Meeting Title: Brainforge x Omni Partnership Discussion Date: 2025-09-30 Meeting participants: Jake Nathan, Omni Notetaker, Tamara John, Jamie Davidson, Uttam Kumaran
WEBVTT
1 00:05:05.870 ⇒ 00:05:07.890 Tamara John: Hello, how are you?
2 00:05:08.530 ⇒ 00:05:10.119 Jake Nathan: Hey, doing well, how are you doing?
3 00:05:10.120 ⇒ 00:05:11.869 Tamara John: Doing well, thanks.
4 00:05:12.990 ⇒ 00:05:17.390 Jake Nathan: Yeah, thanks for making time. Oh, we’ll be able… Cat?
5 00:05:17.610 ⇒ 00:05:20.370 Tamara John: This is, yeah, this is Walter, my shadow.
6 00:05:20.720 ⇒ 00:05:21.860 Jake Nathan: Hello, Walter.
7 00:05:23.020 ⇒ 00:05:23.960 Jake Nathan: Aww.
8 00:05:24.830 ⇒ 00:05:25.380 Jake Nathan: model.
9 00:05:26.480 ⇒ 00:05:28.239 Jake Nathan: Honestly, I need to get a cat.
10 00:05:28.390 ⇒ 00:05:33.949 Jake Nathan: Seems like it would be a great companion when I… because I’m just working by myself so much, so it would be great.
11 00:05:34.480 ⇒ 00:05:44.300 Tamara John: Oh my god, I cannot… we have two, but I had one cat who, like, passed away after, like, she was super old, and then we got another one
12 00:05:44.680 ⇒ 00:05:50.340 Tamara John: She seemed… Like, she wanted a friend, and we got… we ended up getting two, and like, ugh.
13 00:05:50.900 ⇒ 00:05:55.380 Tamara John: Jamie’s a cat person, too. Cannot say enough. Cats are great. Strongly recommend.
14 00:05:56.670 ⇒ 00:05:58.090 Jake Nathan: Hey, Jamie, how’s it going?
15 00:05:58.090 ⇒ 00:05:59.930 Jamie Davidson: For me, I’m good. Have yourself.
16 00:06:00.660 ⇒ 00:06:04.570 Jake Nathan: Good, good. Are you a… are you a cat person as well? Do you have any cats?
17 00:06:04.570 ⇒ 00:06:20.689 Jamie Davidson: I don’t have any cats anymore. I had two for a long time. They passed away a little bit ago, too. I’ve got a dog now, but yeah, I do love cats. I think… yeah, one… I think we’ll get cats again at some point.
18 00:06:20.910 ⇒ 00:06:32.509 Jake Nathan: I love it, I love it. Well, yeah, thank you again for making time. I think Tom might join us in a little bit here as well. Sure. But, yeah, I’m excited, just gonna…
19 00:06:32.540 ⇒ 00:06:49.959 Jake Nathan: roll through some questions, and then, yeah, we can… we can go from there, and then after our interview, I’m gonna, you know, go listen to it again, rewatch, and come up with a content piece that I’ll send over to y’all, just to make sure you feel great about it, and then we’ll go from there.
20 00:06:50.090 ⇒ 00:06:50.850 Jamie Davidson: Sounds great.
21 00:06:50.850 ⇒ 00:07:09.719 Tamara John: And then I think also just a little context. I know, Jake, that you just joined, the Brainforge team, but I think, like, your founders and stuff have experience with GB, and, like, Greg and some of our sales leadership and stuff, and so this was kind of an opportunity for us to start talking a little bit more about the partnership.
22 00:07:10.640 ⇒ 00:07:12.879 Tamara John: At least that’s my understanding.
23 00:07:12.880 ⇒ 00:07:29.099 Jake Nathan: Yep, definitely, yeah. Perfect. Time for Utom, to join, and, he’d definitely love to talk more about that, and yeah, from every… everything that I’ve heard from him, you know, they’re obviously big fans of Omni, so, and we’ve implemented it in quite a few of our clients, so…
24 00:07:29.100 ⇒ 00:07:38.979 Jake Nathan: We’re kind of excited, yeah, like you said, to expand that partnership and just learn more about the features that y’all are excited about, because we obviously want to then bring that to our clients.
25 00:07:39.620 ⇒ 00:07:40.550 Tamara John: Awesome.
26 00:07:41.310 ⇒ 00:07:42.840 Jake Nathan: tool.
27 00:07:42.840 ⇒ 00:08:02.269 Jake Nathan: Well, yeah, I also did a quick interview with Utam as well, to kind of get a sense from him, from how he uses Omni, but, yeah, Jamie, with your, you know, background at Looker, too, we have a lot of customers, we’ve noticed, that we’re, talking to that are deciding between something like Looker, Tableau, Omni.
28 00:08:02.370 ⇒ 00:08:05.199 Jake Nathan: Why would you choose Omni over those solutions?
29 00:08:05.200 ⇒ 00:08:10.989 Tamara John: Oh, actually, one second, I don’t think… oh, okay, never mind, we are recording. Mine didn’t say recording yet, so I just wanted to make sure…
30 00:08:10.990 ⇒ 00:08:15.180 Jake Nathan: Okay, I’m recording on my end, so, I could send you the recording afterwards.
31 00:08:15.180 ⇒ 00:08:20.209 Tamara John: Okay, no, perfect. I saw my record button was still, like, open, and I was like, oh no, I don’t want you to…
32 00:08:20.210 ⇒ 00:08:23.839 Jake Nathan: Oh, yeah, no, I appreciate that. Yeah, no, thank you, that’s always.
33 00:08:24.140 ⇒ 00:08:25.320 Jamie Davidson: Amazing, brother.
34 00:08:25.320 ⇒ 00:08:28.040 Jake Nathan: experiences, and that’s on the phone afterwards.
35 00:08:28.040 ⇒ 00:08:35.440 Tamara John: Like, I do… I do these interviews, too, so I was like, oh, that would have been a surprise. Okay, all good. I will… I will stop messing things up. I will be quiet.
36 00:08:35.890 ⇒ 00:08:36.510 Jamie Davidson: I agree.
37 00:08:36.510 ⇒ 00:08:37.160 Jake Nathan: Thank you.
38 00:08:37.169 ⇒ 00:09:02.139 Jamie Davidson: Yeah, so, I mean, the highest order bit, you know, we are a BI tool and data analytics platform, like, you know, Looker, Tableau or a bunch of others, too. But really, the core difference for Omni, we want to make it, you know, super easy for everyone to consume data, and the way we do that is basically have interfaces that anyone can access, so sort of the most technical, least technical.
39 00:09:02.139 ⇒ 00:09:26.339 Jamie Davidson: folks, you know, and really the way that sort of we’re differentiated is that all of these interfaces are actually really, you know, interrelated. So, like, you know, we’ve got, like, Tableau is a very strong visual discovery tool. You know, Looker is a really strong governed tool, sort of got a semantic model at its, at its core, sort of allows for democratized access, too. You know, it turns out.
