Meeting Title: Uttam: AI-driven analysis and strategy Date: 2026-04-14 Meeting participants: Greg Stoutenburg, Advait Nandakumar Menon, Uttam Kumaran, Amber Lin, Jasmin Multani
WEBVTT
1 00:00:14.300 ⇒ 00:00:15.330 Greg Stoutenburg: Hey, Avate.
2 00:00:16.540 ⇒ 00:00:17.919 Advait Nandakumar Menon: Hey, Greg, how’s it going?
3 00:00:19.230 ⇒ 00:00:20.610 Greg Stoutenburg: Pretty good, how are you?
4 00:00:21.100 ⇒ 00:00:21.970 Advait Nandakumar Menon: Feeling good.
5 00:00:23.250 ⇒ 00:00:28.479 Greg Stoutenburg: I am hungry, which is one of my least favorite feelings, and .
6 00:00:28.670 ⇒ 00:00:29.490 Advait Nandakumar Menon: Alright.
7 00:00:29.490 ⇒ 00:00:39.529 Greg Stoutenburg: My girlfriend gives me a hard time about it. She’s like, she’s like, I’ve never met anyone who thinks so much about what you’re going to eat next, wants to tell people about what you ate last. I’m like, yeah.
8 00:00:43.430 ⇒ 00:00:44.290 Greg Stoutenburg: So I’m hungry now.
9 00:00:44.290 ⇒ 00:00:51.740 Advait Nandakumar Menon: I, myself, have a kind of a late lunch, so not yet there, but pretty sure in an hour or two, I might…
10 00:00:52.040 ⇒ 00:00:52.500 Greg Stoutenburg: Yeah.
11 00:00:52.500 ⇒ 00:00:54.349 Advait Nandakumar Menon: Be in the same state as you.
12 00:00:54.350 ⇒ 00:00:59.820 Greg Stoutenburg: Yeah, it’s no good. Yeah, and let’s see, I’ve got…
13 00:01:01.360 ⇒ 00:01:06.269 Greg Stoutenburg: Yeah, I’ve got the next hour and a half. No… yeah, I’ve got the next hour and a half back to back. I’m gonna die.
14 00:01:06.270 ⇒ 00:01:06.970 Advait Nandakumar Menon: Uno.
15 00:01:07.120 ⇒ 00:01:07.970 Advait Nandakumar Menon: Who knows?
16 00:01:08.240 ⇒ 00:01:09.499 Greg Stoutenburg: This is no good.
17 00:01:12.110 ⇒ 00:01:16.360 Greg Stoutenburg: Cool. You may have seen, I pinged,
18 00:01:16.960 ⇒ 00:01:22.920 Greg Stoutenburg: Nandika. I tagged Nanda to just get an update on that. Yep.
19 00:01:24.380 ⇒ 00:01:24.910 Advait Nandakumar Menon: Yeah.
20 00:01:28.450 ⇒ 00:01:29.409 Greg Stoutenburg: Hey there, Tom.
21 00:01:29.410 ⇒ 00:01:30.200 Uttam Kumaran: Hey, everyone.
22 00:01:31.290 ⇒ 00:01:32.120 Advait Nandakumar Menon: Notham.
23 00:01:36.930 ⇒ 00:01:37.740 Greg Stoutenburg: Hey, Amber.
24 00:01:38.530 ⇒ 00:01:39.720 Amber Lin: Hello!
25 00:01:40.020 ⇒ 00:01:40.670 Advait Nandakumar Menon: Hey.
26 00:01:40.670 ⇒ 00:01:41.370 Greg Stoutenburg: Oh.
27 00:01:48.900 ⇒ 00:01:57.070 Uttam Kumaran: Let me know if we’re waiting for anyone else, but mainly this is just gonna be, like, kind of a brainstorm session. I’m gonna play the role of,
28 00:01:57.790 ⇒ 00:02:03.070 Uttam Kumaran: Platform team, and talk a little bit about how I’m thinking about enabling
29 00:02:03.170 ⇒ 00:02:14.470 Uttam Kumaran: this group to, you know, use AI to just execute, you know, more and more complicated tasks, faster and more accurately.
30 00:02:14.640 ⇒ 00:02:16.679 Uttam Kumaran: So really, my goal is just to, like.
31 00:02:17.030 ⇒ 00:02:24.840 Uttam Kumaran: have a little bit of a brainstorm session. I’ve thought pretty deeply about this problem, and so I have a couple of, like.
32 00:02:25.380 ⇒ 00:02:33.919 Uttam Kumaran: okay, if you were able to do something like this, would it be helpful? But I actually wanted to start even just to, like, go around the room a little bit and
33 00:02:34.350 ⇒ 00:02:41.640 Uttam Kumaran: and have everybody turn off their, like, okay, what do I need to do today, brain? And reflect a little bit on, like.
34 00:02:41.760 ⇒ 00:02:59.700 Uttam Kumaran: okay, what is it that we’ve… you’ve done over the last 3-6 months in terms of data and strategy? And, like, where have you found, personally, opportunities for automation or using AI to augment? And then I can start to fill in the gaps. Does that make sense?
35 00:03:00.790 ⇒ 00:03:01.260 Amber Lin: Yep.
36 00:03:04.230 ⇒ 00:03:10.780 Uttam Kumaran: Okay, cool. Are we waiting for anybody else? Sorry, I’m just on my phone, I just have to head out the door, but…
37 00:03:11.420 ⇒ 00:03:13.260 Amber Lin: Nope. I think that’s on the team.
