Meeting Title: Insomnia Planning Date: 2025-10-13 Meeting participants: Casie Aviles, Uttam Kumaran, Samuel Roberts, Mustafa Raja, Robert Tseng
WEBVTT
1 00:00:52.540 ⇒ 00:00:53.880 Uttam Kumaran: Hello!
2 00:00:54.910 ⇒ 00:00:56.000 Uttam Kumaran: Good evening.
3 00:00:56.460 ⇒ 00:00:57.330 Casie Aviles: And the tongue.
4 00:01:35.890 ⇒ 00:01:40.890 Samuel Roberts: a, oh, no.
5 00:01:55.510 ⇒ 00:01:58.720 Samuel Roberts: Might disappear for a second if I can’t get this charger to work.
6 00:02:00.220 ⇒ 00:02:07.710 Samuel Roberts: For some reason, my… Charger is not working, and I didn’t… Realize it?
7 00:02:08.340 ⇒ 00:02:09.939 Samuel Roberts: And then my battery’s very low.
8 00:02:11.470 ⇒ 00:02:11.990 Uttam Kumaran: You’re good.
9 00:02:11.990 ⇒ 00:02:15.019 Samuel Roberts: I don’t know what’s happening, why this charger’s not working.
10 00:02:15.470 ⇒ 00:02:17.650 Samuel Roberts: freaked me out. It should just be working.
11 00:02:27.810 ⇒ 00:02:29.560 Samuel Roberts: There we go, almost.
12 00:02:33.890 ⇒ 00:02:41.469 Uttam Kumaran: Okay, gang, so I was hoping today I could just, basically, we could just do, like, a run-through of Scorecard.
13 00:02:41.630 ⇒ 00:02:44.769 Uttam Kumaran: And… the weekly stuff…
14 00:02:45.030 ⇒ 00:02:53.719 Uttam Kumaran: Like, if I could just get literally an overview of everything, and I guess to start with, I just wanted to look… talk about, like, current state.
15 00:02:53.930 ⇒ 00:03:00.570 Uttam Kumaran: And then… we can lay on all the problems, and I think we can work on some solutions.
16 00:03:00.720 ⇒ 00:03:05.620 Uttam Kumaran: if helpful, even the context, Casey, on, like, the issue that came up today, and, like.
17 00:03:06.000 ⇒ 00:03:10.230 Uttam Kumaran: That is a good way to frame, like, what the current problems are, that’d be great, but…
18 00:03:10.590 ⇒ 00:03:17.120 Uttam Kumaran: Yeah, and then in parallel, I’m gonna make sure that I have access to everything that you’re seeing, too, so I know there’s just a lot of links and stuff, so…
19 00:03:18.790 ⇒ 00:03:22.820 Casie Aviles: Sure, I can go and share my screen and just give a…
20 00:03:23.580 ⇒ 00:03:24.860 Uttam Kumaran: an overview of…
21 00:03:24.860 ⇒ 00:03:25.909 Casie Aviles: how I do it?
22 00:03:28.750 ⇒ 00:03:30.599 Casie Aviles: Can you guys see my screen now?
23 00:03:30.780 ⇒ 00:03:31.290 Mustafa Raja: Yes.
24 00:03:31.290 ⇒ 00:03:32.010 Uttam Kumaran: Yes.
25 00:03:32.940 ⇒ 00:03:33.620 Casie Aviles: Alright.
26 00:03:34.050 ⇒ 00:03:35.639 Robert Tseng: Okay, sorry, I’m here.
27 00:03:37.160 ⇒ 00:03:37.830 Uttam Kumaran: Hey.
28 00:03:38.280 ⇒ 00:03:43.649 Uttam Kumaran: Basically, I’m just asking for, like, a complete overview of, like, kind of scorecard. We could talk through one of the issues, and then…
29 00:03:43.770 ⇒ 00:03:47.100 Uttam Kumaran: Play in the week, so… Feel free to just…
30 00:03:47.480 ⇒ 00:03:51.660 Uttam Kumaran: listen, and then I can call on you, Robert, if you… if we need anything, or… or…
31 00:03:51.660 ⇒ 00:03:52.650 Robert Tseng: Sure.
32 00:03:52.650 ⇒ 00:04:04.060 Uttam Kumaran: Yeah, let me know if you have any thoughts. I just want to get an overview of the current state. I want to kind of hear, like, when an error comes up, like, what’s our process, and then I can then take all that and propose sort of, like, what the architecture is going to be.
33 00:04:04.760 ⇒ 00:04:16.190 Robert Tseng: Okay, yeah, I also kind of have… I’ve done some more thinking on how I think we should kind of move on from this format, because right now, Casey kind of updates both the Marketing Performance Tracker and this scorecard.
34 00:04:16.300 ⇒ 00:04:30.940 Robert Tseng: And, anyway, like, I have some thoughts on how, you know, because we have a ticket that’s, like, redesigned this, like, process. Like, I kind of have… I have thoughts on that now, but I guess he can share, kind of, what’s breaking first, and then we can go back to that.
35 00:04:33.000 ⇒ 00:04:39.320 Casie Aviles: Yeah, sure, so… currently what, this… there’s this tracker that I have to fill out.
36 00:04:39.870 ⇒ 00:04:46.309 Casie Aviles: So, I take… so basically what I did for the automations is…
37 00:04:47.530 ⇒ 00:04:49.730 Casie Aviles: It should… it’s in this sheet.
38 00:04:51.890 ⇒ 00:04:57.860 Casie Aviles: for each source, like, for FDA, which is DoorDash and Uber.
39 00:04:58.290 ⇒ 00:04:59.810 Casie Aviles: I would have, like, a sheet.
40 00:05:00.050 ⇒ 00:05:09.139 Casie Aviles: Where, the pipelines would dump the data there. So, for example, for DoorDash, the data gets dumped here.