40 00:09:26.339 ⇒ 00:09:28.439 Jamie Davidson: You want rigorous…
41 00:09:28.439 ⇒ 00:09:53.059 Jamie Davidson: visual discovery on top of government analytics, and you want to go even further. You want to be able to go and augment the, you know, the visualizations with, like, Excel-like, you know, syntax to make it super easy. You may want to use SQL to directly, you know, augment. You may want to use natural language to create new calculations, or to create new queries, or to summarize data, or recommend new questions, and whatnot. All of these things are kind of, you know, tightly interrelated, and you don’t want to
42 00:09:53.059 ⇒ 00:10:17.919 Jamie Davidson: sort of tailored, siloed solutions for them, because they’re so, so, so tied together. And so, you know, we want to be better than, you know, Looker at governance, we want to be better than Tableau at, you know, Viz, we want to be better than Excel at, you know, spreadsheets, we want to be better than ThoughtSpot at Natural Language. Not because we’re necessarily better in any one dimension, but because we’ve got all these interrelated functionality and sort of this… it’s a tightly integrated paradigm.
43 00:10:17.919 ⇒ 00:10:22.979 Jamie Davidson: The little aspirits. That’s sort of the core difference for folks.
44 00:10:23.500 ⇒ 00:10:30.090 Jake Nathan: Yeah, so yeah, at a high level, you’re able to kind of connect those pieces, and… and how… how are you able to do that?
45 00:10:30.480 ⇒ 00:10:55.040 Jamie Davidson: I mean, basically, the core of Omni is a semantic model, so you’ve defined data relationship and business logic in a sort of single, shareable place, and then we allow for, basically, augmentation of that semantic model in basically any interface you want. So, like, you can use SQL, you can write Excel functions, you can do, like, you know, lightweight visual discovery where, you know, create a new
46 00:10:55.740 ⇒ 00:11:20.639 Jamie Davidson: count distinct, or create a new sum on top of a numeric field, and all of it effectively is done in a siloed workbook-like environment, but that workbook is actually tied to that sort of data model, and there’s a workflow to say, hey, I find, you know, I’m changing the definition of our gross margin, I’m going to update it in this workbook, I’m not ruining anyone’s reports, but if I decide that this is something that’s actually, you know, in fact.
47 00:11:20.640 ⇒ 00:11:45.620 Jamie Davidson: usable. There’s basically a mechanism to sort of take that, recommend it for what we call as promotion, but basically sharing with the rest of the organization, you know, that sort of shared logic and whatnot. That’s the sort of the core piece, where you basically have a mechanism to sort of have these sort of, you know, single definitions, but then, you know, end users can go beyond them, and then there’s sort of a feedback loop to basically improve that experience.
48 00:11:45.620 ⇒ 00:11:47.289 Jamie Davidson: Basically, through usage.
49 00:11:47.720 ⇒ 00:11:59.589 Jake Nathan: Yeah, that makes sense to me, and you kind of alluded to this a little bit, but it seems like, with a tool like Omni, there’s probably multiple different user personas who might be touching it. So, like, for example, a finance team.
50 00:11:59.760 ⇒ 00:12:00.180 Jamie Davidson: Yeah.
51 00:12:00.180 ⇒ 00:12:07.210 Jake Nathan: and they don’t have as big of a data background? Like, how does Omni support that for users who might not be as data savvy?
52 00:12:07.600 ⇒ 00:12:20.649 Uttam Kumaran: Yeah, and I just want to chime in with one thing. Hey, Jamie, and hey Tamara, nice to meet you, by the way. Thanks for taking the time. We’re big fans of the product. I mean, for me, what sold me on Omni in the first demo was the spreadsheets feature, because.
53 00:12:20.650 ⇒ 00:12:21.180 Jamie Davidson: Yeah.
54 00:12:21.180 ⇒ 00:12:24.819 Uttam Kumaran: We’ve led data teams for a while, and as I’m sure as you know.
55 00:12:24.820 ⇒ 00:12:28.299 Uttam Kumaran: The finance user is, like, the elusive person that never gets…
56 00:12:28.300 ⇒ 00:12:44.920 Uttam Kumaran: into the system, and you guys crushed it by just basically building spreadsheets into the tool, so it really, like, crushes that objection, and, like, it’s obviously something that, as a Looker user and a Tableau user for, you know, almost 10 years, not something that we could have ever done.
57 00:12:44.940 ⇒ 00:12:48.409 Uttam Kumaran: Right, we, you know, and the ability to also
58 00:12:48.550 ⇒ 00:13:04.679 Uttam Kumaran: a big, also, problem is syncing from Google Sheets back into the database as well. So, yeah, I think to Jake’s, you know, question, I think we’d just be interested in hearing, like, how you’re thinking about supporting, you know, those people as well, and as, like, kind of, like, where the vision for that came from.
59 00:13:05.130 ⇒ 00:13:30.049 Jamie Davidson: Yeah, I mean, very literally the, sort of the start for the company, and even the name, Omni, is like, all of your data for all of your users, effectively, is really what, sort of, Omni is meant to be an allusion to. So, sort of, wanting to satisfy, really, that full spectrum, you know, folks writing SQL to Excel to point-and-click, natural language, too. We think, you know, the default feature for BI up till now really has been that export
60 00:13:30.050 ⇒ 00:13:54.619 Jamie Davidson: to Excel. It’s like download as CSV, export to Excel. You know, that is sort of the, standard, you know, BI approach. And so, we want to basically bring an interface to folks that, you know, unlike other tools, like, there are certainly other tools that are, you know, sort of spreadsheet-like, but basically taking, you know, what feels sort of like a spreadsheet, but basically redefining it.
61 00:13:54.620 ⇒ 00:14:19.169 Jamie Davidson: creating new functions, creating it so that it’s only, like, commoner, you know, displays of data, as opposed to, like, hey, I could reference, you know, row, you know, A1, B1, or something, and, like, do… do, you know, sort of completely custom things. We… we wanted to say, hey, let’s make it very literally… the goal is we could point people to Excel docs, like, literally the functional definitions,
62 00:14:19.170 ⇒ 00:14:44.119 Jamie Davidson: from Excel, like, publish, and be able to use them, and basically build up, sort of, the sort of the tight enough integration so that it’s actually, you know, basically, you can go from that Excel function to a SQL-backed, you know, query to a governed data model, too, and sort of have it be tightly integrated so that you can kind of have sort of the best of all of the worlds. So, sort of you can enable that operations person, the finance person, too, that knows Excel.
63 00:14:44.120 ⇒ 00:15:05.039 Jamie Davidson: in an interface that makes most sense to them, too. And then, sort of the similar, you know, problem statement, or similar, sort of user base, like that, you know, bringing the data back into the workflow. We’ve got sort of the same process as well there. So, like, there’s lots of reasons why data might not be present inside the data warehouse. Maybe they’re, you know.