38 00:03:13.260 ⇒ 00:03:13.730 Greg Stoutenburg: There’s a lot.
39 00:03:14.510 ⇒ 00:03:15.190 Uttam Kumaran: Cool.
40 00:03:15.270 ⇒ 00:03:28.010 Uttam Kumaran: Yeah, so maybe, like, you know, I think probably Amber and Greg, you guys have, out of the group here, been here the longest, so maybe I’ll ask you guys just to reflect a little bit, you know, briefly for, like, a minute or two on, like.
41 00:03:28.040 ⇒ 00:03:46.200 Uttam Kumaran: the work that you’ve been doing across data and strategy, which I will, you know, loosely define as, like, gathering information, putting together presentations or decision reports, you know, understanding, you know, like, actually going deep into a subject area.
42 00:03:46.360 ⇒ 00:03:57.200 Uttam Kumaran: And extracting information to draw the client towards a decision, or, you know… So, I would love to hear you guys just reflect on, you know, where you found
43 00:03:57.450 ⇒ 00:04:04.819 Uttam Kumaran: you know, automation and AI to be helpful, and yeah, just, like, give your feedback so we can kind of set the stage.
44 00:04:06.480 ⇒ 00:04:07.270 Amber Lin: I see.
45 00:04:07.390 ⇒ 00:04:09.460 Amber Lin: Do you want to go first?
46 00:04:09.860 ⇒ 00:04:11.199 Greg Stoutenburg: No, I was gonna say, you can go first.
47 00:04:11.420 ⇒ 00:04:12.440 Amber Lin: Oh, okay.
48 00:04:12.810 ⇒ 00:04:28.610 Uttam Kumaran: I mean, it could literally be as, like… I want to actually keep… it could be as, like, good at… as narrow as, like, hey, I use speech-to-text now, and my workflows are faster. It could also be as advanced as, like, yeah, I have a skill that, like, loops through Snowflake, combines tables, and is, like.
49 00:04:28.740 ⇒ 00:04:37.150 Uttam Kumaran: trying to optimize for some metric, like, so I want you to be… feel free to be as specific or as broad as you want.
50 00:04:38.520 ⇒ 00:04:40.989 Amber Lin: Yeah, some…
51 00:04:41.350 ⇒ 00:04:59.229 Amber Lin: how AI works in my workflow comes in a few ways. So there’s stuff that’s with analysis, I think it’s a little bit less relevant for… for, I think, what our team is working on right now, and then there’s a lot of
52 00:04:59.370 ⇒ 00:05:02.819 Amber Lin: Dashboarding-related tasks.
53 00:05:02.980 ⇒ 00:05:04.820 Uttam Kumaran: I think both are relevant, both are relevant.
54 00:05:04.820 ⇒ 00:05:17.840 Amber Lin: Okay, okay. So, I think I’ll still start with the dashboard-related tasks, so that includes, okay, let’s explore what’s related in a database. Let’s see what the modeling
55 00:05:17.840 ⇒ 00:05:35.320 Amber Lin: looks like, and then let’s see how we can take that and build a dashboard. And each step of the chain requires… can have AI participate, and it can have AI participate better. So, the main tools I use right now is mostly
56 00:05:35.620 ⇒ 00:05:48.170 Amber Lin: cursor, which is linked to, say, the dashboarding tool and our database. And then that enables me to say, look at what modeling exists.
57 00:05:48.360 ⇒ 00:05:52.669 Amber Lin: in… In dbt to…
58 00:05:52.920 ⇒ 00:06:05.459 Amber Lin: say, quickly view the work that our DE team has done, so I don’t have to take up too much of their time and explain, hey, explain this field, to me when I can ask it to crochet.
59 00:06:05.520 ⇒ 00:06:18.290 Amber Lin: Other things include, say, given these metrics, or given this dashboard I’m trying to build, I do a first pass of, hey, suggest some…
60 00:06:18.430 ⇒ 00:06:35.729 Amber Lin: suggest what fields I need to use in a new model, and then it usually gives me some good suggestions to start off with, and then I do a manual parse before I send it to the DE team to say, hey, please build this model for me, and then I just take that to
61 00:06:36.160 ⇒ 00:06:37.739 Amber Lin: To build a dashboard.
62 00:06:39.260 ⇒ 00:06:46.509 Uttam Kumaran: Okay, so saying it back, there’s, like, one part of, like, hey, I’m… I’m… I need to interact with dbt in more of a…
63 00:06:46.710 ⇒ 00:06:50.910 Uttam Kumaran: seamless way, like, what metrics should I use? How are things defined?
64 00:06:51.130 ⇒ 00:07:04.809 Uttam Kumaran: Another thing is, like, how do you actually create new requirements for the analytics engineering team? So that’s both great. I also heard a little bit about, like, it’s easier to create dashboards, and it’s probably easier to just, like, you know.
65 00:07:04.880 ⇒ 00:07:12.660 Uttam Kumaran: even just a year ago or two years ago, you’d have to run these queries directly on Snowflake. You can run a lot of them now just from your IDE.
66 00:07:13.200 ⇒ 00:07:13.840 Amber Lin: Yeah.
67 00:07:14.720 ⇒ 00:07:15.720 Uttam Kumaran: Okay, cool.
