Meeting Title: Brainforge x UrbanStems: Forecast Project Proposal Date: 2026-03-17 Meeting participants: Zack Gibbs, Uttam Kumaran, Robert Tseng
WEBVTT
1 00:00:24.050 ⇒ 00:00:24.720 Uttam Kumaran: Exactly.
2 00:00:25.060 ⇒ 00:00:27.330 Uttam Kumaran: Sorry, it’s the last thing just ran a little over.
3 00:00:27.850 ⇒ 00:00:30.169 Zack Gibbs: It’s all good. Maddie’s… I just hopped on.
4 00:00:31.310 ⇒ 00:00:32.330 Uttam Kumaran: How’s everything?
5 00:00:32.960 ⇒ 00:00:35.289 Zack Gibbs: It’s going alright. How are you doing?
6 00:00:35.660 ⇒ 00:00:36.979 Uttam Kumaran: Good, good.
7 00:00:37.130 ⇒ 00:00:40.419 Uttam Kumaran: Yeah. How was, you were just on vacation?
8 00:00:41.240 ⇒ 00:00:57.090 Zack Gibbs: Yeah, we, every year… every year, my brother comes out from Chicago with his two kids, his two boys that are older, than my kids, and we all go skiing, so we were… we were in Steamboat… Steamboat Springs, had a nice, had a nice time, so…
9 00:00:57.370 ⇒ 00:01:07.319 Zack Gibbs: But yeah, just everything’s stacked up from that week that I was gone to last week, so a little bit less busy this week, but still quite busy.
10 00:01:07.730 ⇒ 00:01:08.330 Uttam Kumaran: Yeah.
11 00:01:09.530 ⇒ 00:01:10.080 Zack Gibbs: Whoa.
12 00:01:10.470 ⇒ 00:01:11.150 Uttam Kumaran: Perfect.
13 00:01:11.740 ⇒ 00:01:27.299 Uttam Kumaran: I mean, we, yeah, I just reached out just to say hi again, and I know we kind of put this… last time we talked, we put this, you know, forecasting sort of scope in front of you. Mainly just wanted to kind of stay hi again and see, like, is there anything that we can be helpful with? I’ve sort of…
14 00:01:27.670 ⇒ 00:01:35.840 Uttam Kumaran: just been sort of on the sidelines watching. It seems like things are still running on the, like, data team side, so I’m… I’m curious to kind of hear
15 00:01:36.190 ⇒ 00:01:52.300 Uttam Kumaran: how you feel about, like, what we talked about, I think it may have been in November or December, about, like, the data team running, but of course, I think also this sort of forecast that we… forecast project that we put together, you know, we think that it directly lines up to some
16 00:01:52.880 ⇒ 00:01:58.299 Uttam Kumaran: actual revenue and some ROI, but, like, those are kind of the two topics on my mind.
17 00:01:59.830 ⇒ 00:02:03.340 Zack Gibbs: Yeah, I mean, I think for…
18 00:02:03.530 ⇒ 00:02:07.489 Zack Gibbs: Recently, we put this together for, pricing.
19 00:02:07.890 ⇒ 00:02:10.110 Zack Gibbs: Let me find that, let me find that worksheet.
20 00:02:14.950 ⇒ 00:02:24.420 Zack Gibbs: So, yeah, I think it would be good, high level, for you guys to maybe share the, like, the Notion doc that you put together, on the proposal. There’s…
21 00:02:25.110 ⇒ 00:02:29.359 Zack Gibbs: There’s another, kind of, path that I personally see.
22 00:02:29.540 ⇒ 00:02:32.530 Zack Gibbs: when it comes to our S&OP,
23 00:02:32.650 ⇒ 00:02:37.420 Zack Gibbs: forecasting process. Like, we could… we could absolutely go…
24 00:02:37.820 ⇒ 00:02:43.579 Zack Gibbs: like, third-party SaaS here, but I also feel like, we absolutely could…
25 00:02:44.190 ⇒ 00:02:55.000 Zack Gibbs: we could go, like, agent orchestration route as well, using… using something like LaneChain. Like, we’ve already done experimentation with N8N, on the reviews automation front.
26 00:02:55.240 ⇒ 00:02:58.260 Zack Gibbs: Yeah, we’ve seen some good success with, and…
27 00:02:58.600 ⇒ 00:03:08.720 Zack Gibbs: Now that we have that under our belt, around, you know, something like SNOP, which is, like, there’s some uniqueness to it,
28 00:03:09.170 ⇒ 00:03:17.350 Zack Gibbs: Having, like, a multi-agent approach that we… that we control all of the inner workings on, may be a better investment.
29 00:03:17.690 ⇒ 00:03:25.219 Zack Gibbs: Then some type of, like, just out-of-the-box SaaS that we then have to customize, and then support, but…
30 00:03:25.870 ⇒ 00:03:27.150 Zack Gibbs: I don’t know, I don’t know…
31 00:03:28.210 ⇒ 00:03:30.299 Zack Gibbs: We have to have time to make that.