64 00:15:05.040 ⇒ 00:15:14.080 Jamie Davidson: plans, you know, financial plans that you’ve got, and you’re comparing them to actuals, and you want to compare them together in the same visualization, or, like, maybe there’s
65 00:15:14.140 ⇒ 00:15:38.280 Jamie Davidson: you know, sales quota numbers, or, marketing campaign mappings from, like, you know, these are brand campaigns, and these are direct consumer, and these are, you know, experimental, you know, so there’s, like, lots and lots of reasons why, that basically it won’t be present in any of the source systems, like, maybe they’re not coming from Salesforce, it’s not coming from, like, your ERP system, or it’s not coming from, you know, your marketing systems, too, and basically what we allow for
66 00:15:38.280 ⇒ 00:15:43.889 Jamie Davidson: It’s sort of easy upload so that people can do it, and honestly, it’s done in a way where,
67 00:15:43.890 ⇒ 00:16:07.829 Jamie Davidson: it sort of has this progressive hardening as well. So, you know, folks can, you know, you can have the marketer or the finance person add plans in a workbook environment that is siloed. It is, you know, sort of one-off, and, you know, they can collaborate, of course, in sort of a modern way, but, you know, it’s not like polluting the CMO’s dashboard or the CFO’s dashboard, but then the
68 00:16:07.830 ⇒ 00:16:31.820 Jamie Davidson: to the extent that, you know, this is something that’s actually supposed to be reusable and should be standardized and governed, there’s sort of a mechanism to take it from that workbook and basically add that directly to the data warehouse, so you can do things like, you know, join it directly in SQL inside Snowflake, or in BigQuery, or Databricks, or whatever the tool might be, to be able to sort of have standardized definitions of these metrics as well.
69 00:16:35.420 ⇒ 00:16:39.509 Jake Nathan: Awesome. Utam, do you wanna… I know you had a few questions, do you wanna…
70 00:16:39.820 ⇒ 00:16:43.720 Uttam Kumaran: Yeah, I guess my next question was gonna kind of be,
71 00:16:44.920 ⇒ 00:16:54.389 Uttam Kumaran: I guess we can go maybe two ways. Maybe one of them, like, a more simpler one is, like, if you guys are innovating at all on the data security side, like, I mean, I think most of our clients are familiar with
72 00:16:54.390 ⇒ 00:17:08.900 Uttam Kumaran: role-based access control, typical Looker governance, like, is there… are there any sort of things you guys are thinking about doing differently on the security and governance side, whether that’s on observability, whether that’s actually, like, the mechanism? Yeah, interested to hear about that.
73 00:17:08.900 ⇒ 00:17:33.229 Jamie Davidson: Yeah, so, I mean, at our core, we’ve got sort of the, you know, row-level, column-level access controls tied to roles that are, you know, going to be integrated with, you know, SCIM, or SSO, or sort of basically whatever mechanism users might want. And it’s all governed by a data model that can be integrated with software, you know, software development best practices. So, it can be integrated with conversion control and steps or change management on it, too.
74 00:17:33.230 ⇒ 00:17:37.770 Jamie Davidson: You can have… we’ve got… certainly got customers that have, you know, CICD,
75 00:17:38.000 ⇒ 00:17:50.470 Jamie Davidson: sort of frameworks built in, too, so they require review of a data person, and, like, you know, and, you know, the, sort of the piece, I’d say, there’s sort of, like, data permission controls, there’s also, content.
76 00:17:50.500 ⇒ 00:18:14.949 Jamie Davidson: controls or sort of metadata, you know, controls that are… it’s a bit orthogonal to data directly, like, hey, just because a user has access to data doesn’t necessarily mean that they are allowed to see the dashboards, you know, necessarily on it. Or, like, some dashboards may have sensitive things. Even the metadata can be sensitive, too. So we sort of have robust controls along those lines as well. I’d say the pieces that are a little bit different, we’ve got basically a much clean… sort of in the
77 00:18:14.950 ⇒ 00:18:23.800 Jamie Davidson: like, Looker World, we’ve got a much cleaner, abstraction, of… Okay. So you can do things like departmental models, or like, honestly.
78 00:18:23.830 ⇒ 00:18:32.409 Jamie Davidson: group models, or per-user models. So, you know, we’ve got lots of customers, in particular in embedded contexts, where, you know, they’re building
79 00:18:32.410 ⇒ 00:18:47.500 Jamie Davidson: customer-facing dashboards, so Omni powers it to show, you know, the analytics to their customers, where customer A for sure can’t see customer B’s data, where they will… and they may even have things like slightly differing data models and the like.
80 00:18:47.500 ⇒ 00:18:52.650 Jamie Davidson: And so we’ve got sort of a much, much more robust control over that than, like.
81 00:18:52.650 ⇒ 00:18:53.070 Uttam Kumaran: Okay.
82 00:18:53.070 ⇒ 00:19:12.849 Jamie Davidson: you know, anyone else too, but, like, at the highest level, you know, we started from day zero thinking about, you know, enterprise-grade BI, and really that’s sort of, you know, we’ve, you know, we’ve done it, we’ve done it before, so sort of trying to improve upon some of the, you know, sort of the paradigms that we’ve, you know, we’ve been part of.
83 00:19:12.850 ⇒ 00:19:26.399 Tamara John: Yeah, and I think kind of just something just to call out on the security front, which was covered in everything that Jamie said, but I think a lot of other companies kind of glance over it, is because of the semantic model and all the security and permissions and everything that Jamie just talked about, like, all of that applies to AI as well.
84 00:19:26.400 ⇒ 00:19:35.829 Tamara John: Versus it being, like, thinking about BI, and then thinking about, kind of, like, your AI access as an add-on, or, like, something different, or, like, a different platform, like.
85 00:19:35.830 ⇒ 00:19:54.589 Tamara John: all of that is going to transform and just kind of come through, like, however you’re interacting with the product. So, if, you know, someone’s using spreadsheets, if someone… or spreadsheets in AMI, if no one’s writing SQL or doing point-and-click, or if someone’s, like, interacting with the AI chat, or an AI, like, agent, one of our AI agents in a dashboard, or through, like, our MCP server.
86 00:19:55.010 ⇒ 00:19:59.590 Tamara John: all of those, you know, security and permissions that Jamie just talked about are going to be respected.
87 00:20:00.300 ⇒ 00:20:10.790 Uttam Kumaran: Okay. Yeah, that’s a good point. I mean, that’s what we have been telling clients, too, is that if we centralize our… all of our governance, semantic layer, and Omni, and then the MCP basically comes out of that.
88 00:20:10.790 ⇒ 00:20:20.300 Jamie Davidson: Yeah, yeah, exactly, and it’s nice because unlike, like, basically every, literally every other tool that is writing, you know, custom SQL effectively, like text-to-text.
89 00:20:20.300 ⇒ 00:20:20.760 Uttam Kumaran: Yes.
90 00:20:20.760 ⇒ 00:20:45.690 Jamie Davidson: even though people are sort of trying to constrain. It means that we sort of… we tie this stochastic, non-deterministic system to a purely deterministic execution, so, like, you cannot, you know, we’ve got a bunch of customers using us in embedded analytics, where, you know, they’re providing to their customers, and the LLM cannot hallucinate and show customer B, you know, customer A’s data because it forgot a filter or something like that. Like, it cannot, like, very literally can’t do that.