41 00:05:10.070 ⇒ 00:05:13.539 Casie Aviles: For Google here, you know, I have sheets for each.
42 00:05:14.010 ⇒ 00:05:21.050 Casie Aviles: And I created, like, a mock scorecard, which should let me just… For a given date.
43 00:05:21.920 ⇒ 00:05:26.180 Casie Aviles: For example, Like, for Sunday.
44 00:05:26.450 ⇒ 00:05:32.189 Casie Aviles: I would just get, like, the data here, and I would just paste it Like, for example, this one.
45 00:05:33.440 ⇒ 00:05:36.349 Casie Aviles: I would just paste it here.
46 00:05:38.650 ⇒ 00:05:41.430 Casie Aviles: So that’s… that’s how it works right now.
47 00:05:42.200 ⇒ 00:05:47.899 Casie Aviles: And then for those that I don’t have access, which is Meta, I have to go to Looker for that.
48 00:05:52.120 ⇒ 00:05:57.910 Casie Aviles: What else? So, that’s pretty much the process, and… For, for…
49 00:05:58.280 ⇒ 00:06:02.499 Casie Aviles: the own marketing, and also the FDA promotion, we have, like.
50 00:06:02.900 ⇒ 00:06:11.239 Casie Aviles: intermediate sheets for those, so that would be the marketing performance tracker, and then the FDA projections tracker.
51 00:06:11.670 ⇒ 00:06:15.479 Casie Aviles: So, to start, we have the Marketing Performance Tracker.
52 00:06:15.980 ⇒ 00:06:21.400 Casie Aviles: And… Basically, this contains, like, the Braze data that we need.
53 00:06:22.880 ⇒ 00:06:27.969 Uttam Kumaran: And one thing, Casey, the Daily Impact Scorecard SOP that you have.
54 00:06:28.320 ⇒ 00:06:34.290 Uttam Kumaran: this is also, like, basically has the same thing you’re explaining, like, each of the SOPs, right?
55 00:06:34.910 ⇒ 00:06:38.459 Casie Aviles: Yes, this, this the automated process.
56 00:06:39.060 ⇒ 00:06:40.290 Uttam Kumaran: Trying to this?
57 00:06:40.510 ⇒ 00:06:45.950 Casie Aviles: There’s, like, a new thing that they’re doing, which is not yet there.
58 00:06:46.160 ⇒ 00:06:50.799 Casie Aviles: That was just from last week, so I’ll also get to that.
59 00:06:51.240 ⇒ 00:06:55.189 Casie Aviles: But yeah, like, this is, the overview, like, for…
60 00:06:55.720 ⇒ 00:07:01.690 Casie Aviles: We also do, like, a 7-day attribution, which is, like, refreshing it for the past week.
61 00:07:02.060 ⇒ 00:07:04.300 Casie Aviles: Which, I have, like, this…
62 00:07:05.170 ⇒ 00:07:12.589 Casie Aviles: Last updated column here to show, like, when was it last updated. So we tried to update, like, for 7-day windows.
63 00:07:13.450 ⇒ 00:07:14.510 Casie Aviles: And…
64 00:07:14.950 ⇒ 00:07:24.029 Casie Aviles: there are some pieces that are still manual, like, for example, this part, this one is manual, because I’ve asked with Braze support, and
65 00:07:24.510 ⇒ 00:07:29.779 Casie Aviles: I just couldn’t, like, get, like, an endpoint to get their, scheduled date.
66 00:07:30.050 ⇒ 00:07:35.070 Casie Aviles: So I have to go into Braze and add it for, like, each row that I have to…
67 00:07:35.440 ⇒ 00:07:38.499 Casie Aviles: Add here, but pretty much the rest, like.
68 00:07:39.020 ⇒ 00:07:43.030 Casie Aviles: These are already automated. These are just copy-paste.
69 00:07:44.650 ⇒ 00:07:47.379 Casie Aviles: Yeah, other than that.
70 00:07:49.580 ⇒ 00:07:57.670 Casie Aviles: We also have the FDA Performance Tracker, or, sorry, I mean the FDA Projections Tracker, and basically this just…
71 00:07:58.080 ⇒ 00:08:07.349 Casie Aviles: So they usually have, like, two campaign types. They have, like, an ads or sponsored listing, and then they have one with promo… promotions.
72 00:08:07.960 ⇒ 00:08:12.149 Casie Aviles: And then I just go here, basically, to this tracker.
73 00:08:12.380 ⇒ 00:08:17.280 Casie Aviles: I would just copy, basically, this, and then I would just paste it there.
74 00:08:18.340 ⇒ 00:08:22.470 Casie Aviles: For example, this one, so October 12…
75 00:08:24.810 ⇒ 00:08:27.349 Casie Aviles: So yeah, I just do that, pretty much.
76 00:08:27.930 ⇒ 00:08:36.900 Casie Aviles: And… Yeah, I think that’s… that’s all from… for the… Daily stuff,
77 00:08:37.220 ⇒ 00:08:39.659 Casie Aviles: I think where it broke is…
78 00:08:40.760 ⇒ 00:08:43.570 Casie Aviles: Like, the data here was not updated.
79 00:08:45.850 ⇒ 00:08:49.530 Casie Aviles: What they saw was an outdated one, but…
80 00:08:51.660 ⇒ 00:08:53.619 Casie Aviles: And let me go to that,
81 00:08:56.630 ⇒ 00:08:57.460 Casie Aviles: Red.
82 00:09:05.570 ⇒ 00:09:10.460 Casie Aviles: Yeah, so it was just lower than what DoorDash was, showing.
83 00:09:11.760 ⇒ 00:09:14.879 Casie Aviles: That was just because it wasn’t updated then.
84 00:09:16.930 ⇒ 00:09:22.859 Casie Aviles: But it should be, like, a match now. Let’s see here, 1045.