32 00:03:30.470 ⇒ 00:03:38.640 Zack Gibbs: To do that work, and then maintain it, versus, you know, the… Buy and… buy and customize.
33 00:03:38.810 ⇒ 00:03:42.150 Zack Gibbs: on the more… on the more, SaaS front, but…
34 00:03:42.150 ⇒ 00:03:52.570 Uttam Kumaran: Yeah, Robert, do you want to talk about… I mean, we were… we were talking about… I pulled together some SaaS that, like, we found people were… some of our other clients were discussing. I think…
35 00:03:52.980 ⇒ 00:04:11.599 Uttam Kumaran: Robert and I’s conversation was, like, you could actually… you could actually build all this, and your assumption is right. I think also, like, we would love to consider, like, how we can leverage AI to support, you know, not only the development of it, but actually, like, being able to get more bang out of an internal system that you manage.
36 00:04:11.690 ⇒ 00:04:20.159 Uttam Kumaran: we are seeing that probably a lot of these vendors… a lot of… I would say not a lot of these are actually, like, going more AI route. They’re still doing a lot of the same
37 00:04:20.250 ⇒ 00:04:34.770 Uttam Kumaran: things, and so you’re right in that you may see… it may take them another year or two to ever implement any AI solution, and even then, it won’t kind of be governed, but yeah, maybe, Robert, I can hand it to you to sort of talk about this, and from… from your past experience, and yeah.
38 00:04:35.150 ⇒ 00:04:45.779 Robert Tseng: Yeah, I mean, I guess, like, kind of my observation of kind of how AI has been helping these forecasting projects is that, like, I mean, the current state forecasting flow, like, that data flow doesn’t change, like.
39 00:04:45.780 ⇒ 00:05:02.409 Robert Tseng: I mean, maybe the marks are a little bit different as we continue to evolve it, but generally, yeah, like, you need… you need inputs from your operators, and those… those inputs need to be normalized, put into data models, and it needs to be kind of refreshed. Like, that… that’s bread and butter, right? That’s, like, the kind of…
40 00:05:02.660 ⇒ 00:05:22.850 Robert Tseng: garbage-in, garbage-out situation, that… that these most out-of-the-box kind of SaaS tools need a calibration period. They don’t promise any results for 60 days, because they require you to basically do the same thing, which is just, like, load all your SOPs and your knowledge into… and give them data models in a way that their algorithms can consume.
41 00:05:22.850 ⇒ 00:05:34.030 Robert Tseng: So, you know, it could be as rudimentary as just, like, Google Sheets with a bunch of different tabs at first, and having all your people build that out. That’s kind of how I did forecasting when I was at Ruggable.
42 00:05:34.300 ⇒ 00:05:47.209 Robert Tseng: But then once we passed $100 million in revenue, then we needed to actually kind of build out the, kind of the underlying data models, which we had already done. We already had all of that, and with the dbt and the Looker stuff, with Urban Stems.
43 00:05:47.210 ⇒ 00:06:10.109 Robert Tseng: So, kind of… and then from there, like… but maintenance is hard, because the model continues to drift, which is why I had two full-time analysts, one on the supply side, one on the demand side, that… whose only job was just to maintain this. And I think where Udam is pointing out, like, where AI’s really able to help us is I actually believe that you don’t need those analysts anymore. Like, you can actually have multiple agents to,
44 00:06:10.110 ⇒ 00:06:26.089 Robert Tseng: continue to track where the model’s going to drift, what inputs need to be updated, like, it’s easy to populate across all the models. I don’t need somebody logging in at, you know, 8 a.m. every day, kind of making sure that everything is, like, good to go by 10 a.m. for leadership to review. So,
45 00:06:26.090 ⇒ 00:06:35.329 Robert Tseng: Yeah, and I think, like, we’ve had a lot of experience in building, kind of, multi… multi-cloud agents, kind of, maintenance across multiple data workflows.
46 00:06:35.600 ⇒ 00:07:00.189 Robert Tseng: And so I think it’s a great application for us to really just build… build… to build the initial model to just make sure that the performance is, you know, I think we could get it close… we could get it to 10% accuracy with very… with not much effort. Sorry, not much effort is the wrong phrase, but, like, I think that’s pretty baseline, that, like, any strategic finance team, their directive coming in would be to build you a forecast
47 00:07:00.190 ⇒ 00:07:23.320 Robert Tseng: access within 10%. And then we would just pretty much be adding all the AI wrappers around it to, continue to maintain it, which could be just a better, like, an easier maintenance solution than buying something out of the box. So, that’s why I was excited to come back and see, like, hey, is this something that you guys would be open to considering now? Because I think this would be… this would be the time to try.
48 00:07:27.640 ⇒ 00:07:34.749 Zack Gibbs: And was that… was that option, if you scroll back up there, I think you had looked at, like, SEN7 and…
49 00:07:34.750 ⇒ 00:07:35.220 Uttam Kumaran: Yeah.
50 00:07:35.220 ⇒ 00:07:39.560 Zack Gibbs: Fishbowl. So, option 4.