91 00:20:45.690 ⇒ 00:20:50.219 Jamie Davidson: operating within, sort of, the constraints of, the sort of the deterministic query execution system.
92 00:20:50.630 ⇒ 00:21:00.869 Uttam Kumaran: Great. Yeah, and I guess my next question was gonna kind of be about the journey of, like, stitching together semantic layer, LLM and sort of the visual product. I mean, again, like.
93 00:21:01.010 ⇒ 00:21:07.990 Uttam Kumaran: I’ve been sort of demoing text to SQL, data analysis, AI products.
94 00:21:08.090 ⇒ 00:21:24.829 Uttam Kumaran: You know, for the last 3 years, nothing has been so great. We have every client now. I mean, the fun thing about our job, though, is we were way ahead of clients in that, like, I was looking at that a few years ago. Now, I think, finally, when we come into a client.
95 00:21:24.830 ⇒ 00:21:28.249 Uttam Kumaran: And these are typically, you know, medium to large-sized private businesses.
96 00:21:28.250 ⇒ 00:21:40.030 Uttam Kumaran: once we have their stuff organized, they’re like, hey, can we… now they’re… I think 6 months ago, I don’t think this was even the case, now they’re like, hey, wonder if we could chat over the data, or I can use it for point lookups instead of having to go into my Looker.
97 00:21:40.030 ⇒ 00:21:57.349 Uttam Kumaran: you know, just to look up a person ID in Salesforce or something like that. But those are great use cases, but kind of, like, interested to hear about that journey from your side, and, like, what you guys found from looking left to right at other people probably pursuing the same thing, and of course, we’ve used… we’ve been using the features for a few months now, but yeah, curious.
98 00:21:57.350 ⇒ 00:22:12.169 Jamie Davidson: I’d say, like, I mean, the way I’d sort of describe it, or even, like, the story, like, you know, I’ve been in data for 20 years, I know data is hard, and, like, contextual, and, like, natural language is imprecise, too, so, like, sort of…
99 00:22:12.170 ⇒ 00:22:32.340 Jamie Davidson: you know, I think a lot of folks want magic, and magic doesn’t exist, and so, like, in general, I’d say I’ve been a little bit skeptical, honestly, of the ability to sort of sprinkle AI or rub some LLM on some, you know, a data problem, and, like, you know, suddenly everything is magic. The sort of… the…
100 00:22:32.340 ⇒ 00:22:57.299 Jamie Davidson: way I sort of think about it is we started Omni, you know, like, I spent a long time at Looker, like, absolutely loved the product and the team and the experience there, but, like, wanting to basically make it an order of magnitude easier to build data model and use data model. That was sort of the core piece, and we did it very literally to help users. So, like, we wanted to serve every user inside an organization, and it turns out that, like.
101 00:22:57.300 ⇒ 00:23:16.450 Jamie Davidson: like, building data models is effectively how you define context for your business. It’s how you define, you know, your data relationship, operational metrics, how you basically marry the technical implementation to the context or the language that people will end up using for it. And so, we ended up basically building the best way to build
102 00:23:16.450 ⇒ 00:23:34.509 Jamie Davidson: context for… for data and, you know, AI to access data, too. And the sort of… the… I’d say the biggest, like, the key insight was really not… it was not to generate text-to-SQL, which is actually… we built text-to-SQL, too, probably when everyone else started doing it also, like, as soon as whatever chat GPT2
103 00:23:34.510 ⇒ 00:23:59.500 Jamie Davidson: 2.5, or whatever it was, I remember the demo, and, like, everyone got so excited, and so, like, we built that into the product as well, but realized, like, that the opportunity was actually not to do, text to SQL, the opportunity is to do text-to-semantic query, because that actually turns on… and, like, the reason why we realized this is, at the time, the core focus of the product, and we still have this in the product today, basically, we go from SQL, we parse that SQL, we pull out reusable objects.
104 00:23:59.500 ⇒ 00:24:21.159 Jamie Davidson: of it. We know, basically, what is part of the data model, and when you have the data model lighting up, it basically means it’s interactive for a user, and it means it’s interpretable to a non-technical person. So, like, a technical person can read SQL, but they can also write SQL. So, like, that’s not really realizing the potential to remove friction for, like, kind of the, you know, for the ops people, for the executives that want to use it.
105 00:24:21.160 ⇒ 00:24:29.429 Jamie Davidson: Non-technical people, though, they have, like, the text-to-SQL paradigm is just fundamentally broken. Like, they cannot interpret a number because the number, you know.
106 00:24:29.430 ⇒ 00:24:52.879 Jamie Davidson: basically because natural language is imprecise, you don’t know what that number means. Like, if I ask how many users do we have, there’s, like, 15 different definitions any organization might have that are reasonable, and reasonably different. And so, you know, you need to be able to make it so that that person can say, oh, I see, you’re showing me how many people have used the product on this cadence, or, like, with what constitutes usage, or how many people have signed up in the product, or whatever it might be.
107 00:24:52.880 ⇒ 00:25:16.759 Jamie Davidson: because we have, effectively, that mapping, that sort of deterministic system that, that, you know, is that semantic layer, too, and then it also just gives you a bunch… it gives you, like, that row-level, comm-level permissions, it gives you kind of control over, you know, you’re not hallucinating, you know, a new definition of gross margin just because, like, the LLM forgot about sales tax, or your data model’s a little bit different, or your operations are a little bit different. You know, it turns out,
108 00:25:16.760 ⇒ 00:25:35.409 Jamie Davidson: That’s really important as well. So, like, that was sort of the core pieces, like, hey, how do we help people build data model? How do we help them expose that data model in a thoughtful way to LLMs? How do we make it sort of deeply integrated into the UI so it’s interpretable, so it’s not just, you know, SQL or a number coming back to them? And then, the nice part about it, too, is it’s…
109 00:25:35.410 ⇒ 00:26:00.300 Jamie Davidson: it removes a ton of the friction. So, like, it’s really good at summarizing data, it’s really good at even suggesting questions to the end user, because it, like, knows what the data model is, and can reason from, you know, the… the sort of sum total of all of the human intelligence that the LLMs are trained on. You know, what… what do, you know, like, if you’re a, you know, a B2B, you know, dog food, you know, e-commerce company, like, what are the questions you might want to ask? And, like, they’re really good
110 00:26:00.300 ⇒ 00:26:01.919 Jamie Davidson: They’re doing those types of things.
111 00:26:02.620 ⇒ 00:26:05.160 Uttam Kumaran: Yeah, I guess my other question is, like, how are you thinking about
112 00:26:05.260 ⇒ 00:26:24.009 Uttam Kumaran: getting more context into the product. You know, I know that, you know, typically, again, you can have, like, column metadata. I see you also guys are adding more instruction-based text where users can add it, but what I talk to our clients and our team about is, like, we are the biggest bottleneck for bad context.
