yes.
139 00:24:42.670 ⇒ 00:24:53.330 Meg Bevilacqua: And I think that if there’s a real market in helping them do that, and helping them identify the use cases where it can actually be a differentiator, as opposed to the ones that are all hype.
140 00:24:53.610 ⇒ 00:24:58.099 Meg Bevilacqua: And… I think that…
141 00:24:58.110 ⇒ 00:25:06.430 Meg Bevilacqua: the one… the ones that I’m most excited about are sort of the more focused and specific LLMs. I think that you…
142 00:25:06.430 ⇒ 00:25:18.290 Meg Bevilacqua: you can’t just toss ChatGPT at a problem and call it good, right? If you have a robust… if you have a good, robust data set, or if you have the possibility of a… possibility of a good, robust data set if you do the right cleansing and transformations.
143 00:25:18.290 ⇒ 00:25:26.929 Meg Bevilacqua: that you can train something on. I think that there’s a lot of opportunities in those types of engagements. One of the ones we were looking at specifically was, …
144 00:25:27.210 ⇒ 00:25:43.369 Meg Bevilacqua: on-device tech support for remote field engineers, right? So if you’ve got somebody who’s in a place with zero internet connection, just a laptop, you want to be able to have a local on-device LLM that they can use to help them process the
145 00:25:43.370 ⇒ 00:25:51.429 Meg Bevilacqua: Like, a couple of thousand pages of technical documentation on all of the different, systems that they might be called upon to use.
146 00:25:51.610 ⇒ 00:25:54.990 Meg Bevilacqua: And that’s the kind of… not…
147 00:25:55.460 ⇒ 00:26:00.689 Meg Bevilacqua: like, the very narrow, but actually useful use case for AI that I get excited about.
148 00:26:00.920 ⇒ 00:26:11.049 Uttam Kumaran: But, you know, that would have taken how many people to arrive on that, and, like, how many steps to get there, versus, yeah, we can just chat over documents. Like, it’s a really nice, …
149 00:26:11.170 ⇒ 00:26:27.000 Uttam Kumaran: like, demo, but beyond that, it’s not effective. And so we run evals for everything, we measure, you know, response time, we also do, like, leaderboards of, like, which users are using it, interview them, like, build the actual product feedback cycle.
150 00:26:27.320 ⇒ 00:26:42.119 Meg Bevilacqua: Yeah, and that adoption piece, I feel like, is critical, especially when you’re looking… like, and I’m gonna switch gears a little bit back to a different thing, the marketing engagement that I mentioned, because adoption there was a key… was a crucial metric that we were looking at, right? Because
151 00:26:42.460 ⇒ 00:26:58.820 Meg Bevilacqua: we were dealing with a bunch of teams that were used to spinning up their own analytics, and there was a lot of concern about, sort of, like, gray technology, right? But people do… they’re… them doing that again, right? And not telling us about it, and not using it. So we were paying very close… we had a…
152 00:26:58.990 ⇒ 00:27:11.699 Meg Bevilacqua: basically a health of the program dashboard that we were looking at for each of the major dashboards. Like, what’s the usage rate? Is it spiking when we think it’s supposed to spike? Like, right before, like, quarterly or monthly business meetings?
153 00:27:11.700 ⇒ 00:27:15.950 Uttam Kumaran: Are we seeing the focus on stuff that we need?
154 00:27:16.030 ⇒ 00:27:24.100 Meg Bevilacqua: And there were some of them that just weren’t getting adopted. And when we went in and had the conversations with the users, they’re like, well, it’s like…
155 00:27:24.220 ⇒ 00:27:30.659 Meg Bevilacqua: it’s too much, right? There’s… it doesn’t… there’s… it’s got… given me all of the information, and it’s not what I actually need to know.
156 00:27:30.790 ⇒ 00:27:35.450 Meg Bevilacqua: So I worked with some very talented,
157 00:27:35.960 ⇒ 00:27:40.530 Meg Bevilacqua: data analysts and a junior to put together a
158 00:27:40.530 ⇒ 00:27:59.700 Meg Bevilacqua: basically a full suite of, spoon-feeding reporting, right? So, if you were a marketer, you’d get an email every, like, week with, here’s all of the stuff for your specific territory, that was basically just pulling on the back end, like, a number of filters for each marketer based on their territory assignments and everything else.
159 00:27:59.700 ⇒ 00:28:08.670 Meg Bevilacqua: So they would get exactly what they needed to know, and then there’d be, of course, the link to the dashboard being like, if you want to know more, or if you want to dig into this, here it is.
160 00:28:08.880 ⇒ 00:28:12.089 Meg Bevilacqua: And that sort of, like, full-scale, like, adoption piece.