51 00:07:41.660 ⇒ 00:07:43.319 Zack Gibbs: like, I guess in my mind.
52 00:07:43.750 ⇒ 00:07:51.399 Zack Gibbs: You know, we’ve done some automation around our reviews, so prior, our customer reviews,
53 00:07:51.750 ⇒ 00:08:04.939 Zack Gibbs: Those would come in a few different, like, ways and channels, but it would just be a bunch of string data. And so, we’ve automated… we’ve automated a lot of the insights and, like, categorization of those, insights of those, recommendations of those.
54 00:08:05.050 ⇒ 00:08:24.879 Zack Gibbs: in Aiden, in, like, a custom, you know, a custom, set of workflows where you have different agents that have very distinct tasks, within that. And in Aiden is an orchestration layer, that would then, you know, hit other APIs or normalized data, add, you know, tagging, whatever may need to be done.
55 00:08:26.930 ⇒ 00:08:30.029 Zack Gibbs: it seems like You know, based on…
56 00:08:30.420 ⇒ 00:08:37.969 Zack Gibbs: what I know about our S&OP process, and, like, how manual it is, it seems like we could do something a little bit more…
57 00:08:38.330 ⇒ 00:08:49.970 Zack Gibbs: complicated, but more durable longer term, in, like, using an orchestration tool like Langchain. Have you guys… do you have experience with Langchain, specifically?
58 00:08:50.150 ⇒ 00:08:55.960 Zack Gibbs: There’s other ones that are out there that are similar in, like, complexity of what you can build.
59 00:08:56.410 ⇒ 00:09:03.230 Zack Gibbs: But LangChain’s the one that I’ve, like, done some dabbling in, so I’m kind of leaning more towards that path.
60 00:09:03.230 ⇒ 00:09:12.829 Uttam Kumaran: Yeah, so… I mean, so part of our business is, actually, we do a lot of AI application and, like, workflow automation. Our business, we started doing a lot of N8N,
61 00:09:12.870 ⇒ 00:09:27.760 Uttam Kumaran: But we probably have stopped doing primarily N8N work maybe, like, a year ago, and we actually are moving to some more, like, agent-based frameworks. One in particular that’s, like, very common is this company called, like, Mastra. It’s a JavaScript
62 00:09:27.900 ⇒ 00:09:32.759 Uttam Kumaran: agent framework. I… I think this is sort of, like, one of the leading edge
63 00:09:33.010 ⇒ 00:09:50.379 Uttam Kumaran: ones, but you can consider LaneChain similar, but Python-based. I would say it sort of depends on, like, two things. If you use sort of a no-code builder tool, yes, of course, like, maybe a broader audience can edit it, but
64 00:09:50.690 ⇒ 00:09:57.110 Uttam Kumaran: part of the difficulty with N8N is, like, observability is really tough, and having any type of, like,
65 00:09:57.350 ⇒ 00:10:08.519 Uttam Kumaran: like, being able to change things on the fly, you have to kind of go through a UI. So we typically build all of our things for clients, all of our agent frameworks through… through Mastra.
66 00:10:08.690 ⇒ 00:10:16.350 Uttam Kumaran: there’s a couple of these that are, like, sort of, like, the leading edge. Just found, like, it was really great. It’s completely open source as well.
67 00:10:16.510 ⇒ 00:10:34.149 Uttam Kumaran: And then additionally, like, there also is a lot of other ways to actually develop these, so if there are people in… at Urban Stems that want to use an agentic environment to build this, right, or edit pieces, you can actually use something like Cursor or Claude Code to actually make those changes through natural language.
68 00:10:34.220 ⇒ 00:10:41.929 Uttam Kumaran: And we tend to build the… the deployment, the CICD, in a way where it’s a lot less fragile.
69 00:10:42.030 ⇒ 00:10:43.629 Uttam Kumaran: So…
70 00:10:43.820 ⇒ 00:11:02.179 Uttam Kumaran: Totally, we actually have a lot of clients that we’re building these agentic systems on. We also internally use a lot of AI agents to do our data work, and so whether it’s writing dbt models, querying, helping with analysis, so very, very familiar, not just about building agents to hit APIs or
71 00:11:02.300 ⇒ 00:11:18.259 Uttam Kumaran: send prompts, but actually querying a database, like, writing data somewhere, doing an analysis, is kind of the ways that we’ve been trying to push things forward. And so I feel, like, more than capable of building, at least from what I heard, the system.
72 00:11:18.620 ⇒ 00:11:29.570 Robert Tseng: Can I put this another way? Yeah. Like, so I think, like, the NA… why we moved on from the NA8 framework? NAN is basically, like, Zapier for, like, like, I guess for the NAI world, right?
73 00:11:29.570 ⇒ 00:11:33.250 Zack Gibbs: It was a good… it was a good… it was a good… it’s like, from a…
74 00:11:34.220 ⇒ 00:11:45.100 Zack Gibbs: figuring out how this works, and it ends a great starting point, then you really quickly realize how limited, limited, you know, you are after that, after that initial build.