113 00:26:24.010 ⇒ 00:26:26.120 Uttam Kumaran: So, I’m kind of curious about…
114 00:26:26.120 ⇒ 00:26:48.160 Uttam Kumaran: you know, one of the advantages of us as consultants going in is in order to do our job, we have to collect so much context, right? We go in and we do a ton of documentation. We have meetings where we have a ton of Zoom meetings, and we use that with AI to create project plans, and for me, you know, one of the things that I talk to our team about is, like, look.
115 00:26:48.160 ⇒ 00:27:02.380 Uttam Kumaran: for every client, we’re gonna have a trove of meetings that we’ve had with a client, our understanding of the project, the LLM’s understanding of what we’re doing here, the code base, all of that I want to be made available to the AI when someone’s asking a question.
116 00:27:03.360 ⇒ 00:27:18.409 Uttam Kumaran: And for our clients, when they’re asking a, okay, how much… how many orders did we sell today? I want Omni to have all the contacts that we’ve already gathered for them, which… which is, like, that may be in documents, that may be in structured some way. Like, how do you guys think about,
117 00:27:18.410 ⇒ 00:27:31.210 Uttam Kumaran: for your users, like, helping them shove more context into there. And, like, input validation is probably, like, one way of just being, like, you need to add more information, this is not enough. But, like, I’m kind of… I think a lot about, for our clients, like.
118 00:27:31.220 ⇒ 00:27:39.789 Uttam Kumaran: we’re coming in and collecting so much context, it’s really powerful for us to spin that around and hand it to AI for more and more use cases.
119 00:27:40.740 ⇒ 00:27:41.390 Jamie Davidson: Yeah.
120 00:27:44.200 ⇒ 00:27:58.240 Jamie Davidson: Yeah, so the… I think it’s a super interesting question, and, like, very literally was talking with, kind of, our AI team about, like, very literally this topic, this morning, because I think this is where almost everyone falls over. I think we’ve actually got almost all of the right primitives, assuming.
121 00:27:58.240 ⇒ 00:27:58.740 Uttam Kumaran: Yes.
122 00:27:58.740 ⇒ 00:28:00.379 Jamie Davidson: The context today.
123 00:28:00.380 ⇒ 00:28:00.820 Uttam Kumaran: Yes.
124 00:28:00.820 ⇒ 00:28:04.160 Jamie Davidson: We don’t… we do,
125 00:28:04.490 ⇒ 00:28:29.480 Jamie Davidson: we do some, but not enough yet. And actually, it’s something that I’ve been personally, like, experimenting with a bunch, too. The… like, there are… like, actually, maybe it’s, like, a good example. We had a customer that was… I think they e-fouled something like 25 different, you know, AI data tools, too, and, like, ultimately picked Omni. And the way that they created… they basically… they had, like, a super rigorous process. They basically… they created… they interviewed a bunch of their business users.
126 00:28:29.480 ⇒ 00:28:54.410 Jamie Davidson: they recorded all the transcripts, they actually used ChatGPT to pull out, effectively, what are the questions that the user has, and they basically generated what sort of amounts to an eval set. So, like, I think it was something like 25 to 50 queries, or, like, prompts that they wanted to ask each of these things, and they basically, you know, they used those prompts, they iterated on it a little bit, built data model, added some context and stuff to an
127 00:28:54.410 ⇒ 00:29:17.990 Jamie Davidson: to sort of try and get the right answer, and ultimately, that was their mechanism for evaluating sort of their AI. I think we want to sort of standardize that. So, like, one, make it really easy for you to do things like import, context, whether it’s in, like, free text, or, like, connecting to Notion docs, or, like, connecting to, you know, like, Markdown files, or, like, I don’t know what the right format is, to be fair, that we can kind of pull in.
128 00:29:17.990 ⇒ 00:29:34.380 Jamie Davidson: We also want to use usage. You know, I think it’s a really… it’s subtle, but it’s sort of super valuable. Most of the value you get when you’re actually building out these sort of contexts is that hiding fields that aren’t used, actually. So, like, removing, basically, semantically similar topics, ideas.
129 00:29:34.380 ⇒ 00:29:35.120 Uttam Kumaran: Yes.
130 00:29:35.120 ⇒ 00:29:36.130 Jamie Davidson: dates.
131 00:29:36.130 ⇒ 00:30:00.309 Jamie Davidson: measure, and just picking the ones that actually end users mean, and so, like, you know, for, like, if you’re looking at revenue, you never need to look at, like, canceled orders or something. Like, I don’t know, but, like, sort of, like, kind of hardening, effectively, the types of questions, or sort of the pre-built sort of context that’s implicit, even though you have the canceled orders in the database, like, you never want to show them to the users, because that’s never part… like, only if they ask
132 00:30:00.310 ⇒ 00:30:03.889 Jamie Davidson: only if, like, I don’t know, finance cares about cancer, or operations cares about.
133 00:30:03.890 ⇒ 00:30:04.220 Uttam Kumaran: Sure, sure.
134 00:30:04.220 ⇒ 00:30:29.119 Jamie Davidson: like, you know, revenue generation things don’t. And so, like, helping people use, like, literally the dashboards, use the existing usage to be able to say, like, hey, let’s hide all of the unused fields, or unused views, or unused topics from the LLMs. Let’s use the, sort of, usage patterns on queries to say, hey, can we auto-generate?
135 00:30:29.120 ⇒ 00:30:34.290 Jamie Davidson: auto-generate context to be able to say, like, you know, you always query
136 00:30:34.290 ⇒ 00:30:43.469 Jamie Davidson: for this filter, you know, canceled orders, not canceled orders, right? I don’t know what, you know, whatever the example might be, and providing that as contextual. So that’s sort of…
137 00:30:43.560 ⇒ 00:31:07.979 Jamie Davidson: That’s, like, the high-level framing for it, but I actually think we want to sort of get into a systematic process where it’s like, hey, can we do… like, give us your 100 questions that you… your business asks. Let’s go through and improve it, and then let’s watch all the questions that your end users have, too, over time, because it’s going to change, like, in the way the language even changes. You know, their operations change, the questions will change, too, and sort of have a mechanism to be like, hey.
138 00:31:07.980 ⇒ 00:31:18.400 Jamie Davidson: you know, if people are asking these new questions in, like, a broad category, and they’re being thumbs down like it’s a bad result, let’s, like, go and see what we can do to improve the… and I think it’s…
139 00:31:18.400 ⇒ 00:31:31.199 Jamie Davidson: I think it’s without a doubt human in the loop, you know, it needs to be, but but, like, sort of building these sort of feedback loops, or building the sort of usage tracking, and, and then ultimately the kind of the toolset to be able to drive that context, too.
140 00:31:32.500 ⇒ 00:31:48.639 Tamara John: And kind of on that, just two things that I shared that might be helpful as you’re working with clients. I’m not sure if you’re aware of them, but one is a community article that we have on just kind of, like, best practices for improving AI answer quality, and just, like, how we kind of think about it, and then another one is we did actually an OmniLive session,
141 00:31:48.740 ⇒ 00:32:13.610 Tamara John: with Colin, one of our other co-founders, where he just, like, goes through in the product and shows how he would do it. So those would definitely be helpful, but then also, just kind of to add to what Jamie was saying, like, we have a lot of customers, I know, like, documentation is one of the first things to go, especially for a fast-moving company. But one of our customers is a very fast-growing software company. What they did is they just used an LLM to do… to, like, kick off an
142 00:32:13.610 ⇒ 00:32:14.740 Tamara John: Auto Doc.