68 00:07:17.720 ⇒ 00:07:19.220 Uttam Kumaran: I guess, Greg, do you want to go?
69 00:07:19.410 ⇒ 00:07:28.340 Greg Stoutenburg: Yeah, so, the things that I’ve done that have been, like, the most helpful are, like, preparing weekly update decks. Should I go into that sort of thing, or is that too operational?
70 00:07:28.340 ⇒ 00:07:38.990 Uttam Kumaran: Sure, sure, yeah, yeah, yeah. I’m more interested in, like, how you guys are using it to impact your time. Whichever way we want to take it, it’s all relevant.
71 00:07:38.990 ⇒ 00:07:52.989 Greg Stoutenburg: Yeah, okay. Yeah, I… I had been using Cursor for deck updates, but then, you know, still ended up having to copy and paste everything. So, as of a few weeks ago, I started using just my own Cloud Pro account, because the co-work feature will actually
72 00:07:52.990 ⇒ 00:08:08.949 Greg Stoutenburg: make it so that I could just… I just took the most recent… I’ll take the most recent deck, give it to co-work, and then say, look at my linear, where I’ve got connected to Notion MCP and Linear MCP, and Slack MCP, and, say, you know, give me an update based on the last week’s work.
73 00:08:08.950 ⇒ 00:08:13.920 Greg Stoutenburg: And then when it creates the deck, I just… I then upload it back to…
74 00:08:13.920 ⇒ 00:08:23.600 Greg Stoutenburg: slides and, ask everyone to weigh in. And that has just sped me up so much by taking that extra step. I’m like, I’m… I’m happy to pay for tokens for this.
75 00:08:24.080 ⇒ 00:08:29.529 Greg Stoutenburg: So, that’s been valuable. Would love to be able to keep everything in the cursor environment.
76 00:08:29.530 ⇒ 00:08:29.850 Uttam Kumaran: Sure.
77 00:08:29.850 ⇒ 00:08:33.419 Greg Stoutenburg: I’m sure the day will come that they, you know, they have the same connectors.
78 00:08:33.490 ⇒ 00:08:52.359 Greg Stoutenburg: Okay, so that’s one. And then I think the places where AI has sped me up the most, that’s actually made me go, like, wow, I’m really impressed here, are on, are on things that had to do with product analytics, event tracking plans, and also with, I created, like, an experimentation framework for Eden.
79 00:08:52.360 ⇒ 00:09:05.089 Greg Stoutenburg: Now, in both cases, it took a lot of… it took a lot of filling in context that doesn’t live in the vault, and I think that’s… that’s okay, but, like, we’re heading more in the direction of having playbooks for these… these things. In the case of Eden.
80 00:09:05.510 ⇒ 00:09:18.209 Greg Stoutenburg: I… I got a whole bunch of information out of their Monday board. I got, a whole bunch of information from Mixpanel and from VWO. I…
81 00:09:18.210 ⇒ 00:09:32.559 Greg Stoutenburg: fed in guidelines from Intercom on rice scoring for prioritization, and then said, like, give me a two-page, sort of, like, philosophical rationale on why we’re gonna start
82 00:09:32.560 ⇒ 00:09:38.880 Greg Stoutenburg: roadmapping experiments in a certain way, and then also, like, give me a 3-month roadmap.
83 00:09:38.880 ⇒ 00:09:58.150 Greg Stoutenburg: And I ended up with this, like, 14-page document. I was like, okay, this is too much. So, you know, had to delete a bunch of stuff, had to work with, Claude a little bit, but ended up with something that I was pretty much able… in the end, I was able to copy and paste it into Notion and share it with Eden, and they loved it, and they’ve been using it since. So, that’s something that, if I had done that on my own, would have taken…
84 00:09:58.150 ⇒ 00:10:03.129 Greg Stoutenburg: you know, in the old days, honestly, like, that would have been, like, a few days’ work.
85 00:10:03.910 ⇒ 00:10:12.129 Greg Stoutenburg: For an event tracking plan, there’s still a whole bunch of manual effort that has to go into really understanding a product and how a user would interact with it.
86 00:10:12.130 ⇒ 00:10:37.089 Greg Stoutenburg: But once that part of it is done, I was able to use, Amplitude’s best practices for product analytics and client context that exists in the vault, and then, just use AI and a template and go, alright, with this information I’ve given you, create an event tracking plan. And just got… and, like, it was good. It was pretty good. I had to delete a bunch of stuff, it created, like.
87 00:10:37.090 ⇒ 00:10:38.310 Greg Stoutenburg: Too much stuff.
88 00:10:38.310 ⇒ 00:10:49.970 Greg Stoutenburg: But it also captured everything, and it even captured things that I would have missed, because, like, so I did this for Global VetLink, because, you know, I don’t really… I don’t know that much about
89 00:10:50.170 ⇒ 00:10:55.790 Greg Stoutenburg: stuff vets do, you know? But, but, but Claude did, so…
90 00:10:56.130 ⇒ 00:10:59.579 Greg Stoutenburg: Those are some places where I use AI,
91 00:11:00.290 ⇒ 00:11:12.819 Greg Stoutenburg: Consistently. I mean, much of the rest of the time, like, my cursor is gonna show me asking questions, it’s gonna be planning docs, it’s gonna be looking for updates and things like that, you know, like, creating that weekly update and things like that.