85 00:09:30.100 ⇒ 00:09:34.340 Casie Aviles: That was for… the 9th?
86 00:09:35.100 ⇒ 00:09:35.920 Casie Aviles: Let’s see…
87 00:09:39.480 ⇒ 00:09:42.920 Casie Aviles: Yeah, I think it’s… yeah, this one, DoorDash Promotions.
88 00:09:45.940 ⇒ 00:09:49.269 Casie Aviles: So yeah, it should be the same line, then 045.
89 00:09:52.760 ⇒ 00:10:00.949 Casie Aviles: It’s, yeah, it’s just that this sheet wasn’t updated, which, I am updating now.
90 00:10:01.900 ⇒ 00:10:04.430 Casie Aviles: After that, but yeah, I think that’s pretty much.
91 00:10:04.430 ⇒ 00:10:05.060 Uttam Kumaran: Okay.
92 00:10:06.710 ⇒ 00:10:12.829 Uttam Kumaran: So talk me through the… talk me through, like, an issue, like, that came up this morning. Like, what is your triage process?
93 00:10:14.300 ⇒ 00:10:18.749 Casie Aviles: Yeah, basically, I go and check, like, if the automation is…
94 00:10:19.670 ⇒ 00:10:22.329 Casie Aviles: working correctly, and then I would…
95 00:10:23.960 ⇒ 00:10:30.339 Casie Aviles: I would check, like, the values, I would go to Uber, I go to Uber Ads…
96 00:10:30.620 ⇒ 00:10:34.030 Casie Aviles: When there’s, like, an issue, I would go directly to the source.
97 00:10:34.780 ⇒ 00:10:39.739 Casie Aviles: And then I would check… oh, sorry, this is DoorDash, and I… yeah, should be DoorDash.
98 00:10:43.590 ⇒ 00:10:47.330 Casie Aviles: And I would go directly here, and I will just double-check, and…
99 00:10:49.240 ⇒ 00:10:51.529 Casie Aviles: So, like I mentioned, there are different…
100 00:10:51.840 ⇒ 00:10:56.620 Casie Aviles: campaign types, so I would go to the… Or a campaign type.
101 00:10:58.160 ⇒ 00:11:06.199 Casie Aviles: And then… I’ll try to diagnose, like, what’s the reason why, and…
102 00:11:06.500 ⇒ 00:11:12.710 Casie Aviles: If it’s just, like, it wasn’t updated, but it’s showing up correct here, then I would just update it here.
103 00:11:13.760 ⇒ 00:11:17.380 Casie Aviles: So they see the same… Updated values.
104 00:11:18.290 ⇒ 00:11:23.650 Casie Aviles: If it’s not, then… I would check, like, Dagster, like, if…
105 00:11:24.490 ⇒ 00:11:30.300 Casie Aviles: it’s… you know, if there’s, like, an issue with the script, then… I’m not logged in, but…
106 00:11:30.720 ⇒ 00:11:32.310 Casie Aviles: I would go there and check.
107 00:11:33.320 ⇒ 00:11:39.949 Casie Aviles: And if it’s, if it’s, like, a pipeline problem, then I’ll have to make, like, a fix there.
108 00:11:40.750 ⇒ 00:11:43.900 Casie Aviles: But I think we’ve… we’ve…
109 00:11:45.580 ⇒ 00:11:52.309 Casie Aviles: We’ve done a lot of fixes for most of, like, the issues that are occurring, so that was just an update issue.
110 00:11:53.340 ⇒ 00:11:58.249 Casie Aviles: But yeah, I think that’s… that’s, like, the gist of it.
111 00:11:58.850 ⇒ 00:12:04.529 Casie Aviles: of the issues that, like, how I would… Typically go and fix…
112 00:12:05.580 ⇒ 00:12:08.330 Casie Aviles: those issues that they would mention.
113 00:12:09.080 ⇒ 00:12:09.700 Uttam Kumaran: Okay.
114 00:12:11.680 ⇒ 00:12:21.880 Uttam Kumaran: Cool, so there’s a lot of paths here. I guess, yeah, Robert, can you set the, like, vision for me? And then I can kind of… I… I…
115 00:12:22.450 ⇒ 00:12:28.130 Uttam Kumaran: there’s just, like, I want to know what we can do. This is just an absolute mess.
116 00:12:28.530 ⇒ 00:12:34.550 Uttam Kumaran: And so I want to know, like, kind of where you… what you think we can affect short-term, like.
117 00:12:34.690 ⇒ 00:12:40.069 Uttam Kumaran: end of the year, where do you think this can go, like, next year? And then I can start to think about, like, architecture.
118 00:12:40.070 ⇒ 00:12:54.119 Robert Tseng: Yeah, so this FDA projections and the marketing performance tracker, they’re structured very similarly. To me, it’s just channel, like, day, and then, like, I don’t know, some basic, like, performance metrics, like.
119 00:12:54.580 ⇒ 00:13:06.960 Robert Tseng: whether it’s impressions or messages that were sent, then, like, open rates, conversion rates, and then the revenue, right? And, like, the problem with pulling the data out of Braze and also for
120 00:13:07.730 ⇒ 00:13:15.929 Robert Tseng: and FDA, why they don’t go to these platforms natively, is that these numbers get updated, because each one of these channels has an attribution window
121 00:13:16.310 ⇒ 00:13:26.169 Robert Tseng: probably 7 days average. So, like, once a campaign goes live, like, on Tuesday at 1.45 PM on October 7th, I’m just reading off the screen.
122 00:13:26.390 ⇒ 00:13:40.090 Robert Tseng: that it’s our 22nd birthday email campaign. Yeah, that… that campaign will… that number of… that number… the performance metrics will continue to change for, like, over… over one week. And they want to, like.