75 00:11:45.960 ⇒ 00:12:04.110 Robert Tseng: So, when we say agentic orchestration, like, now you’re in an environment that’s basically back to, like, being, like, chat-first against, right? And, you know, and because of MCP, kind of, developments, like, you know, you’re not constrained by how many connectors, like, NAN already has built out. You can just custom build an FTC pretty much
76 00:12:04.130 ⇒ 00:12:08.639 Robert Tseng: for pretty much most data sources. And so, just, like, as far as, like.
77 00:12:08.900 ⇒ 00:12:32.230 Robert Tseng: kind of how robust it is, like, we think that going… that’s why… that’s why we’ve… we’ve really invested in just building the scaffolding and trusting that the infrastructure of just building on top of something that’s agent… agent-first is going to be… is going to be longer-lasting. We don’t, like, we just don’t believe that people are going to be using N88 in a couple years, just because, like, you won’t… you won’t need to. So, but I think, like, yeah, that…
78 00:12:32.230 ⇒ 00:12:35.119 Robert Tseng: that’s kind of… I feel like NAN was a good starting point.
79 00:12:35.120 ⇒ 00:12:39.019 Robert Tseng: And, you know, for what you’ve described, and maybe there’s more to it, but, like.
80 00:12:39.210 ⇒ 00:13:00.239 Robert Tseng: which sounds like you’ve got something hooked up that just does, like, automated feed, like, kind of fee categorization, which, yeah, I think is super valuable for kind of categorizing reviews, especially in anything else, but if you’re using that same kind of approach to, let’s say, launching new product lines, you would have to go back to the platform, you’d have to go and, like, kind of
81 00:13:00.240 ⇒ 00:13:03.019 Robert Tseng: Make all the different edits within the orchestration
82 00:13:03.020 ⇒ 00:13:08.840 Robert Tseng: A layer, you know, in order to go and continue to make…
83 00:13:08.840 ⇒ 00:13:32.589 Robert Tseng: to force those categories to be more open, whereas an agent approach is a lot more, kind of, like, you can do it on the fly, you could still have the same guardrails, and also, like, if you need… if you need to interact with it for ideas, like, NAN’s not going to give you any, like, ideas for how to make your system better, right? It’s really just kind of… you have to… you’re kind of in the driver’s seat of setting it up, not really having
84 00:13:32.700 ⇒ 00:13:37.659 Robert Tseng: Kind of, like, that feedback loop that an agent-based system is going to give you.
85 00:13:40.030 ⇒ 00:13:40.660 Zack Gibbs: Yeah.
86 00:13:41.260 ⇒ 00:13:41.970 Zack Gibbs: Yep.
87 00:13:42.180 ⇒ 00:13:47.260 Zack Gibbs: So in… Anything… anything from, you know, anything tactically…
88 00:13:47.570 ⇒ 00:14:05.869 Zack Gibbs: around, you know, our S&OP processes, we wouldn’t consider kicking them off until after Mother’s Day. Like, right now, we’ve, you know, we’ve got, we have a, I don’t know, say 30 or so distinct items, about, I don’t know, there’s 10 or 12 of those that are reporting-based items that are, like.
89 00:14:05.870 ⇒ 00:14:18.400 Zack Gibbs: tactically, we need to do these items before Mother’s Day, and so that’s kind of the general focus area. I’m thinking about just post-Mother’s Day, there’s a couple… there’s a couple things that are opportunities. One is,
90 00:14:19.250 ⇒ 00:14:24.910 Zack Gibbs: Just simple cost reduction and some more consolidation, so…
91 00:14:25.380 ⇒ 00:14:29.490 Zack Gibbs: Let me go back and find the… this is all in a Fig… FigJam file.
92 00:14:29.720 ⇒ 00:14:33.999 Zack Gibbs: Some BI Pipelines…
93 00:14:42.700 ⇒ 00:14:46.230 Zack Gibbs: So, I mean, my mindset is, like, post-Mother’s Day, what is the…
94 00:14:48.130 ⇒ 00:14:51.080 Zack Gibbs: what is the opportunity? One of the opportunities is…
95 00:14:52.030 ⇒ 00:14:56.580 Zack Gibbs: You know, a tool, which seems less likely now, versus something custom-built.
96 00:14:58.190 ⇒ 00:15:07.540 Zack Gibbs: to support our SNOP team, but also to reduce the amount of A manual work, and B, headcount in that area of the business.
97 00:15:08.230 ⇒ 00:15:23.700 Zack Gibbs: So let me share… this file… So… You know, this is the…
98 00:15:25.880 ⇒ 00:15:36.020 Zack Gibbs: this is the rough… this is the rough cost of our current BI tooling. So, number one…
99 00:15:37.580 ⇒ 00:15:41.400 Zack Gibbs: we drop a lot of data into Redshift that
100 00:15:41.710 ⇒ 00:15:45.690 Zack Gibbs: we’re paying for, that we don’t necessarily use downstream. So…
101 00:15:46.010 ⇒ 00:15:48.960 Zack Gibbs: Cutting out what is dropping into Redshift.