92 00:11:14.740 ⇒ 00:11:32.959 Uttam Kumaran: Anyone else have any areas not covered there? So, Greg talked about presentation, creation, plan creation, he’s talked about filling in knowledge where maybe we don’t have the subject matter expertise. You know, you nicely teed me up for some stuff I’ll talk about in terms of, like.
93 00:11:33.000 ⇒ 00:11:42.529 Uttam Kumaran: a different sort of interface instead of cursor for interacting with sort of some of this and integrations. That makes a lot of sense. Anyone else have anything?
94 00:11:44.070 ⇒ 00:11:52.759 Jasmin Multani: I think in the past, in previous companies, what’s been helpful is, benchmarking the client’s
95 00:11:53.880 ⇒ 00:11:57.390 Jasmin Multani: Data and benchmarking it against industry practice.
96 00:11:57.510 ⇒ 00:12:02.989 Jasmin Multani: I feel like we will eventually have this issue come up in Element, because…
97 00:12:03.150 ⇒ 00:12:22.440 Jasmin Multani: one senior production manager is about to leave, and the other production manager just started a month ago, and even with the questions I directly asked the client, the one who’s new, I asked for, like, benchmarks, like, hey, how often do you have to change this? And, I could tell, like, she was just…
98 00:12:23.300 ⇒ 00:12:29.560 Jasmin Multani: nervous to be put on the spot. So, pulling in industry benchmarks also helps when we’re flying.
99 00:12:29.560 ⇒ 00:12:30.140 Uttam Kumaran: Right.
100 00:12:30.360 ⇒ 00:12:48.610 Uttam Kumaran: That’s perfect. So another piece is industry benchmarking. It sort of goes into a little bit of, like, okay, research, broadly, of best practices, but this is something that, a lot of our clients ask us about, and in addition to, you know, our ability to say, like, hey, Brainforge has done this a bunch of times.
101 00:12:48.610 ⇒ 00:12:59.380 Uttam Kumaran: it’s also, I think, fairly… I feel that there’s a path for me to enable for you guys, more around research. And research is… I will broadly define as, like.
102 00:12:59.600 ⇒ 00:13:15.119 Uttam Kumaran: looking through internet, looking through Brainforge, like, what has Brainforge done, potentially, in the past? Like, you may find that there is a, like, someone you’re working with who worked on a client, like, 2 years ago, right? Or Robert and I did something, so how are you able to
103 00:13:15.120 ⇒ 00:13:20.720 Uttam Kumaran: research what we’ve done? Are you able to research benchmarks? And then how are you able to apply that
104 00:13:20.850 ⇒ 00:13:27.619 Uttam Kumaran: In a specific scenario, whether you’re preparing for a meeting, whether you’re looking at a piece of data, so that’s really helpful.
105 00:13:27.770 ⇒ 00:13:33.140 Uttam Kumaran: Maybe just to fill out a couple of other pieces. One is, like, I think,
106 00:13:33.400 ⇒ 00:13:49.049 Uttam Kumaran: being able to actually execute on work items is really, really great. Second is, I’ve always… I’m always gonna share that this team should be the one team that’s really focused on spending time changing the behavior of clients, and I want to get you guys as much out of, like.
107 00:13:49.050 ⇒ 00:14:00.300 Uttam Kumaran: spending time behind the IDE working as possible. Like, this crew really has the power and the leverage to go change the behavior and impact decisions that are happening.
108 00:14:00.300 ⇒ 00:14:07.810 Uttam Kumaran: the more time that this crew is spending, you know, like, futzing around with SQL, or, like, changing a line chart.
109 00:14:07.880 ⇒ 00:14:20.729 Uttam Kumaran: it’s… that’s not time well spent, that’s not money well spent by the client. So that’s always what I’m gonna push us towards. I want you to keep that in mind when I talk about automation, because it’s not automating this group.
110 00:14:20.760 ⇒ 00:14:40.049 Uttam Kumaran: out of a job or anything. Really, the automation is to move us toward doing the data work that we all know really matters, but is oftentimes not the majority of our time, right? Like, in past life, you may spend 80% of your time, like, doing the table on a PowerPoint instead of actually, like.
111 00:14:40.150 ⇒ 00:14:48.660 Uttam Kumaran: thinking about, I have an hour with the CEO, what am I supposed to say? So that’s what I want to really, really express. So a couple things that, like.
112 00:14:48.890 ⇒ 00:15:07.579 Uttam Kumaran: I’m working on on the platform team to help you guys, and I’ll give you some ideas. Really, what I would like to understand on… is on two levers. I want to know where, like, what would be most impactful in terms of time savings, and then what would be most impactful in terms of going above and beyond with the client.
113 00:15:07.580 ⇒ 00:15:19.579 Uttam Kumaran: So, like, I have those two axes in mind, which is, like, if I can deliver something for you this week that, like, okay, this shaves off an hour a week, that’s really helpful for me to know. But if you’re also, like, hey.
114 00:15:19.590 ⇒ 00:15:25.520 Uttam Kumaran: you know, if I can produce an industry benchmark report, about a specific, like.
115 00:15:25.710 ⇒ 00:15:31.360 Uttam Kumaran: Like, peace, and you can build me a skill that helps me do that, like.