161 00:28:12.950 ⇒ 00:28:22.629 Meg Bevilacqua: helped juice the numbers, right? Because once they could see how it could be relevant to them, they were able to kind of, like, okay, maybe I’ll dip my toes in a little bit more.
162 00:28:22.630 ⇒ 00:28:36.730 Uttam Kumaran: Yeah. Now, we’re seeing in several clients where we’ve deployed things, the adoption piece is the number one problem. I don’t think we’ve… I don’t… I think the technical piece is difficult in building the systems, but we’re often not asked for things that
163 00:28:36.960 ⇒ 00:28:54.960 Uttam Kumaran: that I haven’t seen, or that we haven’t built ourselves. It’s purely, like, how do we convince people that this isn’t taking their job, that the first time they used it and it was wrong, that they should give it another chance, because we listened to their feedback, and that we’re actually scoring things, and there is a place for, sort of, human-in-the-loop feedback.
164 00:28:54.990 ⇒ 00:29:08.060 Uttam Kumaran: So all those things are really, really important to us, and those are the conversations that we’re having. You know, I don’t think we found a specific use case that’s, like, works for every company, but, you know, when I think about my company.
165 00:29:08.060 ⇒ 00:29:17.199 Uttam Kumaran: you know, I was… we do a weekly AI meeting where I… we sort of present the whole company things that we’ve worked on, and ways where each individual is using AI, and…
166 00:29:17.200 ⇒ 00:29:28.630 Uttam Kumaran: I said, look, we’re going across sales, finance, ops, engineering, project management, and we are breaking down what are the core, areas of responsibility or procedures that happen.
167 00:29:28.630 ⇒ 00:29:38.329 Uttam Kumaran: And we are starting to, one, build just SOPs for what exists today, and then starting to do the AI augmented piece, right? And that’s how systematic we’re gonna go, and
168 00:29:38.470 ⇒ 00:29:43.799 Uttam Kumaran: I mean, there’s also, like, what I would describe as, like, levels of automation, so I think about it a lot like
169 00:29:43.810 ⇒ 00:30:02.199 Uttam Kumaran: the level of automation for, like, self-driving cars, you know, they have, like, this L1, L2, L3. The baseline is just, like, I have a… I just asked ChatGPT. The next version is, I have a prompt, right? That, like, pretty much works. Then it’s like, okay, but the problem is I have to always drag in these documents. Okay, you need some type of workflow where
170 00:30:02.200 ⇒ 00:30:15.739 Uttam Kumaran: you can get that thing in. And then the next level is also, like, I need to get it somewhere, right? Maybe you’re not copying and pasting anymore, or it’s a structured output. Okay, so there’s these, like, cascading levels, but it has to start at foundationally understanding
171 00:30:15.740 ⇒ 00:30:33.179 Uttam Kumaran: what is the existing process, you know, before we can start to look at a lot of that. But writing the prompt is usually the easiest part. It’s like getting accents, it’s getting it out of people’s heads, it’s making sure that, like, once they tried it, and they’re like, it doesn’t do this, I’m like, well, you never mentioned to us that that’s even needed.
172 00:30:33.180 ⇒ 00:30:39.850 Uttam Kumaran: You know, and so it’s building that camaraderie. And then the last piece, you know, and I gave a… I actually gave a talk about this last week, is
173 00:30:39.850 ⇒ 00:30:46.980 Uttam Kumaran: where I want to find out how these continue to emerge in our business, AI and data. And so, for us, it’s decision making.
174 00:30:46.980 ⇒ 00:30:48.040 Meg Bevilacqua: And so….
175 00:30:48.040 ⇒ 00:31:02.260 Uttam Kumaran: I don’t know if there’s, you know, I think we’re really well poised to solve this problem, which is allowing AI to actually assist, our data analysts and our operators in our company to actually make decisions.
176 00:31:03.450 ⇒ 00:31:21.410 Uttam Kumaran: This isn’t just, like, tell me how many orders I got yesterday and, like, doing text to SQL. This is actually ingesting meeting notes, Slack, using the internet, and the structured data from queries to actually act as a co-pilot for revenue analysis, contract analysis.
177 00:31:21.410 ⇒ 00:31:32.930 Uttam Kumaran: for marketing analysis, and I think you’re seeing some of these happen in products, where they’re saying, we have an AI data analyst, but they lack all this rich context that is not in the…
178 00:31:33.210 ⇒ 00:31:52.829 Uttam Kumaran: CSV. That’s just not there. But as consultants, I… we’ve now gone through all those meetings, we document all of that, we have access to all that data, because we have to ask those questions, and that can now be provided as really rich context to an agent that, in addition to having, you know, the
179 00:31:52.940 ⇒ 00:31:56.309 Uttam Kumaran: The data can actually probably answer some of these questions. So.