123 00:13:40.230 ⇒ 00:13:49.590 Robert Tseng: they’re trying to understand, well, like, how much of this… how, like, how is this changing, like, day to day, right? So, like, this is, to me, like,
124 00:13:50.440 ⇒ 00:13:56.530 Robert Tseng: Like, if this were, you know, properly flowing into a…
125 00:13:57.090 ⇒ 00:13:59.470 Robert Tseng: Like, a data warehouse, and we view
126 00:13:59.630 ⇒ 00:14:10.430 Robert Tseng: like, the day… the daily change, and I was able to, you know, pull out the insight that says, like, okay, for this birthday campaign that we ran on October 7th.
127 00:14:10.710 ⇒ 00:14:14.599 Robert Tseng: You know, we got 80% of the revenue in the first 2 days.
128 00:14:14.700 ⇒ 00:14:27.700 Robert Tseng: And then it took, like, you know, the rest of the week to fully realize, like, the maximum potential of the campaign. Like, that’s… that’s the insight that needs to come out of, like, the data, not just, like.
129 00:14:27.810 ⇒ 00:14:40.000 Robert Tseng: what, yeah, like, otherwise they could just download static reports and just always do historical reporting on campaign performance. Like, the purpose of updating this daily is to be able to measure, like.
130 00:14:40.190 ⇒ 00:14:54.730 Robert Tseng: kind of how these metrics are changing over the course of a campaign’s life cycle, which is not very long. It’s only one week. So, yeah, contrary to a lot of other, like, marketing channels,
131 00:14:54.790 ⇒ 00:15:08.269 Robert Tseng: like, I guess, I don’t know if Awish is on this call, but with Eden, when we look at paid performance marketing, the attribution windows are much longer. It could be, like, up to a month, and so they’re not actually having to look at this on a daily cadence. A weekly cadence is fine.
132 00:15:08.270 ⇒ 00:15:20.270 Robert Tseng: But for lifecycle channels specifically, these are emails, like, text messages, push notifications. They run a bunch of these campaigns every, you know, they run thousands a year, and they’re sending millions of messages.
133 00:15:20.270 ⇒ 00:15:39.999 Robert Tseng: And… and the life cycle of each campaign is very short. It’s only, like, a week long, max. For those push notifications, it’s even shorter. It’s probably, like, one day. It’s, like, an exploding offer. Hey, like, tap into the app and order something, you’ll get it. So, that’s try… that’s… that’s kind of… I’m just articulating, that’s, like, the purpose of, like, what…
134 00:15:40.000 ⇒ 00:15:46.400 Robert Tseng: they’re, they’re, building out of this. And so, what we’re doing here is we’re simulating, like.
135 00:15:46.440 ⇒ 00:15:51.450 Robert Tseng: Exporting daily data out of these platforms, and then being able to, like.
136 00:15:51.580 ⇒ 00:16:01.360 Robert Tseng: have some sort of change log on, like, the past week of, of, campaign data, and that, that needs… and that, and that changes, like.
137 00:16:01.360 ⇒ 00:16:13.630 Robert Tseng: for… for up to 7 days before a campaign officially ends. And that’s, like, that’s… that’s the… that’s the insight they’re trying to capture out of… out of, this work. So it is a lot of work to update, but, like.
138 00:16:13.630 ⇒ 00:16:19.940 Robert Tseng: They do all of that just so they can know, like, for that campaign that was launched on October 7th.
139 00:16:20.230 ⇒ 00:16:36.440 Robert Tseng: you know, what was the return on October 8th, on October 9th, on the 10th? And, like, being able to track that daily, and see, like, did it, you know, did it stop making money after the 10th, and it made all of the money in the first 2 or 3 days? Like, that’s…
140 00:16:36.480 ⇒ 00:16:42.229 Robert Tseng: Like, anyway, so that’s… that’s the point of doing this for across their different platforms.
141 00:16:42.850 ⇒ 00:16:56.360 Uttam Kumaran: Okay, okay, so it makes sense. So, the use case makes sense. So, tell me about the… when you… when we walked in here, they were both affected because they couldn’t get this data, right? So did we unlock the Uber and DoorDash data for them?
142 00:16:56.360 ⇒ 00:17:10.910 Robert Tseng: Yes, we unlocked the DoorDash data for them. We have it somewhat automated. We’re able to scrape most of it off and then into a separate spreadsheet that Casey can, you know, paste into this. You know, he’s not logging into everything and exporting everything.
143 00:17:10.910 ⇒ 00:17:17.550 Uttam Kumaran: But we’re not yet doing the… we’re not yet doing the real, like, hey, campaign started every day.
144 00:17:17.690 ⇒ 00:17:20.180 Uttam Kumaran: how are those changing? We’re doing the…
145 00:17:20.510 ⇒ 00:17:23.100 Uttam Kumaran: this is just, like, what it is. Like, we’re not…
146 00:17:23.109 ⇒ 00:17:24.209 Robert Tseng: Yeah, so…
147 00:17:24.210 ⇒ 00:17:33.180 Uttam Kumaran: Great. And so, do you feel that that’s, like, a… you wake up in the morning, and you’re given, like, basically, like, the velocities, and, like.
148 00:17:33.320 ⇒ 00:17:38.389 Uttam Kumaran: the projections of ActiveCampaigns, then you could send a message, or like, what do you think.
149 00:17:38.390 ⇒ 00:17:39.230 Robert Tseng: That would be ideal.
150 00:17:39.230 ⇒ 00:17:39.740 Uttam Kumaran: ends up being.
151 00:17:39.740 ⇒ 00:17:42.570 Robert Tseng: But that’s, yeah, that’s what we need to get to, so…
152 00:17:42.570 ⇒ 00:17:43.100 Uttam Kumaran: Okay.
153 00:17:43.260 ⇒ 00:17:43.730 Uttam Kumaran: Okay.