116 00:15:31.530 ⇒ 00:15:48.320 Uttam Kumaran: although that… that maybe isn’t something we were already doing, this would help me take this client above and beyond. So, I want you guys to consider those axes, and I’m gonna… I’m gonna talk about strategy analytics as part of the supply chain of data at Brainforge. So, we’re moving from data engineering to modeling to then
117 00:15:48.480 ⇒ 00:15:55.850 Uttam Kumaran: data landing in March, and then y’all kind of take over from there. So, let’s first start with what Amber mentioned, which is the
118 00:15:55.920 ⇒ 00:16:09.490 Uttam Kumaran: relationship between, the relationship between the strategy team and the analytics engineers. So one is, like, to take Amber’s analogy further, I actually think you guys are very close to being able to write
119 00:16:09.820 ⇒ 00:16:24.629 Uttam Kumaran: dbt yourselves. Like, I actually think dbt is just an abstraction of SQL. You guys have the requirements on what a table needs to look like, what the data… where the data needs to be sourced from, and also have the requirements on, like, what accuracy looks like.
120 00:16:24.860 ⇒ 00:16:43.969 Uttam Kumaran: Those are all the inputs that typically go into writing dbt code, and I’ll tell you, I haven’t written dbt in Scratch in more than a year. So, I actually… one piece is I want to start to have you guys bleed in a little bit into the analytics engineering world, and start to take on those tasks. One, we are… we’re…
121 00:16:44.300 ⇒ 00:17:01.379 Uttam Kumaran: getting more… getting really sophisticated and successful analytics engineers is hard right now. There’s not a lot of them, and we’re having difficulty finding a lot of great ones, so the more that this team can take on, like, Advait, for example, like, I think you could have done that pylon work yourself.
122 00:17:01.420 ⇒ 00:17:12.760 Uttam Kumaran: Not saying you should have, but I think I could make it easier, and I could have shown you, okay, there’s some helpful skills for you to actually write dbt code, and I know you know that. I think another piece is also
123 00:17:12.760 ⇒ 00:17:23.429 Uttam Kumaran: changing, like, pieces of models, QAing, I want to make that a lot easier. That way, the analytics engineers can themselves go work on really tougher modeling challenges, right? Like.
124 00:17:23.430 ⇒ 00:17:32.489 Uttam Kumaran: Building cohort models, using window functions for sophisticated things, thinking about running models faster. So that’s, like, one piece.
125 00:17:33.610 ⇒ 00:17:40.799 Uttam Kumaran: I guess, like, does anyone want to reflect or give me, like, two cents on that? Like, what a series of skills that allow you to
126 00:17:41.190 ⇒ 00:17:49.649 Uttam Kumaran: you know, like, work on dbt or modeling be… be helpful? Like, would that… or would be sort of a nice-to-have?
127 00:17:51.270 ⇒ 00:17:59.660 Jasmin Multani: I think it’d be helpful. I think it would close the loop of depending on engineers, especially because we’re each in different time zones.
128 00:18:00.620 ⇒ 00:18:03.200 Amber Lin: Yeah, and bare minimum, I think.
129 00:18:03.360 ⇒ 00:18:19.600 Amber Lin: we can suggest a PR, because in order for me to tell them what to model, I know what… what they need to do. Like, I already know… I have to spell out that, I might as well put it in Kirscher and have them review the PR. Like, I’d rather do that.
130 00:18:19.760 ⇒ 00:18:22.239 Amber Lin: And I think it’ll just be so much faster.
131 00:18:22.610 ⇒ 00:18:27.419 Uttam Kumaran: And if I can ask you one thing, Amber, why haven’t you done it so… why haven’t you done it yet?
132 00:18:28.150 ⇒ 00:18:41.070 Amber Lin: Because it feels like a different field that I… I feel like if I do it, I might break some certain joints that I’m not aware of, so if the skill can…
133 00:18:41.820 ⇒ 00:18:57.600 Amber Lin: do the checks that a DE, a senior DE would… AED would do, that would be really helpful, because, like, I’ve had to go back and tell them, hey, this… this joint broke, and then this report broke, so…
134 00:18:57.850 ⇒ 00:19:03.149 Amber Lin: I need confidence that that won’t happen if I’m the one pushing the PR, because that’s responsibility.
135 00:19:05.820 ⇒ 00:19:25.369 Uttam Kumaran: Okay, makes sense. So there’s something around dbt. The second piece is, like, everything around deck creation, like, slide creation. I feel like that’s pretty self-explanatory. I think we… I think what I need is from, probably from Jasmine and Garrett, like, some sort of standards around decks.
136 00:19:25.490 ⇒ 00:19:37.479 Uttam Kumaran: And I’m more than happy to build skills that allow you to go from data or a plan to a deck pretty fast. I think we’re already pretty close to there. I think there’s probably some training required, and I think
137 00:19:37.600 ⇒ 00:19:52.540 Uttam Kumaran: Greg, you mentioned that you’re having some success with Cowork. Like, there’s some ways to emulate that same behavior in Cursor. We’re coming out with, like, our own version of Cowork here shortly that should hopefully take some of that on. So, heard you on slide creation.
138 00:19:53.390 ⇒ 00:19:59.670 Uttam Kumaran: that, I think, is a one that we’ll just… I feel like we’re pretty close to just nailing, and I think would take off a lot of time.
139 00:20:01.210 ⇒ 00:20:11.099 Uttam Kumaran: The other piece I want to talk about is actually just, like, analyzing data. And so the way I look at this is, like, hey, let’s say you’re just given a mart. How do you use…
140 00:20:11.100 ⇒ 00:20:23.629 Uttam Kumaran: like, text-to-SQL in Kersher to actually help you identify trends. I think, Amber, you running this for ABC was, like, the first show of us actually doing this across, like, a wide amount of domains.