102 00:15:49.260 ⇒ 00:15:54.330 Zack Gibbs: is Lower… lower hanging fruit, but would require some effort.
103 00:15:57.080 ⇒ 00:16:04.769 Zack Gibbs: Polytomic is by far… you know, we already pay… we already pay NetSuite for this ability to extract data, which is…
104 00:16:04.950 ⇒ 00:16:11.550 Zack Gibbs: sad, in itself. But Polyatomic is… Really expensive for what we…
105 00:16:11.550 ⇒ 00:16:12.140 Uttam Kumaran: Yeah.
106 00:16:12.740 ⇒ 00:16:18.590 Zack Gibbs: So for what we… how we… for how… how we leverage it. And…
107 00:16:19.300 ⇒ 00:16:35.269 Zack Gibbs: I don’t… I don’t know if we had a great number of options back when this decision was made. However, I don’t think that we… I think that we made a bad… I think we made a mistake here. And now we… now we are going to look to unwind that post-Mother’s Day.
108 00:16:35.610 ⇒ 00:16:51.730 Zack Gibbs: Now, there… there are… there’s new connectors. Hivo has one, Stitch has one that we could evaluate. I think that the value that Polytomic provided was some of the customization that was needed. So…
109 00:16:51.900 ⇒ 00:16:54.449 Zack Gibbs: But Polytomic is a big, like, it’s a big…
110 00:16:54.590 ⇒ 00:16:59.340 Zack Gibbs: source of cost, when we look at just, like, overall.
111 00:16:59.340 ⇒ 00:16:59.870 Uttam Kumaran: Yeah.
112 00:17:00.890 ⇒ 00:17:08.180 Zack Gibbs: The other side of it is, like, we’re gonna continue to whittle down Looker, like, our Looker usage internally. Like, some of these teams…
113 00:17:08.180 ⇒ 00:17:19.040 Zack Gibbs: they have all the data that they would ever need just in Shopify, natively. They don’t necessarily… we don’t necessarily need them to be in Looker, or they’re only hitting certain Looker reports for certain reasons.
114 00:17:19.040 ⇒ 00:17:28.269 Zack Gibbs: We’ve done kind of, like, a step one of whittling down, but there’s a step two, step three, that we had already expected, to have. But, like, the…
115 00:17:28.500 ⇒ 00:17:34.690 Zack Gibbs: The high-level cost opportunity area is absolutely on the polyatomic… polyatomic and redshift side.
116 00:17:34.690 ⇒ 00:17:37.479 Uttam Kumaran: You can get… you can delete dbt Cloud, too.
117 00:17:38.130 ⇒ 00:17:39.330 Uttam Kumaran: Pretty easily.
118 00:17:40.110 ⇒ 00:17:55.989 Zack Gibbs: Well, dbt Cloud, we… I mean, we have, like, a super legacy plan, that they’ve just said, you guys can continue on this legacy plan, because, we… we signed up, like, soon after they were… they, released some of their products.
119 00:17:56.390 ⇒ 00:17:57.140 Zack Gibbs: So…
120 00:17:57.140 ⇒ 00:18:02.739 Uttam Kumaran: So, like, it’s a completely open source product, so, like, for a lot of our clients, we run it completely free.
121 00:18:03.000 ⇒ 00:18:05.049 Uttam Kumaran: Either on GitHub or on, like, a cheap
122 00:18:06.090 ⇒ 00:18:09.580 Uttam Kumaran: Redshift EC2 instance, so that’s, like.
123 00:18:10.690 ⇒ 00:18:15.750 Uttam Kumaran: that’s an easy thing for us to remove, too, but I can see that it’s not a… it’s not a big part.
124 00:18:16.350 ⇒ 00:18:16.980 Zack Gibbs: Yeah.
125 00:18:17.370 ⇒ 00:18:28.379 Zack Gibbs: So, like, high level, when I look at our opportunities, outside of this, there’s the S&OP side, and…
126 00:18:28.540 ⇒ 00:18:31.800 Zack Gibbs: Inside of just the general costing,
127 00:18:32.120 ⇒ 00:18:37.239 Zack Gibbs: it’s redshift and polyatomic. Like, those are the two pieces that stick out, right? So…
128 00:18:37.370 ⇒ 00:18:42.659 Zack Gibbs: Do you guys agree, disagree, see things… anything slightly differently?
129 00:18:44.950 ⇒ 00:18:54.509 Robert Tseng: Well, I guess, like, my first thought is that, like, maybe as, like, a overall share of your IT budget, like, seems pretty reasonable for a company your size, like.
130 00:18:54.620 ⇒ 00:18:59.860 Robert Tseng: I mean, I can see why you would want to continue to consolidate, I just feel like there are probably bigger…
131 00:18:59.890 ⇒ 00:19:06.629 Robert Tseng: bigger wins that you can get elsewhere. I mean, if you move polyatomic decision, yeah, sure, there’s probably cheaper ways now that you have
132 00:19:06.630 ⇒ 00:19:20.730 Robert Tseng: the customization down, you know exactly how the connector needs to be set up. Let’s say you slashed that by 50%, you know, that’s still, like, I don’t know, like, a thousand bucks a month or something, maybe? And, on the Looker side, you could remove some licenses, maybe cut it down.