154 00:17:43.800 ⇒ 00:17:50.890 Robert Tseng: I don’t… I don’t really think… no one’s really looking at this, opening this sheet every day. They look at it on Monday, and they’re like, hey.
155 00:17:50.940 ⇒ 00:18:06.569 Robert Tseng: when I call out, like, hey, Casey, why is something, like, off? It’s because the dude logged on, Matt logged on last week, and, like, he saw that it’s different from today. It might have actually changed, like, we’re not necessarily wrong, but he’s just, like, he doesn’t necessarily get it. So, like, that’s…
156 00:18:06.620 ⇒ 00:18:16.520 Robert Tseng: Okay. You know, that’s why I’m saying a future version of this, like, I don’t really think we need to be doing this every day. Like, if we’re able to just communicate the velocities of, like.
157 00:18:17.530 ⇒ 00:18:31.539 Robert Tseng: Yeah, like, kind of what you were saying, if I were to add two more, like, fields to, like, a… if we could create one single model that captures all the different channels, all the different performance metrics, and then I would add two more columns of, like, percent,
158 00:18:31.650 ⇒ 00:18:42.429 Robert Tseng: like, like, like, present to fully realized camp… like, fully, like, fully realized revenue, projected fully realized revenue. Like, that way they can have some, like,
159 00:18:43.980 ⇒ 00:18:57.650 Robert Tseng: more, clarity on, like, oh, we expected this campaign to, you know, get all the money in the first two days, but then it actually lasted longer than we thought, and, like, that opens up more interesting conversations, so…
160 00:18:58.030 ⇒ 00:19:08.160 Robert Tseng: Like, that… I think that’s what we’re missing. I think there’s one more level of consolidation we can do here in terms of, like, from a modeling perspective, and then I need to, like, dial in on what, like, what are those, like.
161 00:19:08.300 ⇒ 00:19:13.939 Robert Tseng: two… I think it’s, like, two or three more metrics that we can look at,
162 00:19:14.120 ⇒ 00:19:17.580 Robert Tseng: To be able to communicate the velocity and the direction.
163 00:19:18.620 ⇒ 00:19:25.409 Uttam Kumaran: Okay, okay. Okay, so that makes sense. So, I think roughly…
164 00:19:25.710 ⇒ 00:19:32.039 Uttam Kumaran: Like, we probably still need to support this, but we have to move this to an environment where it’s literally, like.
165 00:19:32.270 ⇒ 00:19:35.739 Uttam Kumaran: I can just see this in one or two tables, like…
166 00:19:36.290 ⇒ 00:19:37.010 Robert Tseng: Yeah.
167 00:19:37.170 ⇒ 00:19:42.460 Uttam Kumaran: like, a data warehouse, like, I’m trying to get us out of… like, this…
168 00:19:42.570 ⇒ 00:19:49.530 Uttam Kumaran: Like, this spreadsheet stuff is going to be really hard for us to run that query that you need every morning to sort of show you details.
169 00:19:50.540 ⇒ 00:19:51.050 Robert Tseng: Yeah.
170 00:19:51.050 ⇒ 00:19:52.799 Uttam Kumaran: So, so, yeah.
171 00:19:54.010 ⇒ 00:20:12.780 Uttam Kumaran: like, part of this is, like, nailing the automations and scraping it, that’s fine, but I don’t want to… like, it can’t live in here. It has to… I either… it has to live in, like, a CSV in a data lake, or I have to push it to some tables where I can do a… run a simple query for you in the morning after we… our automations run.
172 00:20:12.780 ⇒ 00:20:31.369 Uttam Kumaran: I can calculate for you whatever you need, which is, like, the velocities, the difference between spend and revenue, so you can see, like, what is higher than expected, what is lower, and then that gives you a bunch of details, and then ideally, you should just throw that shit into AI, and then help you form, like, the analysis, you know, that’s, like.
173 00:20:31.370 ⇒ 00:20:32.080 Robert Tseng: Yeah.
174 00:20:32.690 ⇒ 00:20:33.560 Robert Tseng: Yeah.
175 00:20:35.820 ⇒ 00:20:41.320 Uttam Kumaran: So, is a good next… is a good next step for us to just architect, like, okay, what is, like.
176 00:20:41.530 ⇒ 00:20:50.129 Uttam Kumaran: one, I just want to architect, like, what the schema could look like, like, what are the different tables? What are the different dimensions and measures? Do they have, like, a…
177 00:20:51.150 ⇒ 00:20:59.020 Uttam Kumaran: like, a data warehouse, like, I can already tap into, or, like, what do you think is, like, a short-term thing for us to do to, like, try to make this real?
178 00:20:59.460 ⇒ 00:21:03.850 Robert Tseng: Yeah, I think the architecture piece would be great.
179 00:21:04.280 ⇒ 00:21:11.299 Samuel Roberts: you know, I told you they have Holistics, that’s kind of like their BI platform on top of their Azure warehouse.
180 00:21:11.300 ⇒ 00:21:22.079 Robert Tseng: it’s completely, like, unmodeled. They just… they have, like, a hundred kind of tables sitting there, so I think whatever we… and then that would be a good conversation to bring to…
181 00:21:22.190 ⇒ 00:21:24.750 Robert Tseng: the Robert on their team, and…
182 00:21:25.020 ⇒ 00:21:31.079 Robert Tseng: Just be like, this is what… like, just tell them, like, this is what we’re trying to do, and
183 00:21:31.390 ⇒ 00:21:45.459 Robert Tseng: And, you know, either he’ll… he’ll just try to take it on and do it himself, which would be fine with me, or he’ll be like, okay, whatever, I’ll let you in, you can go and add these models to the warehouse, like, which is what I want.
184 00:21:45.890 ⇒ 00:21:47.000 Samuel Roberts: So…
185 00:21:47.000 ⇒ 00:21:50.159 Robert Tseng: I think that’s… that’s probably the next step.