141 00:20:23.760 ⇒ 00:20:31.430 Uttam Kumaran: And I think, you know, I think we saw some things that were really positive, but we also saw, like, there were some hiccups, you know?
142 00:20:31.620 ⇒ 00:20:36.250 Uttam Kumaran: Like, I think we were… we were basically trying to say, like, okay.
143 00:20:36.490 ⇒ 00:20:50.110 Uttam Kumaran: you were able to get to the data, but then maybe the charting was difficult, or the data actually, like, we just didn’t have all the definitions. And so one thing that I can help on is, like, how do you actually do structured analysis?
144 00:20:50.240 ⇒ 00:21:08.479 Uttam Kumaran: you know, via, like, an agentic system, right? A couple ways that I’m thinking about this is you actually don’t want… what I want people to get challenged by is that the AI should actually grill you a little bit on, like, what it is you’re trying to solve for, right? You guys should come with what is the business outcome that you’re trying to drive.
145 00:21:08.840 ⇒ 00:21:23.330 Uttam Kumaran: the AI, you want to take advantage as, like, imagine you had a team of 10 analysts that could go, like, identify opportunities. What happens commonly with AI is if you come to the table with, hey, I want to look at how revenue is affecting this number.
146 00:21:23.370 ⇒ 00:21:32.130 Uttam Kumaran: you’re… you’re setting yourself up to just experiment on one thing. Instead, if you’re like, hey, Element wants to grow revenue sustainably.
147 00:21:32.330 ⇒ 00:21:35.570 Uttam Kumaran: Like, let’s start with a brainstorm on just with that.
148 00:21:35.840 ⇒ 00:21:37.700 Uttam Kumaran: Given the data you have.
149 00:21:37.830 ⇒ 00:21:51.359 Uttam Kumaran: work with me on, like, what are some hypotheses that, you know, we can come up with. Then you can go through those hypotheses, pick a few, and then go deeper. Pick a few, go deeper. Do you see how that, like, kind of changes the
150 00:21:51.690 ⇒ 00:22:00.180 Uttam Kumaran: the way that we typically use AI in more of a brute force versus, like, actually, we’re gonna do, like, a broader exploration.
151 00:22:00.410 ⇒ 00:22:01.899 Uttam Kumaran: Does that make sense?
152 00:22:05.590 ⇒ 00:22:08.190 Amber Lin: An example, please?
153 00:22:08.480 ⇒ 00:22:14.320 Uttam Kumaran: Yeah, so, like, if you were to… if you were to look at Brainforge, right, and if I was to go to Brainforge,
154 00:22:14.440 ⇒ 00:22:34.419 Uttam Kumaran: if I was to go to Brainforge-related, data and say, hey, like, tell me, what our margin is, right? And just, like, show me what our margin is per product, per service, I may get an answer. But actually, what I’m trying to solve is, like, how do I expand Brainforge’s margin by 5%?
155 00:22:34.630 ⇒ 00:22:40.750 Uttam Kumaran: So instead, what I should do is I said, hey cursor, you have access to Snowflake, here’s all the data that’s in it.
156 00:22:40.980 ⇒ 00:22:46.079 Uttam Kumaran: My goal today is to try to find opportunities to expand Brainforge’s margin by 5%.
157 00:22:46.720 ⇒ 00:23:01.500 Uttam Kumaran: let’s work on a plan and a series of experiments and a series of analyses that we can run in this session to help me identify those. Do you see that that’s different than the old way, which is like, give me the margin and I’ll figure it out. Instead, you’re like.
158 00:23:01.680 ⇒ 00:23:03.330 Uttam Kumaran: Here’s the problem.
159 00:23:03.510 ⇒ 00:23:11.310 Uttam Kumaran: Now, I’m gonna steer… like, I’m gonna steer this series of analyses around the problem.
160 00:23:13.510 ⇒ 00:23:33.010 Amber Lin: Yeah, gosh, I think I tried similar things before when I was running more analysis-based. Giving it a goal does change it a lot, and having… using the plan feature does change it a lot, versus, like, single task-based executions,
161 00:23:33.110 ⇒ 00:23:42.479 Amber Lin: can… I just… it can get confused, or it might take certain routes that only completes a task, but it doesn’t really help you
162 00:23:43.080 ⇒ 00:23:47.260 Amber Lin: think about the next steps, but then sometimes I also realized, like.
163 00:23:47.810 ⇒ 00:23:54.059 Amber Lin: it might even be better to think about the plan than I think I would be, so…
164 00:23:54.060 ⇒ 00:24:13.599 Uttam Kumaran: Yes, that’s exactly it. What we’re good at is looking at a series of plans and making a judgment on what could be more effective than others. What I want to move us out of is coming up with the entire plan ourselves, and then just having AI execute. Like, AI you should start using as a thought partner, is the rough
165 00:24:13.920 ⇒ 00:24:15.830 Uttam Kumaran: Sort of, like, sell here.
166 00:24:18.310 ⇒ 00:24:22.100 Amber Lin: Yeah. I, I use it that way, so I, I think…
167 00:24:22.480 ⇒ 00:24:30.539 Amber Lin: The folks here also probably have been using it that way, so if you guys want to share how you use it, that would be interesting.