133 00:19:20.760 ⇒ 00:19:39.229 Robert Tseng: yeah, like, we manage, like, BI… like, kind of BI tool budgets for companies a bit larger that under 20K, so maybe you get another… let’s… I think you could reasonably see a path to 20% cuts within a quarter, or two quarters, but, like, I don’t know, just… it feels like it’s not that big of a deal.
134 00:19:39.230 ⇒ 00:19:41.209 Robert Tseng: If I were to just be honest, yeah.
135 00:19:41.210 ⇒ 00:19:43.150 Zack Gibbs: Yeah. Yeah, that’s fair.
136 00:19:43.370 ⇒ 00:19:43.980 Robert Tseng: Yeah.
137 00:19:45.800 ⇒ 00:19:48.849 Robert Tseng: Like, I mean, I would be more interested in trying to, like.
138 00:19:49.110 ⇒ 00:19:59.780 Robert Tseng: grow your business by 20% than trying to cut your VI tooling costs by 20%. But that’s… that’s just, I mean, that’s… I mean, obviously, it’s easy to say that from the outside.
139 00:20:00.280 ⇒ 00:20:00.920 Zack Gibbs: Right.
140 00:20:05.820 ⇒ 00:20:15.619 Robert Tseng: But yeah, no, I think it makes sense. Those areas you highlighted, like I said, I think there is a clear way to cut it at least 20%. Like, I do see that.
141 00:20:15.770 ⇒ 00:20:24.160 Uttam Kumaran: Yeah, like, I think… I think we identified some wins on Redshift, too. Like, I think we could do this, for sure, but I agree with Robert. Like, I want to go after the…
142 00:20:24.280 ⇒ 00:20:27.100 Uttam Kumaran: The growth side, like, as well.
143 00:20:28.220 ⇒ 00:20:31.859 Zack Gibbs: Yeah. Yeah. Yeah, I mean, there’s… there’s more…
144 00:20:32.880 ⇒ 00:20:38.019 Zack Gibbs: On, like, just on the, like, cost mitigation front, there’s more… there’s more…
145 00:20:38.210 ⇒ 00:20:44.000 Zack Gibbs: Juice to squeeze out of just the forecasting and planning side of things. No.
146 00:20:44.000 ⇒ 00:20:51.129 Robert Tseng: I mean, even beyond the S&O, like, on the growth marketing side, like, I don’t think we’ve really pitched you much on, like, our work there, but I think, like.
147 00:20:51.440 ⇒ 00:21:05.040 Robert Tseng: I… one of our leading services is, like, basically around tagging and tracking, and, you know, just typically e-com companies, like, they’re maybe identifying, like, 60%, 60% to 80% of
148 00:21:05.040 ⇒ 00:21:19.039 Robert Tseng: Of their… of their visitors, but, like, we are able to help companies identify, like, 95% with, like, edge layer tracking. So, like, you know, there’s… I think there are some really tangible ways that we’re able to kind of break… drive top-line growth.
149 00:21:19.510 ⇒ 00:21:35.710 Robert Tseng: on the marketing side, too. But I know we’ve never really kind of played in that area with your team, and you’re more on the product side, but maybe there’s, like, other, you know, maybe you’re hearing other opportunities across the org that maybe we could also be introduced to.
150 00:21:35.970 ⇒ 00:21:54.369 Zack Gibbs: Yeah, I mean, I think the… the… on the growth side of things, we have… we have Northbeam as a way to have better attribution. However, I would say that our… the growth in marketing, like, how we view growth is inside of the marketing funnel, and…
151 00:21:54.370 ⇒ 00:21:54.880 Robert Tseng: Yeah.
152 00:21:54.880 ⇒ 00:22:05.649 Zack Gibbs: And that organization internally is gonna get shook… it’s gonna… there’s a shake-up there that’s gonna be coming in our, like, next fiscal. And so there’s opportunity there, as well, but…
153 00:22:06.500 ⇒ 00:22:07.330 Zack Gibbs: you know.
154 00:22:07.590 ⇒ 00:22:14.659 Zack Gibbs: Until there… until that… the dust settles on that, it probably is not a great area to focus on, right?
155 00:22:14.660 ⇒ 00:22:15.610 Robert Tseng: Yeah, makes sense.
156 00:22:15.610 ⇒ 00:22:31.670 Zack Gibbs: So, so I’m thinking more tactically of, okay, there’s… we know there’s some opportunities in the cost reduction side, but I would like to pursue, you know, a, you know, more of an agent. Something that we maintain, the engineering staff, like, sets up initially.