180 00:31:56.370 ⇒ 00:32:07.820 Uttam Kumaran: that’s where I kind of think our worlds are going to start to mix, which is, like, how, as part of our data work, is the dashboard not actually the final frontier? Like, I think the dashboard is usually the
181 00:32:07.820 ⇒ 00:32:18.000 Uttam Kumaran: it sucks. It’s like, it gets you all almost the way there, but then if your executive doesn’t like it, it’s very subjective, it’s just not as good as an interface as
182 00:32:18.000 ⇒ 00:32:24.060 Uttam Kumaran: Asking a question and… and exploring, you know, and so that’s sort of what word….
183 00:32:24.060 ⇒ 00:32:24.580 Meg Bevilacqua: Yeah.
184 00:32:24.580 ⇒ 00:32:25.870 Uttam Kumaran: lines to think about.
185 00:32:26.150 ⇒ 00:32:33.980 Meg Bevilacqua: I know, and it’s… it’s one of those things where it’s a really… Tempting idea. It’s so clear.
186 00:32:33.980 ⇒ 00:32:34.650 Uttam Kumaran: So tempting.
187 00:32:34.650 ⇒ 00:32:38.630 Meg Bevilacqua: It’s really… It’s… there’s so many….
188 00:32:38.890 ⇒ 00:32:40.089 Uttam Kumaran: And it’s too….
189 00:32:40.090 ⇒ 00:32:47.440 Meg Bevilacqua: Challenges in trying to walk people through what it really means to ask a question.
190 00:32:47.440 ⇒ 00:32:48.230 Uttam Kumaran: Yes, yes.
191 00:32:48.230 ⇒ 00:32:53.309 Meg Bevilacqua: Because… and that’s, that’s, it’s, like, we’ve… we played around with this a little bit in… at… at the.
192 00:32:53.310 ⇒ 00:32:54.000 Uttam Kumaran: Right.
193 00:32:54.000 ⇒ 00:33:06.130 Meg Bevilacqua: And it’s… it’s so… it’s so much fun to watch people do this, and to see how people’s brains are working on it, because they’ll ask, you know, hey, what, how many, like, MQLs did I have?
194 00:33:06.130 ⇒ 00:33:06.930 Uttam Kumaran: Yes.
195 00:33:06.930 ⇒ 00:33:12.450 Meg Bevilacqua: But if you don’t specify a time frame, if you don’t specify a region, if you don’t specify… like, you have to have….
196 00:33:12.450 ⇒ 00:33:14.369 Uttam Kumaran: A lot of input validation, yeah.
197 00:33:14.370 ⇒ 00:33:24.299 Meg Bevilacqua: Yeah, that an executive has when they’re asking a question in a given meeting, it can be narrowed down. So, and if you don’t have that
198 00:33:24.530 ⇒ 00:33:27.200 Meg Bevilacqua: As the context, when you give the output.
199 00:33:27.490 ⇒ 00:33:30.010 Meg Bevilacqua: Somebody just looks at that output and says, well, that’s wrong.
200 00:33:30.010 ⇒ 00:33:30.530 Uttam Kumaran: Yes.
201 00:33:30.530 ⇒ 00:33:38.349 Meg Bevilacqua: So you have to… when you have that output, you have to include, or you… it’s wise to include, hey, you know.
202 00:33:38.450 ⇒ 00:33:43.200 Meg Bevilacqua: the MQL is this, 4, and then you include all of the assumptions that it made.
203 00:33:43.510 ⇒ 00:33:45.509 Meg Bevilacqua: To get you to that point.
204 00:33:45.510 ⇒ 00:33:45.950 Uttam Kumaran: Yeah.
205 00:33:45.950 ⇒ 00:33:53.320 Meg Bevilacqua: Because if you don’t include that sort of back-end transparency, if you can’t build that in, then all somebody does is look at the numbers and say, well, that’s wrong.
206 00:33:53.320 ⇒ 00:33:53.790 Uttam Kumaran: Yeah.
207 00:33:53.790 ⇒ 00:34:01.659 Meg Bevilacqua: If you include the context, then they can say… you can look at it and be like, well, okay, that’s not what I was going for, and then they can refine.
208 00:34:01.660 ⇒ 00:34:14.720 Uttam Kumaran: quickly tweak, yeah. It’s also suggestions, right? It’s like, it’s a level one question is, like, how many, yeah, what are my MQLs? But it’s like, okay, I saw the spike, like, what were the contributing factors to that spike?
209 00:34:14.719 ⇒ 00:34:24.180 Uttam Kumaran: And then… but you almost want the AI to also start to think, if I was an executive, what would I ask? Okay, what is their seasonality? What was it last year? What was the same time last month?
210 00:34:24.179 ⇒ 00:34:36.719 Uttam Kumaran: And I just think that the… a lot of our work on the data modeling side ends up in this form factor of a dashboard that is just not the best. And I don’t think there was an alternative.