168 00:24:30.540 ⇒ 00:24:39.969 Uttam Kumaran: Yeah, does that… does that, like, help anybody, or are other… are folks using it in that way, or are people still using it more like… you know, I like to describe it as, like, a scalpel versus, like…
169 00:24:40.450 ⇒ 00:24:43.840 Uttam Kumaran: Actually, just, like, let me think through, like, the entirety of a plan here.
170 00:24:45.500 ⇒ 00:24:47.290 Greg Stoutenburg: I mean, that’s similar to what I did for Eden.
171 00:24:47.630 ⇒ 00:24:49.379 Greg Stoutenburg: In a way, right? Like…
172 00:24:50.590 ⇒ 00:24:51.090 Uttam Kumaran: Yes.
173 00:24:51.090 ⇒ 00:24:54.359 Greg Stoutenburg: map, you know? And then I… I mean, there were so many ideas, I deleted so much stuff.
174 00:24:54.770 ⇒ 00:24:55.120 Uttam Kumaran: Yes.
175 00:24:55.370 ⇒ 00:24:56.100 Greg Stoutenburg: Yeah.
176 00:24:56.100 ⇒ 00:25:01.880 Uttam Kumaran: But that’s fine, like, you would… that’s where we… that’s where I actually want you guys to be in the loop.
177 00:25:02.110 ⇒ 00:25:08.149 Uttam Kumaran: Is on qualifying the ideas. Yeah. But don’t limit yourself to just the ideas that we can come up with.
178 00:25:08.300 ⇒ 00:25:12.500 Greg Stoutenburg: Yeah, which is good, because, I mean, one of the things that’s been helpful is it kind of helped me, like.
179 00:25:12.500 ⇒ 00:25:31.979 Greg Stoutenburg: where I’ve had success in the past directly informs the ideas I’m gonna come up with in the first place. Like, some of the things that, AI had proposed around, like, remarketing and stuff, like, if I just wrote it without using… if I just sat down with a word processor and was like, write down a roadmap, remarketing would not have appeared on that.
180 00:25:31.980 ⇒ 00:25:38.210 Greg Stoutenburg: document, you know what I mean? So, but Claude suggested it. So, that was really helpful.
181 00:25:40.410 ⇒ 00:25:42.570 Uttam Kumaran: I guess Advay, Jasmine, does that make sense?
182 00:25:43.820 ⇒ 00:26:00.209 Advait Nandakumar Menon: Yeah, I have been using it in both ways, like, sometimes I might just lay out the plan or whatever and ask it to do it this way, and sometimes, if I’m trying to come up with some visual for the dashboard, I might ask it to give some ideas and pick the best one out of it myself, depending on
183 00:26:00.210 ⇒ 00:26:04.280 Advait Nandakumar Menon: my experience previously, what I worked on, and what seems best for that.
184 00:26:04.460 ⇒ 00:26:10.409 Advait Nandakumar Menon: particular dashboard, so that’s how I go about, like, I try to do it both ways sometimes, so…
185 00:26:10.870 ⇒ 00:26:12.580 Advait Nandakumar Menon: Yeah.
186 00:26:13.790 ⇒ 00:26:21.819 Jasmin Multani: I’ll pitch an example that I think could help everyone. As our dashboards mature.
187 00:26:22.120 ⇒ 00:26:32.120 Jasmin Multani: We’re gonna get clients asking us about alerting schedules, and escalations, alerting schedules are very, very manual, and they end up being very reactionary.
188 00:26:32.310 ⇒ 00:26:39.499 Jasmin Multani: So, which I know the clients will hate. So, in order to get, proactive,
189 00:26:40.630 ⇒ 00:26:56.170 Jasmin Multani: before, like, adoption, I think it’d be helpful would be, running an RCA example through an AI, and saying, like, hey, given these metrics, given this industry, put together some metrics.
190 00:26:56.170 ⇒ 00:26:59.259 Uttam Kumaran: What is an RCA, by the way, for the people in the building?
191 00:26:59.260 ⇒ 00:27:00.330 Advait Nandakumar Menon: Blue Cross.
192 00:27:00.770 ⇒ 00:27:03.220 Jasmin Multani: Okay, okay. That’s my buddy.
193 00:27:03.220 ⇒ 00:27:07.209 Uttam Kumaran: Sorry, that’s me. I don’t know. Okay, yeah, I’m familiar, I’m familiar.
194 00:27:07.210 ⇒ 00:27:11.820 Jasmin Multani: Yeah, yeah, that’s, like, the bread and butter for every analyst in every company. Right.
195 00:27:11.960 ⇒ 00:27:14.719 Jasmin Multani: And the way it works is that…
196 00:27:15.530 ⇒ 00:27:18.199 Jasmin Multani: I’ll give a DoorDash example, is…
197 00:27:18.290 ⇒ 00:27:31.820 Jasmin Multani: Hey, if there’s an increase in cancellation, but, the acceptance rate has flatlined, that tells… that’s gonna be like, okay, given these two metrics, what are the possible scenarios?
198 00:27:31.820 ⇒ 00:27:44.159 Jasmin Multani: Does that mean that, hey, there is no snowstorm? Sometimes there’s a weather report, that’s impacting acceptance rate, and driving up cancellation. Or maybe it’s like,
199 00:27:44.770 ⇒ 00:27:47.980 Jasmin Multani: the backend logic is not correctly,
200 00:27:48.620 ⇒ 00:27:59.659 Jasmin Multani: set up for the catalog. Given those two metrics, and whether which one is moving up, which one is moving down, and which one is flatlining, that helps curate, hey, out of…
201 00:27:59.980 ⇒ 00:28:03.839 Jasmin Multani: Given these two metrics, what are, like, the top two scenarios?