157 00:22:31.670 ⇒ 00:22:42.319 Zack Gibbs: And provides the tooling and handoff to our, S&OP team on, you know, forecast demand planning, and there’s more…
158 00:22:42.780 ⇒ 00:22:47.960 Zack Gibbs: audit, traceability, like, there’s more accountability there, which…
159 00:22:47.960 ⇒ 00:22:50.590 Robert Tseng: Do you guys build internal tooling for your OP3 SNO team?
160 00:22:51.390 ⇒ 00:22:59.319 Robert Tseng: Anything custom, or is it mostly just kind of… I mean, sounds like you’re kind of the gatekeeper for all, like, tech tools for all the teams, possibly?
161 00:23:00.600 ⇒ 00:23:02.470 Zack Gibbs: Yeah, we don’t… I mean, we haven’t built anything…
162 00:23:02.730 ⇒ 00:23:12.810 Zack Gibbs: unique for them. They’ve just been operating… they’ve been oper… they operate inside of, they use Shopify data, they operate inside of NetSuite.
163 00:23:12.850 ⇒ 00:23:32.370 Zack Gibbs: And then they have their own process that’s all Google Sheets-based on what are the… what are the demand patterns, what are we forecasting, what assumptions are we making, and there’s very little auditability, in that, and when mistakes are made, like, there’s big cost implications, which.
164 00:23:32.370 ⇒ 00:23:32.750 Robert Tseng: She happens.
165 00:23:32.750 ⇒ 00:23:35.169 Zack Gibbs: All the time. And so…
166 00:23:35.360 ⇒ 00:23:43.650 Zack Gibbs: having a more mature process there is really, like, the starting point. And like I said, my initial headspace was.
167 00:23:44.110 ⇒ 00:23:50.290 Zack Gibbs: let’s go look at… let’s go look at SaaS vendors that could help solve this problem relatively simply.
168 00:23:50.570 ⇒ 00:23:57.239 Zack Gibbs: But now, with some of these, you know, the emergence of some of these more open source tools, I think it’s better, because our… some of our…
169 00:23:57.360 ⇒ 00:24:01.999 Zack Gibbs: you know, SOPs are unique, I think it’s better to, like…
170 00:24:03.130 ⇒ 00:24:09.590 Zack Gibbs: work on the fundamental side and build it out ourselves. Yeah.
171 00:24:10.320 ⇒ 00:24:29.300 Uttam Kumaran: I think you’re also gonna see that, like, it’s through your suspicion about these software platforms, they’re releasing these AI features at a platform approach, so they do very little for, like, way too many people, right? So they never have any depth. And so that’s similarly, like, we’ve… even in our business, we have consolidated and have built out
172 00:24:29.390 ⇒ 00:24:44.940 Uttam Kumaran: the two… the two or three features that comes in, like, a large platform that we just need, and it’s been really effective. And we’ve learned a lot along the way about how to build agents, how to build agents that can query data, but of course, actually, what it forces you to do is, like.
173 00:24:45.020 ⇒ 00:24:52.249 Uttam Kumaran: what exactly do I actually need? Because you have to instruct the agent to do that, deliver it in the right way, and we found it to be, like.
174 00:24:52.380 ⇒ 00:25:02.290 Uttam Kumaran: really, really effective, not only just from, okay, we don’t have other tooling, right? But it’s actually helped us codify some of our processes and, you know, really specifically.
175 00:25:02.380 ⇒ 00:25:11.170 Uttam Kumaran: And now we can scale that up. So, as a business has scaled, we’re able to, like, use those agents and abstract those away.
176 00:25:11.170 ⇒ 00:25:24.429 Uttam Kumaran: And that’s how we… I… we… to give you this, like, we used N8N really early on, like, 2 years ago, to do a lot of internal business process automation, but probably a year… like, a year after that, I came to the same conclusion, where I was like.
177 00:25:24.430 ⇒ 00:25:43.990 Uttam Kumaran: this is an insane thing to maintain. There was no alternative, really, apart from Langchain 3, 2 or 3 years ago. Now there’s a whole host of agent frameworks that operate very similarly to the standard software development lifecycle, you know? So, you benefit from all the things we learned about developing software to then harness, like, a specific LLM.
178 00:25:44.020 ⇒ 00:25:51.070 Uttam Kumaran: It’s a much… Much better, much faster to develop, and then much faster to actually to tweak, you know?
179 00:25:52.330 ⇒ 00:25:52.870 Zack Gibbs: Yep.
180 00:25:53.280 ⇒ 00:25:54.060 Zack Gibbs: Gotcha.
181 00:25:54.490 ⇒ 00:25:55.280 Zack Gibbs: Okay.
182 00:25:55.450 ⇒ 00:26:07.089 Zack Gibbs: Well, I think where my head’s at is that, you know, we were tactically focused through Mother’s Day. We have a whole bunch of things on, like, what we, you know, view as core services. We have all those storefront initiatives.
183 00:26:07.090 ⇒ 00:26:17.689 Zack Gibbs: you know, items on the BI front that are all prioritized, and we just need to, like, we just need to work through them. There’s also lessons learned from Valentine’s Day.