211 00:34:36.880 ⇒ 00:34:39.380 Uttam Kumaran: Other than just people being, like, good analysts.
212 00:34:39.380 ⇒ 00:34:39.790 Meg Bevilacqua: Yeah.
213 00:34:39.790 ⇒ 00:34:57.770 Uttam Kumaran: That’s always great, which is very hard to come by, but I think that there’s now some opportunity to use an LLM to augment that process. There’s a lot of our clients that are never going to be able to handle… to hire those, like, great, you know, analysts and bring them onto their team. What option do they have?
214 00:34:57.770 ⇒ 00:35:02.080 Uttam Kumaran: Like, yeah, we put together great dashboards for them, but they still just, like.
215 00:35:02.210 ⇒ 00:35:06.480 Uttam Kumaran: Don’t have the training, or don’t have the wherewithal, or the time, you know, to answer the.
216 00:35:06.480 ⇒ 00:35:07.810 Meg Bevilacqua: Ask the right questions.
217 00:35:07.810 ⇒ 00:35:22.510 Uttam Kumaran: And I’m trying to think about how we solve that, and I think we’re well poised because we have both of these sides of the equation, but I also think it’s more about where… where was the decision made? Like, for example, okay, we saw CPMs increase on a marketing campaign.
218 00:35:23.260 ⇒ 00:35:32.120 Uttam Kumaran: if you’re just gonna see that what? That’s gonna lead to you sending a Slack note to your CMO, some… somebody has to get up and, like, go find that out, when… when actually there was a meeting last week.
219 00:35:32.280 ⇒ 00:35:33.880 Uttam Kumaran: Where you decided to…
220 00:35:33.950 ⇒ 00:35:44.879 Uttam Kumaran: to take a risk on something, and that should have been the answer. And, like, why wasn’t that available as context? Like, why wasn’t that meeting transcript, or the notes, or the Slack messages associated with the decision
221 00:35:44.880 ⇒ 00:35:55.750 Uttam Kumaran: tied. The obvious… people say, oh yeah, someone should have, like, gone in and annotated that, and, like, that doesn’t happen. It’s just, these are really quality of life things that just don’t happen.
222 00:35:56.040 ⇒ 00:35:59.970 Uttam Kumaran: And I don’t know, I think there’s gotta be some brain, but this is, like, where…
223 00:36:00.200 ⇒ 00:36:12.149 Uttam Kumaran: I think we’re going to continue to do a lot of the AI work that’s going to be kind of non-data related. We’re going to do a lot of data work that still lives in data land, but I… we have to try to find a way to merge these and leave teams with
224 00:36:12.200 ⇒ 00:36:20.269 Uttam Kumaran: you know, an agent, or some type of interface where they can make decisions faster and understand the context behind them. Like, that for me is…
225 00:36:20.490 ⇒ 00:36:29.979 Uttam Kumaran: is something that’s really, really cool that I think we’re uniquely poised to work on in a services capacity, like, not in a product capacity where it’s a one-size-fits-all.
226 00:36:29.980 ⇒ 00:36:33.670 Meg Bevilacqua: Like, I think we have the ability to sort of make something that’s more bespoke, so….
227 00:36:34.040 ⇒ 00:36:39.709 Uttam Kumaran: That’s, like, my… what I’ve been thinking about a lot. I don’t know when we’re gonna get there, have the budget to….
228 00:36:39.710 ⇒ 00:36:40.070 Meg Bevilacqua: Yeah.
229 00:36:40.070 ⇒ 00:36:43.120 Uttam Kumaran: To invest in that, but we’re definitely thinking about that, so….
230 00:36:44.120 ⇒ 00:36:44.910 Meg Bevilacqua: Okay.
231 00:36:45.310 ⇒ 00:36:53.239 Meg Bevilacqua: All right, I know we’re a little bit over, and I don’t want to eat into your next meeting, but I really enjoyed this conversation, and I’d like to talk a little bit more.
232 00:36:53.240 ⇒ 00:37:11.220 Uttam Kumaran: 100%, yeah, if I can answer any questions, please let me know. Maybe I’ll send you a note, we can grab some more time to chat. Definitely curious on, like, kind of, you know, you mentioned sort of what you’re thinking about next, but also I’m happy to give you a lay of the land of, like, our company structure, the people involved, and, like, our OKRs, and where we’re headed, and so happy to share all of that.
233 00:37:11.680 ⇒ 00:37:12.680 Meg Bevilacqua: Okay, perfect.
234 00:37:12.680 ⇒ 00:37:13.710 Uttam Kumaran: Okay. Alright.
235 00:37:13.710 ⇒ 00:37:14.699 Meg Bevilacqua: It’s good to meet you.