202 00:28:03.970 ⇒ 00:28:10.270 Jasmin Multani: Top 2 out of, like, 15 scenarios that could be driving, this spike.
203 00:28:10.390 ⇒ 00:28:16.300 Jasmin Multani: Let’s, like, put all hands on deck. That is especially helpful.
204 00:28:16.970 ⇒ 00:28:36.690 Jasmin Multani: And even, yeah, so things like Gemini, what they’ll do is they’ll give a few different scenarios, and they’ll… like, Gemini will very, eloquently, like, design a table that, tells you what to focus on, what is the purpose, and who… and where we would lean in and be like, hey.
205 00:28:37.060 ⇒ 00:28:44.199 Jasmin Multani: who are the escalation owners that need to resolve this? Who do we need to bring in? What is the lift, and so forth.
206 00:28:44.380 ⇒ 00:28:55.289 Jasmin Multani: So that’s an example of, like, going back and forth with the AI to, go from something that’s broad and narrow it down, and adapting it to our clients’ needs.
207 00:28:57.160 ⇒ 00:29:15.730 Uttam Kumaran: Yeah, so in that example, this is where, like, it would be helpful for me to get, like, what is the best practice, like, what is a great example of, like, a super thorough RCA? And so I’m putting my platform team hat on. My job is to deliver you a skill that allows you to run really great RCAs faster.
208 00:29:15.990 ⇒ 00:29:17.099 Jasmin Multani: Like, whatever.
209 00:29:17.100 ⇒ 00:29:18.049 Uttam Kumaran: See how that’s…
210 00:29:18.160 ⇒ 00:29:26.249 Uttam Kumaran: Yeah, so that’s… so that’s the exchange of, like, responsibilities here. What I need from you guys is the requirements, like.
211 00:29:26.410 ⇒ 00:29:40.919 Uttam Kumaran: what inputs do you need? But don’t… don’t worry about how it gets done. Because what you’re gonna see is skills in our company are gonna look more like products over time, where they’re gonna get maintained and updated.
212 00:29:41.030 ⇒ 00:29:49.060 Uttam Kumaran: and changed. For one, I don’t know… What was… So, wow.
213 00:29:49.670 ⇒ 00:30:01.499 Uttam Kumaran: Duck, Omni, whatever you need to then build… build that out, and do it in a manner where it, like, brings you into the loop, asks you questions as it goes, and then helps you produce that…
214 00:30:08.720 ⇒ 00:30:20.860 Jasmin Multani: Makes sense. On my end, I’m still putting on my Eden hat and my Element hat, and, I think throughout the week, when I have more downtime, I will put on my platform hat and…
215 00:30:21.310 ⇒ 00:30:22.870 Jasmin Multani: Ship ideas out.
216 00:30:24.530 ⇒ 00:30:25.220 Uttam Kumaran: Okay.
217 00:30:27.900 ⇒ 00:30:28.480 Uttam Kumaran: Oh.
218 00:30:28.970 ⇒ 00:30:48.290 Uttam Kumaran: Okay, this was really helpful. I think I have, like, enough idea about where I can start to cement some skills for each of you on this area. So let me take this on and, like, start to build out a little bit of a path towards, like, I heard deck building, I heard getting more into dbt, I heard skills for specific analytics outcomes, like RCA, so I can help
219 00:30:48.310 ⇒ 00:30:52.940 Uttam Kumaran: you know, kind of take it out on my side, like, some of the integrations that I’ll need to build.
220 00:30:53.560 ⇒ 00:31:02.449 Uttam Kumaran: And then, Jasmine, I can partner with you on, like, if some of those are standardized, like, we have a standard deck, we have a standard RCA template, that’s what I’ll need to sort of deliver.
221 00:31:02.450 ⇒ 00:31:13.809 Uttam Kumaran: on, like, the final outcome, so this is really helpful. I’m… I feel more than comfortable getting you this, and I think where I want it to continue to go deeper is I want to try to solve more problems that seem like the RCA problem.
222 00:31:13.940 ⇒ 00:31:24.879 Uttam Kumaran: You know, where you’re actually driving towards a sophisticated work output using a series of skills, and a series of AI asking us questions, you know? So let me make that happen.
223 00:31:26.220 ⇒ 00:31:27.359 Jasmin Multani: Mmm, thank you.
224 00:31:27.780 ⇒ 00:31:39.480 Jasmin Multani: I have to drop for another meeting, but, Otham, let us know where you want us to document these asks, and, like, this… these brainstorming, as the week goes on, as we catch things.
225 00:31:40.400 ⇒ 00:31:41.509 Uttam Kumaran: Okay, okay, perfect.
226 00:31:41.800 ⇒ 00:31:43.269 Jasmin Multani: Thank you, I gotta go. Bye.
227 00:31:43.270 ⇒ 00:31:44.980 Uttam Kumaran: Okay. Thank you. Bye, guys.
228 00:31:44.980 ⇒ 00:31:45.850 Amber Lin: Thanks, Bike.
229 00:31:45.850 ⇒ 00:31:47.250 Advait Nandakumar Menon: Thank goodness. Bye-bye.