186 00:21:50.730 ⇒ 00:22:06.089 Uttam Kumaran: Okay, okay, cool. So then we’ll do that, and then… I mean, there’s… there’s other short-term things we could do even before that, like, Mustafa and Casey, we can take all of it, we can shove this into DuckDB, and, like, that’s completely open source, and then…
187 00:22:06.230 ⇒ 00:22:21.859 Uttam Kumaran: that can live in a pretty easy, like, that can just live in GitHub or locally. So, like, that’s even a shorter, shorter term thing we can do, where we’ll work on, like, what the architecture… so what are the core tables that you need? We can shove all that into…
188 00:22:21.990 ⇒ 00:22:27.600 Uttam Kumaran: a DuckDB open file format, and then that way, at least I can run the query on that.
189 00:22:27.730 ⇒ 00:22:33.620 Uttam Kumaran: and then where are the… where are these from? Are these all running in Dagster, Casey?
190 00:22:34.330 ⇒ 00:22:39.669 Casie Aviles: Yes, there are some… most of them are working in Dagster, there are some…
191 00:22:40.050 ⇒ 00:22:48.390 Casie Aviles: scripts that do not, live on Dagster, they’re on… on Windmill, because that’s where it works right now.
192 00:22:48.560 ⇒ 00:22:50.339 Casie Aviles: So we settled it.
193 00:22:50.340 ⇒ 00:22:51.499 Samuel Roberts: Those are automation stuff?
194 00:22:51.930 ⇒ 00:22:54.760 Casie Aviles: Yeah, some of the browser automation stuff, like…
195 00:22:54.940 ⇒ 00:22:57.370 Casie Aviles: Yeah, like, that’s one of the…
196 00:22:58.310 ⇒ 00:23:06.680 Casie Aviles: workarounds that we had to do, like, it’s not, like, the best, because it’s being scraped, you know? And then something… if something changes, then…
197 00:23:07.030 ⇒ 00:23:12.590 Casie Aviles: We’ll have to… Yeah, like… Work on that again, and…
198 00:23:13.320 ⇒ 00:23:21.050 Casie Aviles: But yeah, like, unfortunately, we don’t really have, like, the API access for those, like the DoorDash and Uber specifically.
199 00:23:21.370 ⇒ 00:23:27.390 Casie Aviles: So, we’re just, yeah, we’re stuck with just, you know, browser automations.
200 00:23:27.770 ⇒ 00:23:41.470 Uttam Kumaran: Okay, so I have a bunch of things I want to ask. One is, like, I want to move to their infra for as much of this shit as possible, so I gotta ask them, like, what they’re using for orchestration. Ideally, they’re not, and then we can own that. Second piece is I’m gonna ask them, yeah, we need to land this data somewhere.
201 00:23:41.660 ⇒ 00:23:48.029 Uttam Kumaran: So, like, let me know where I can do that. Even sh… that, again, may take a few weeks for me to, like.
202 00:23:48.150 ⇒ 00:23:59.759 Uttam Kumaran: puncture into that. So, short term, I think I’ll propose this, like, we’ll propose a rough schema, and ideally, I’d like to see if we can land this in DuckDB. You can use Daxter to do that.
203 00:23:59.760 ⇒ 00:24:15.700 Uttam Kumaran: pretty easily. And then I, at minimum, want to be able to produce a query for Robert that after we update… after the data updates, it shows the velocities and the accelerations of campaigns. So there’s something on Monday, Tuesday, Wednesday, that just shows, like.
204 00:24:15.840 ⇒ 00:24:20.469 Uttam Kumaran: How much we spent, how much revenue is coming in, and the day-over-day
205 00:24:20.660 ⇒ 00:24:24.740 Uttam Kumaran: week-to-date accelerate… basically, like, the rates of change.
206 00:24:24.960 ⇒ 00:24:43.490 Uttam Kumaran: So that’ll be, like, if we can get a win on that by, like, Monday, that’ll be a huge step forward. The best way for Robert to go sell us on major infra changes is by selling the output. So the output looks really good, and then we’re like, hey, in order to do this, we had to do XYZ, then they’ll be like, do XYZ.
207 00:24:43.600 ⇒ 00:24:54.219 Uttam Kumaran: You know, so… so that’s what I want to try to get to. Can I… can you make… Casey, can you send me these… the URLs? And so I’m just gonna… I just want to make sure that I can access this.
208 00:24:54.820 ⇒ 00:24:56.120 Uttam Kumaran: On my side.
209 00:24:56.690 ⇒ 00:25:00.540 Uttam Kumaran: Yeah, you could just Slack it to me, and so I’ll just double check that I… yeah.
210 00:25:00.540 ⇒ 00:25:02.630 Casie Aviles: They’re also in the SOP, but…
211 00:25:04.550 ⇒ 00:25:05.200 Uttam Kumaran: Okay.
212 00:25:06.150 ⇒ 00:25:07.569 Casie Aviles: We can send it again.
213 00:25:08.630 ⇒ 00:25:09.270 Uttam Kumaran: Okay.
214 00:25:09.400 ⇒ 00:25:12.340 Uttam Kumaran: Cool.
215 00:25:12.450 ⇒ 00:25:22.819 Uttam Kumaran: Alright, I feel like that’s okay. Anything else, like, we want to talk about planning-wise this week? Like, I feel like I have a rough understanding of, like… so, I think, Casey, the biggest thing is just keep…
216 00:25:23.020 ⇒ 00:25:30.660 Uttam Kumaran: pushing on the daily stuff, and then by the end of this week, I want to have a plan forward. I don’t know whether we’ll be able to get you entirely out of
217 00:25:30.830 ⇒ 00:25:33.819 Uttam Kumaran: Some of the daily stuff, until…
218 00:25:34.100 ⇒ 00:25:37.280 Uttam Kumaran: I can get our new, like, architecture proposed.