184 00:26:17.690 ⇒ 00:26:24.489 Zack Gibbs: Last month. And so, like, we’re not gonna start picking up new projects until,
185 00:26:25.060 ⇒ 00:26:29.029 Zack Gibbs: you know, until the end of May, beginning of June.
186 00:26:29.270 ⇒ 00:26:41.150 Zack Gibbs: And, like I said, on the growth side, like, there’s… my expectation is that come July timeframe, there’s gonna be some changes made there, and, you know.
187 00:26:41.200 ⇒ 00:26:49.659 Zack Gibbs: new leaders come in with new ideas and new expectations and new thoughts. Like, I want some of that dust to settle before proposing anything else there, from…
188 00:26:50.260 ⇒ 00:26:52.260 Zack Gibbs: From my end, and…
189 00:26:52.330 ⇒ 00:27:06.140 Zack Gibbs: But, you know, I think one… assuming that we have bandwidth, you know, starting in the June timeframe, to, like, actually pick up and push forward the… the SNOP side of things, like, that’s our lower… that’s our…
190 00:27:06.140 ⇒ 00:27:17.609 Zack Gibbs: less busy time of the year from a, you know, demand perspective is really that, like, June through September timeframe, and so that’s… that is a good time to, like.
191 00:27:17.660 ⇒ 00:27:32.129 Zack Gibbs: get this set up and running. Whereas we could potentially use some of your consulting services during that time on that project. But I… I want us to have, like, a real clear idea of how we would tackle it,
192 00:27:32.200 ⇒ 00:27:37.659 Zack Gibbs: first, because I think that, like, us leading that process internally is going to be helpful.
193 00:27:38.370 ⇒ 00:27:41.309 Zack Gibbs: to really try to get to the best solution. So…
194 00:27:41.750 ⇒ 00:27:57.239 Zack Gibbs: I think there’s an opportunity there, to leverage some of your, you know, your expertise in that project, and then on the growth side, like I said, I don’t… I don’t think… I don’t think there’s anything… I don’t think there’s appetite there until the dust settles, which is likely, you know, say, August, September timeframe.
195 00:27:57.990 ⇒ 00:27:58.530 Uttam Kumaran: Okay.
196 00:27:58.800 ⇒ 00:28:01.110 Robert Tseng: Yeah. Okay.
197 00:28:01.520 ⇒ 00:28:09.249 Robert Tseng: Yeah, makes sense. I think, like, I think we’d be interested in, you know, if you see… if you hear anything on the marketing side, like you said, whether it’s…
198 00:28:09.320 ⇒ 00:28:17.790 Robert Tseng: I mean, growth is pretty broad. It could be, you know, like, life cycle, kind of paid growth, like, all those areas we, we have, we have
199 00:28:17.810 ⇒ 00:28:29.709 Robert Tseng: we have, services for. So, would be interested, you know, if you, if you know of, if you guys are bringing in a new leader, or, you know, there’s a team that you feel like is, is kind of going through some, some kind
200 00:28:29.860 ⇒ 00:28:49.769 Robert Tseng: like a… like a shake-up that, like, may… may just want another opinion. Like, we’re happy to even just be a second opinion on, you know, how are they actually performing compared to what we see, right? Because a good chunk of our business is still e-com, so I feel like we’re pretty dialed in on, like, what’s… what’s actually helpful for driving growth for businesses like yours.
201 00:28:50.240 ⇒ 00:28:51.810 Zack Gibbs: Yeah, gotcha. Okay.
202 00:28:51.810 ⇒ 00:28:52.410 Robert Tseng: Yeah.
203 00:28:54.100 ⇒ 00:28:54.970 Zack Gibbs: Alright.
204 00:28:55.450 ⇒ 00:29:11.179 Uttam Kumaran: Okay, perfect. Then yeah, I think I’ll… Zach, I’ll probably just kind of keep in touch and message you sort of around that time frame. And then, yeah, if there’s anything on the data modeling or normal sort of data side, like, we did write some stuff about minimizing redshift and things like that, so I feel like if…
205 00:29:11.290 ⇒ 00:29:18.710 Uttam Kumaran: there’s some stuff, I think, that should be in the repo, but happy to just to share what we learned, because there could be some wins shorter term, so…
206 00:29:19.030 ⇒ 00:29:20.590 Zack Gibbs: Yeah, gotcha. Okay.
207 00:29:20.770 ⇒ 00:29:27.789 Zack Gibbs: Alright, well, cool. Well, I appreciate it, and yeah, let’s keep in touch as we move into post-Mother’s Day for us.
208 00:29:28.570 ⇒ 00:29:29.240 Uttam Kumaran: Perfect.
209 00:29:29.240 ⇒ 00:29:29.580 Robert Tseng: Cool.
210 00:29:29.580 ⇒ 00:29:30.060 Zack Gibbs: Alright.
211 00:29:30.060 ⇒ 00:29:32.159 Robert Tseng: Alright, good to see you. Thanks, Zach.
212 00:29:32.160 ⇒ 00:29:32.670 Zack Gibbs: Bye.