143 00:32:16.110 ⇒ 00:32:31.650 Tamara John: project for dbt, because they realized that, like, oh my gosh, like, the more documentation and metadata that we have in dbt just actually makes it easier. It’s good for dbt, but it also makes our Omni AI smarter, and so they did that, and then they just pushed that to Omni. I have a lot of customers who I speak with who are using
144 00:32:31.650 ⇒ 00:32:38.620 Tamara John: like, Omni and DBT, and just kind of, like, auto-docs and everything to constantly refresh their,
145 00:32:38.670 ⇒ 00:32:44.740 Tamara John: to, like, add to their Omni context, and then also another thing that I hear from customers, like Jamie was just talking about, is just, like.
146 00:32:44.960 ⇒ 00:32:48.900 Tamara John: It’s a really cool thing about product analytics, like, when you’re looking at product analytics.
147 00:32:49.070 ⇒ 00:33:03.659 Tamara John: I feel like you’re, you know, usually you’re trying to stitch together and understand what the end user is doing, but with our AI questions, you don’t have to guess what someone’s trying to do. They tell you exactly what they want. Yeah. And so, like, literally, our team just, like, looks at what people ask.
148 00:33:03.770 ⇒ 00:33:05.590 Tamara John: And then optimizes for that.
149 00:33:06.580 ⇒ 00:33:18.170 Uttam Kumaran: Exactly. I mean, we’re starting to do the same thing in that, like, for some clients where we have, or I’m trying to get them to go to Omni, I’m like, I have… we have probably, like, 20 hours of meetings with them.
150 00:33:18.210 ⇒ 00:33:36.460 Uttam Kumaran: with every answer on how to do all the data, because we’re running their data teams, right? That’s all the context it needs, and then it’s like, it would be great to work similarly how you would do a migration. You almost do, like, a context migration, where, like, column… column level descriptions can get filled out, table level metadata can get filled out.
151 00:33:36.460 ⇒ 00:33:44.510 Uttam Kumaran: why or when you should use a dashboard in plain text is made available. Like, that’s all super, super important, but again.
152 00:33:44.580 ⇒ 00:33:57.789 Uttam Kumaran: where… what I find now is that because how… because of the benefit of AI, now my team is really incentivized to do documentation. Like, if we’re… if we’re all good engineers, yeah, we do documentation, but now because
153 00:33:57.790 ⇒ 00:34:13.409 Uttam Kumaran: it’s really important. We have cursor rules and everything, we use cursor to write documentation on our repos, like, on pre-commit, and because I know that the next time someone uses AI in the repo, their output will be way better. And I think so much…
154 00:34:13.409 ⇒ 00:34:31.870 Uttam Kumaran: context is lost in meetings and in Slack and things like that that just have to enter, you know, enter the thing. And again, even the best data teams I know, like, are not keeping up with palm-level definitions. Like, nobody’s doing that. It’s, like, such a luxury good prob- it’s, like, such a luxury problem. Yeah. If you’re, like, really focused on
155 00:34:31.960 ⇒ 00:34:37.960 Uttam Kumaran: keeping your column level definitions, like, you know, now, but now it really matters, you know, so…
156 00:34:38.000 ⇒ 00:34:50.529 Jamie Davidson: Yeah, yeah, no, I think… I think that’s accurate. It makes… it makes… but it… I think that that is… it’s, you know, and this is… we saw this at Looker also, too, but, like, it’s, it’s… it’s hard to get people to do data modeling.
157 00:34:50.530 ⇒ 00:35:15.510 Jamie Davidson: Full stop. It’s hard to get you to, go through the work to say, I’m going to enumerate and, like, create some, you know, reusability with better abstractions, because it’s gonna help us, you know, down the road to have, sort of, consistency. It’s, like, kind of one of these things where it’s solving a tomorrow problem, not a today problem, and so, like, most data teams don’t, don’t necessarily get into it. Like, they’re just, like, looking to
158 00:35:15.510 ⇒ 00:35:40.459 Jamie Davidson: how do I rip it out as fast as I possibly can? And only after you run into, like, hey, we’ve got complete chaos, and I gotta change, you know, gross margin in 15 different places, and like, you know, every quarter we’re spending two hours trying to reconcile numbers. That’s the only time when you sort of really go through the trouble of it. I think, similarly, it’s hard to get people to actually enumerate the, like, the language that they’re using, sort of the context that they’re using for it.
159 00:35:40.460 ⇒ 00:35:49.260 Jamie Davidson: or even, like, you know, we’ve… I was, I was helping, helping with a POC, like, a trial that we, we were, were working with, maybe it was last week or the week before.
160 00:35:49.260 ⇒ 00:35:51.770 Jamie Davidson: relatively recently,
161 00:35:51.880 ⇒ 00:35:58.060 Jamie Davidson: where they wanted to eval AI, but they didn’t even, like, they didn’t know what questions they wanted to ask, actually.
162 00:35:58.060 ⇒ 00:35:58.470 Uttam Kumaran: Yeah, yeah.
163 00:35:58.470 ⇒ 00:35:58.949 Jamie Davidson: I was like…
164 00:35:58.950 ⇒ 00:36:17.009 Uttam Kumaran: That’s the… that’s the number one thing. We come into clients, and we build a golden data set, like, day one, basically. And again, like, nobody… that is such a… that is, like, a painful process of even knowing, like, what do you… what are you trying to solve for, and then what is the right answer? The second part is worse.
165 00:36:17.010 ⇒ 00:36:21.439 Uttam Kumaran: You know, because no one has answered this properly, or no one really, like.
166 00:36:21.440 ⇒ 00:36:22.010 Jamie Davidson: Yes.
167 00:36:22.010 ⇒ 00:36:38.650 Uttam Kumaran: They’re just like, no, I want to ask this, but they’re like, I have no way to judge this. And I’m like, okay, so that’s where we start, right? And so we build golden data sets when we walk into every client on the AI side because of this exact problem. I can’t optimize it. There’s nothing to optimize until that happens.
168 00:36:38.650 ⇒ 00:36:51.440 Jamie Davidson: I mean, I think that’s exactly right. I actually… I don’t… I don’t know what the right answer is, too. Like, I don’t know if it could be productized. I’m, like, I’m super curious. Sure. I’m curious, in your view, too, even, like, because it… it does feel like…
169 00:36:51.480 ⇒ 00:37:07.660 Jamie Davidson: every… like, basically, that should be the process for literally everybody. That should be the data team’s process in general, too. I think it… it sometimes feels like pulling teeth and, like, they really don’t want to do it, which is… which is interesting to me. And I don’t know if there’s ways to make it easier or to automate some of it or something. Like, it’s really funny.