219 00:25:38.550 ⇒ 00:25:40.500 Uttam Kumaran: And ideally, basically, again, like.
220 00:25:40.740 ⇒ 00:25:50.029 Uttam Kumaran: that new architecture can support these spreadsheets. Like, the spreadsheets should pull from just one raw source, and we can make those changes as needed, but…
221 00:25:50.430 ⇒ 00:25:56.590 Robert Tseng: And Tom, I’ll book time with the… would you think Thursday would be a good time to book time with, I’m just gonna grab.
222 00:25:56.590 ⇒ 00:25:57.080 Uttam Kumaran: Yes.
223 00:25:57.080 ⇒ 00:25:58.939 Robert Tseng: sometime on the dude’s calendar, okay.
224 00:25:59.520 ⇒ 00:26:00.440 Uttam Kumaran: Yeah, yeah.
225 00:26:00.730 ⇒ 00:26:01.580 Robert Tseng: Right.
226 00:26:07.010 ⇒ 00:26:11.439 Uttam Kumaran: Okay, that’s kind of all I wanted to see here, so yeah, I wouldn’t…
227 00:26:11.630 ⇒ 00:26:23.869 Uttam Kumaran: work on any more optimizations, let me… give me, like, a day or two to, like, put my thoughts down on an architecture, and then we can align on, like, what… what work needs to happen to do that by, like, Wednesday.
228 00:26:24.170 ⇒ 00:26:27.839 Uttam Kumaran: And then… yeah, we should be in a better place here.
229 00:26:30.420 ⇒ 00:26:31.050 Casie Aviles: Okay.
230 00:26:32.120 ⇒ 00:26:32.800 Uttam Kumaran: The lovely?
231 00:26:32.800 ⇒ 00:26:33.420 Samuel Roberts: Think about that?
232 00:26:33.420 ⇒ 00:26:35.820 Uttam Kumaran: It’s like, yeah, yeah, yeah, go, go.
233 00:26:36.150 ⇒ 00:26:40.500 Samuel Roberts: Let me know if I can help with that at all, I’m sure you probably got the data stuff sorted, but if there’s any automation stuff that I can…
234 00:26:40.620 ⇒ 00:26:43.020 Samuel Roberts: help with, like, I’m happy to.
235 00:26:43.020 ⇒ 00:26:45.890 Uttam Kumaran: Totally. Yeah, and the lovely thing about this is, like.
236 00:26:46.090 ⇒ 00:26:54.959 Uttam Kumaran: I want to create an environment where Robert can plug a lot of this into AI, plus his own sense, and then gives him a great, easy output every morning.
237 00:26:55.260 ⇒ 00:27:03.929 Uttam Kumaran: that’s, like, this is gonna… this will buy us a lot of credibility with them, if we can actually act… act on this daily. I guess, Robert, my last question is, like, what are some of the actions, like.
238 00:27:04.130 ⇒ 00:27:06.650 Uttam Kumaran: The team can take if,
239 00:27:06.840 ⇒ 00:27:10.689 Uttam Kumaran: If they notice something is accelerating versus decelerating, like, midweek.
240 00:27:12.090 ⇒ 00:27:14.839 Robert Tseng: Yeah, I think,
241 00:27:16.600 ⇒ 00:27:22.680 Robert Tseng: So yeah, I guess that spreadsheet you were just sharing about, like, the campaigns itself, so…
242 00:27:24.520 ⇒ 00:27:28.659 Robert Tseng: I guess, like, what you would notice is, so certain…
243 00:27:29.300 ⇒ 00:27:33.950 Robert Tseng: like, I’m working on this kind of, taxonomy right now, but, like.
244 00:27:34.160 ⇒ 00:27:50.349 Robert Tseng: Yeah, for all these different images, like, the birthday thing, like, that’s like an evergreen campaign, like, they run a bunch of those, so it’s very easy to benchmark against that, so once we have enough data, it’ll be like, hey, this birthday campaign is, like, underperforming, whatever, like.
245 00:27:50.350 ⇒ 00:27:54.899 Robert Tseng: And, I think that’s, that’s, like, something we can flag to the team, like, right away.
246 00:27:54.900 ⇒ 00:27:59.840 Robert Tseng: So, like, those are great, like, you know, Slack alerts that I want to start flagging to the Insomnia team.
247 00:27:59.840 ⇒ 00:28:19.560 Robert Tseng: For the other stuff, like this caramel Obsessed, this is, like, a new format of campaigns that they’re running. They’re basically trying to just take their core products and kind of design a series around them. Like, they already… they did one for Peanut butter that did pretty well, so now they’re doing it for their caramel, like, products, and they’re probably gonna do it across their other ones, so…
248 00:28:19.620 ⇒ 00:28:25.340 Robert Tseng: I think this is… you know, the whole point of, like, what we’re able to point out is, like.
249 00:28:25.600 ⇒ 00:28:41.069 Robert Tseng: how… how much can we predict the performance of a campaign when it gets launched? Like, using historical data and, like, being able to benchmark against, like, stuff? So, that’s… that’s, like, the rigor that I’m trying to build up for this marketing team. They just kind of just…
250 00:28:41.960 ⇒ 00:28:48.580 Robert Tseng: they just shoot off whatever the heck they want, and they don’t really know, like, what works or not. They’ll just, like, sit in meetings and be like.
251 00:28:48.870 ⇒ 00:29:04.580 Robert Tseng: telling the CMO, oh yeah, the Taylor Swift campaign was really good, and then I showed them in the data, it was like, actually, it was, like, it didn’t… didn’t really… didn’t really generate as much revenue as you think it did. So, like, that’s… that’s… that’s why… that’s kind of where we’re… where we’re headed, so…
252 00:29:05.660 ⇒ 00:29:20.290 Robert Tseng: yeah, I think no one’s gonna know, we’re gonna be able to set the benchmarks and be able to flag, like, when, you know, when something is underperforming, overperforming, whatever, and, like, that… that should drive their team to take some actions off of what we’re sharing with them.