170 00:37:07.660 ⇒ 00:37:25.090 Uttam Kumaran: Yeah, let me tell you how we’re thinking about it. I mean, one is, you know, I started this business 2 years ago, right when, like, right after 3.5 came out, and so we built this whole business with AI, but when we talk about customers, I realized that, again, like, how much context we…
171 00:37:25.090 ⇒ 00:37:37.209 Uttam Kumaran: absorb as consultants, because we have to. We come in and we ask every question, questions that haven’t been asked, questions that haven’t been asked in a while, questions that never thought of… like, we create all this rich context.
172 00:37:37.210 ⇒ 00:37:52.589 Uttam Kumaran: For me, ultimately, what we started to do is I want to build an agent for every client, which has context of their codebase, all of our meetings with them, all of our project plans, and that enables our consultants internally just to speed up. Okay, are we generating a… are we answering a question for them?
173 00:37:52.590 ⇒ 00:38:16.450 Uttam Kumaran: Do we want… are we putting together an analysis? Okay, we have an agent per client, right? That’s something that we’re debating on whether to just give access to the client, or to just keep internally, but either way, like, that is our job, is to collect all of that, and then we… we basically shove all of that, into Superbase and put it behind a chatbot, basically, for each of our clients. But the data piece is where, like, I don’t want to build that.
174 00:38:16.620 ⇒ 00:38:26.519 Uttam Kumaran: Like, I have… we’ve… I know, I’ve seen all the text-to-LM, text-to-SQL stuff, like, I’m not interested as a consultant to build it, which is why I’m really excited about
175 00:38:26.520 ⇒ 00:38:43.249 Uttam Kumaran: Omni, because all of that work we do, now we just have the MCP we can go use wherever we want to use it, in Slack, or whatever, right? And so that’s… that’s, like, our playbook now, is to leverage things that way. But again, like, I’m looking for things we can do… we can do repeatedly across clients.
176 00:38:43.250 ⇒ 00:38:59.140 Uttam Kumaran: And I think it’s such a shame if we walk into a client as consultants, we get all the info, and we don’t put it behind an AI agent for them. Like, what do… it’s just, like, in a world now, like, we should. That’s, like, that seems very natural, right? And so, I agree, like, I don’t know what the product…
177 00:38:59.190 ⇒ 00:39:09.410 Uttam Kumaran: necessarily is. Maybe it is, like, hey, jump all of your transcripts or whatever you got, and you guys will smartly associate or propose associations. Yeah.
178 00:39:09.610 ⇒ 00:39:21.970 Uttam Kumaran: That could be a great way of even just, like… because again, I’m sure that… I know you guys, right when you install the product, you can ask questions of your data, but for example, if no one put anything in there if it’s the order stable, you’re screwed, right? But you’re not…
179 00:39:21.970 ⇒ 00:39:24.250 Jamie Davidson: That’s a good problem.
180 00:39:24.250 ⇒ 00:39:42.260 Uttam Kumaran: Omni’s screwed, because you’re… Omni’s gonna be like, they’re like, oh, it doesn’t work. So it’s… for me, I almost am like, how do you, preempt that by, like, forcing this context gathering so that at least, you know, like, it’s similar to what’s happening with ChatGPT. If you just put the role in the prompt, it’s 40% better. It’s, like, so stupid.
181 00:39:42.270 ⇒ 00:39:48.210 Uttam Kumaran: But, like, that’s how… that’s, like, what I feel like a tool like yours should preempt and say, like, in order.
182 00:39:48.210 ⇒ 00:39:48.660 Jamie Davidson: We’re free to…
183 00:39:48.660 ⇒ 00:39:49.079 Uttam Kumaran: That’s a word.
184 00:39:49.080 ⇒ 00:39:50.959 Jamie Davidson: We can kind of make that easier, yeah, that’s absolutely right.
185 00:39:50.960 ⇒ 00:39:55.809 Uttam Kumaran: Dump everything in, we’ll do some basic association, so the first time they try.
186 00:39:56.010 ⇒ 00:40:03.469 Uttam Kumaran: you win. I think, and to your point about, like, existing data teams, I think it’s just things are changing, so I don’t know, it’s tough, like…
187 00:40:03.500 ⇒ 00:40:18.880 Uttam Kumaran: I feel like we… our business is really, like, we try to move fast, and we can also… we’re also the first to get fired, so it’s really important that we leverage AI to solve customer problems fast. So we are super, super forward on it, but…
188 00:40:19.320 ⇒ 00:40:22.140 Uttam Kumaran: part of that, I don’t know. Like, I just think some teams are just…
189 00:40:22.270 ⇒ 00:40:29.539 Uttam Kumaran: figuring it out, or they… maybe they tried a different product, and now they don’t like LLM and data, and you know, you never know what it is.
190 00:40:29.760 ⇒ 00:40:32.220 Jamie Davidson: Yeah, yeah, no, I think that’s accurate, that’s totally fair.
191 00:40:32.460 ⇒ 00:40:33.170 Uttam Kumaran: Yeah.
192 00:40:33.460 ⇒ 00:40:42.539 Uttam Kumaran: I guess we had one more, one question, kind of, like, maybe more on the commercial side about, like, how you’re thinking about migrating customers from.
193 00:40:42.540 ⇒ 00:40:42.970 Jamie Davidson: Yeah.
194 00:40:42.970 ⇒ 00:40:56.930 Uttam Kumaran: like a Looker, Tableau, Power BI. Like, are you guys thinking about… and this is kind of selfish, but I think a lot of our ICP are all on these legacy tools. Again, typically either Tableau, Looker, Power BI,
195 00:40:57.340 ⇒ 00:41:04.289 Uttam Kumaran: I think that’s probably most of, like, what we’ve seen. And for us, like, it’s very tough for me to pitch a migration project.
196 00:41:04.290 ⇒ 00:41:05.190 Jamie Davidson: Yeah.
197 00:41:05.190 ⇒ 00:41:19.069 Uttam Kumaran: I really don’t even… I don’t even want to take that project, even if we get paid sometimes, because it can be quite demanding, and it’s… so I’m trying to think about, like, how we… how we make that, clear to not only why moving on me
198 00:41:19.340 ⇒ 00:41:34.609 Uttam Kumaran: is helpful, but I don’t know, I’m just curious, like, about the customers that you find that typically are… I mean, like, Looker and Tableau are described as, like, legacy environments, but what… what do you think about when you talk to folks that are considering migrating? Like, how do you ease their stress around.
199 00:41:34.850 ⇒ 00:41:35.280 Jamie Davidson: Rather.
200 00:41:35.280 ⇒ 00:41:35.750 Uttam Kumaran: process.
201 00:41:35.750 ⇒ 00:42:00.660 Jamie Davidson: So, like, there’s, like, a couple of different things. Like, sometimes folks have effectively just gotten frustrated with, like, for Looker and Tableau, like, a complete lack of product development, you know, the last 5 years post-acquisition, and or, like, price increases, which I think has, you know, been very common for both. With Power BI, this year, very literally, since, like, April, when they changed
202 00:42:00.660 ⇒ 00:42:24.670 Jamie Davidson: the, sort of, the pricing paradigm. We’ve seen much more, just, like, and I’d say in these cases, it’s not us motivating migration, it’s them, like, literally coming and saying, we are migrating, and, like, it’s like, our price for Power BI is, like, more than we pay for Snowflake, and, like, it’s insane. And then there, we just sort of, we try and say, hey, we will help with it, you know, there’s some tooling that you can do.