253 00:29:21.240 ⇒ 00:29:22.240 Uttam Kumaran: Okay, okay.
254 00:29:22.630 ⇒ 00:29:23.260 Robert Tseng: Yeah.
255 00:29:25.910 ⇒ 00:29:37.849 Robert Tseng: It is kind of crazy. Like, I don’t think I’ve ever run… to me, this is just, like, a portfolio management exercise. Like, I’ve never really run it at this frequency before. Like, I don’t know if really, like, weekly optimization makes any sense, but…
256 00:29:38.220 ⇒ 00:29:42.290 Robert Tseng: I don’t know, like, this is something that we’re… I’m just, like, applying…
257 00:29:42.590 ⇒ 00:29:45.160 Robert Tseng: kind of how I would approach… No, I mean, ideally…
258 00:29:45.160 ⇒ 00:29:45.570 Uttam Kumaran: Yeah.
259 00:29:45.570 ⇒ 00:29:46.730 Robert Tseng: in this. Yeah.
260 00:29:46.730 ⇒ 00:29:48.539 Uttam Kumaran: Ideally, they should…
261 00:29:48.720 ⇒ 00:29:59.589 Uttam Kumaran: they should quickly be able to understand, like, within 2 days, whether a campaign’s working, and cut spend, you know, so there’s probably some savings. And then second is it’s gonna be all the, like.
262 00:29:59.870 ⇒ 00:30:18.359 Uttam Kumaran: doing the taxonomy, getting all the data in one place will allow you to run easy queries to say, like, what’s performed well, and what should we just continue to do? And it sort of attacks, like, because marketing people always come up with the next hottest thing, and then it’s up to, like, the data team to be like, well, is this going to work given our past performance, right? So…
263 00:30:18.870 ⇒ 00:30:24.879 Uttam Kumaran: Yeah. Intraday optimizations are not that often, unless you’re doing really heavy programmatic advertising.
264 00:30:26.170 ⇒ 00:30:26.690 Robert Tseng: Yeah.
265 00:30:26.690 ⇒ 00:30:27.070 Uttam Kumaran: like.
266 00:30:27.070 ⇒ 00:30:27.789 Robert Tseng: So, I mean, like, thank you.
267 00:30:27.790 ⇒ 00:30:28.510 Uttam Kumaran: Facebook up and down.
268 00:30:28.510 ⇒ 00:30:33.939 Robert Tseng: There aren’t that many levers. Like, I, you know, if I could just, like, kind of soapbox for, like, another minute on this, like.
269 00:30:34.050 ⇒ 00:30:46.099 Robert Tseng: you know, we’re always, like, thinking about our business. What are some, like, low-lift ways for us to get big wins? Like, I feel like this is almost, like, I’m willing to take on this word, because this, to me, is like the…
270 00:30:46.320 ⇒ 00:30:50.500 Robert Tseng: It’s, it’s like the conversion rate optimization, like.
271 00:30:50.960 ⇒ 00:31:00.279 Robert Tseng: gimmick for… for the… for us as data people working with marketers. There’s no cost, like, they’re just sending emails and messages, like, it’s not… it doesn’t cost very much to send.
272 00:31:00.280 ⇒ 00:31:24.859 Robert Tseng: So they’re not really making any, like, adjustments. They’re not changing spend or whatever off of our insights, but… so this is… because these are all own channels, like, the risk is super low, and we’re just, like, looking for ways to basically help them, like, get money that doesn’t really cost them much to go and get. So… and there aren’t that many levers. Is it changing subject lines, changing the creative, adding a header, a footer, or maybe they need to, like, add something more interactive
273 00:31:24.860 ⇒ 00:31:26.689 Robert Tseng: into the email. So,
274 00:31:26.690 ⇒ 00:31:27.259 Uttam Kumaran: It’s like…
275 00:31:27.630 ⇒ 00:31:31.459 Robert Tseng: But, like, yeah, I think that’s kind of interesting to me.
276 00:31:31.610 ⇒ 00:31:42.000 Robert Tseng: that if we can become, you know, a team that can advise them on that stuff, like, that, to me, seems like an easy win we can bring to a lot of other companies. So, anyway, that’s.
277 00:31:42.000 ⇒ 00:31:42.500 Uttam Kumaran: Yeah.
278 00:31:42.790 ⇒ 00:31:43.489 Robert Tseng: I thought.
279 00:31:44.290 ⇒ 00:32:03.389 Uttam Kumaran: Okay, cool. You know, this is great. I mean, this is a complete mess, so I’m glad. Well, we’re gonna help them, like, make this a lot easier. But again, like, I want… I want to put the data stuff on rails. Like, I don’t want… I want us to drive towards one consistent schema, and then, ideally, you can start to help that team also query the database and
280 00:32:03.510 ⇒ 00:32:16.080 Uttam Kumaran: get these answers to, like, versus just waiting on you to do it, like, so… Okay, cool. So then I’ll… I’ll find out, once I meet with this IT guy, I’ll find out, like, what’s possible in Fresight. There’s a lot of ways for us to get around this, though, so… yeah.
281 00:32:16.550 ⇒ 00:32:18.550 Robert Tseng: Cool. Appreciate it. Thanks, guys.
282 00:32:18.550 ⇒ 00:32:20.379 Uttam Kumaran: Okay, great. Thank you, guys.
283 00:32:20.380 ⇒ 00:32:21.390 Samuel Roberts: Talk to you soon.
284 00:32:21.390 ⇒ 00:32:22.190 Casie Aviles: Thank you.
285 00:32:22.460 ⇒ 00:32:23.340 Mustafa Raja: Thank you.