203 00:42:24.670 ⇒ 00:42:32.980 Jamie Davidson: We can help. So, you know, we, like, we honestly, we want to partner, you know, super deeply on these fronts, too, and so, the manual effort, but, like, really.
204 00:42:33.110 ⇒ 00:42:45.389 Jamie Davidson: you kind of want to… it’s like, you know, it’s a good opportunity to do spring cleaning and, like, you know, clean up the mess and sort of rethink things, you know, often. I’d say for folks that,
205 00:42:46.010 ⇒ 00:43:10.889 Jamie Davidson: that haven’t decided, and they’re kind of looking at us, sometimes it’s, you know, we can consolidate, so, like, take your Looker and your Tableau and your other… or, like, take your Tableau and, like, we’ll clean it up and give you governance, and, like, we’ll replace it, and it’s sort of like a classic, like, rip-replace. Those are hard. I fully agree that those are hard. I’d say, more often, it’s we typically come in with a net new use case, and it’s, like, maybe it’s, maybe it’s embedded use case and doing customer-facing stuff.
206 00:43:10.890 ⇒ 00:43:35.309 Jamie Davidson: Maybe it’s AI, be it’s, like, Customer 360, or they’re using a new data warehouse, or they’re, you know, whatever, whatever, you know, whatever it might be, this sort of new initiative, and we sort of say, like, hey, yes, we can replace this portion of it, and then over time, you know, you’re gonna just build all of your new stuff in Omni, and you’ll retire a bunch of your Tableau seats, and you’re gonna retire a bunch of your Looker seats, or like.
207 00:43:35.310 ⇒ 00:43:47.180 Jamie Davidson: you may be able to completely deprecate it, too, and it’s a migration of, you know, a year, and so, like, you’re, like, you’re using both tools alongside, you maybe have a few seats in Omni, and, like, a lot of seats in Looker still, but, like.
208 00:43:47.180 ⇒ 00:44:10.800 Jamie Davidson: by the end of the year, you’re expecting to, you know, we’ve got a couple of big, big customers that… sometimes it’s… it’s, you know, much faster, like, hey, they will… like, our Looker renewal is up in a quarter, we are going to turn it off, and we need to get everything over. Like, I can’t remember, but the… I think BuzzFeed was, like, a great example. It was, like, a customer that we’d had on stage with us at, like, a bunch of Looker conferences.
209 00:44:10.800 ⇒ 00:44:22.579 Jamie Davidson: Like, absolutely love them, early looker customer, and I think we migrated them, like, so they had a huge estate, but they migrated in something like six weeks or something like that. So, like, and the reality is it’s not, it’s not just
210 00:44:22.580 ⇒ 00:44:41.960 Jamie Davidson: kind of copy-paste from Looker. It is… it does require a rethinking. It’s like, hey, you know, how do… what are the operational questions we want to ask? What are the… what’s the cadence we want to have? Like, what are the dashboards that we really need? And then, you know, not bringing over Jamie’s copy of a copy of a copy… probably not worth… worth it, you know?
211 00:44:43.750 ⇒ 00:44:44.210 Uttam Kumaran: Makes sense.
212 00:44:44.210 ⇒ 00:45:02.910 Tamara John: Yeah, and I was gonna say, I can’t name this particular customer, but, like, I have a customer right now who I’m finalizing a presentation with them to submit for today, and again, it’s like, they had been on Looker for over 5 years, had been using it org-wide, and they did a 10-week migration with us, and that even includes, like, extra buffer for having both tools on.
213 00:45:03.330 ⇒ 00:45:10.399 Uttam Kumaran: Great. Yeah, I mean, in our size, again, I think we’re at… we’re… what companies we go into, there’s typically no…
214 00:45:10.400 ⇒ 00:45:27.589 Uttam Kumaran: like, owner for data. There are either random people scattered across, or there’s, like, they’re just, like, just starting. And so, they don’t even have an understanding of, like, the playing field of tools. And again, what do they think about? They know Power BI, Tableau, Looker, just from
215 00:45:27.590 ⇒ 00:45:31.039 Uttam Kumaran: their brand value, right? And so for us, I think…
216 00:45:31.040 ⇒ 00:45:53.689 Uttam Kumaran: it’s… it’s tough, because for clients that are existing in Looker, they really want to hear, like, okay, what else am I going to get? Especially some of the folks that are making the procurement decisions are not always the folks using the data tools, which makes it even harder. But the AI thing for me is, like, a real wedge. If I can say, look, we’re gonna… their net new use case is actually, like, being able to chat with your data through an MCP in Slack.
217 00:45:53.690 ⇒ 00:46:02.940 Uttam Kumaran: that is a really great use case. Are you guys thinking about pricing that differently? Or, like, is it… you think it… is it… are you considering bundling
218 00:46:03.300 ⇒ 00:46:06.079 Uttam Kumaran: the AI piece with the old platform, or…
219 00:46:06.080 ⇒ 00:46:31.049 Jamie Davidson: Yeah, I mean, the way we do it today is basically, you know, I think AI is table stakes, and, like, is going to be expected, and it’s going to be part of it, and so, like, a certain amount of use, without a doubt, is going to be included in sort of the base product, no matter what. We’re in the process of rolling out, basically dashboards, summarization, and then, and then sort of a deep research, like Agentic
220 00:46:31.050 ⇒ 00:46:46.759 Jamie Davidson: where it’s, like, doing things… it’s really crazy. I was actually just a little using this earlier. Like, look at our pipeline this year, you know, summarize, like, tell me where, like, it’s growing most, and, like, summarize trends, and it’s crazy, and it does, like, 10 quarters.
221 00:46:46.760 ⇒ 00:46:48.800 Uttam Kumaran: Is it, like, doing reasoning steps, where it’s, like, writing.
222 00:46:48.800 ⇒ 00:46:49.310 Jamie Davidson: Yes.
223 00:46:49.310 ⇒ 00:46:49.830 Uttam Kumaran: of worries.
224 00:46:49.830 ⇒ 00:46:55.299 Jamie Davidson: Basically, like, it’ll create a plan, it’s like, okay, I’m gonna look at this, I’m gonna look at this trend, here I’m gonna.
225 00:46:55.300 ⇒ 00:46:58.219 Uttam Kumaran: I’ve started to do some of that in Cursor, actually, because I have cursor…
226 00:46:58.220 ⇒ 00:46:58.730 Jamie Davidson: Yes.
227 00:46:58.730 ⇒ 00:47:05.900 Uttam Kumaran: can now run queries, and I have it hooked up to dbt, so it will start to do that, but it’s a couple of ops, right, it has to do.
228 00:47:05.900 ⇒ 00:47:29.200 Jamie Davidson: Yes, and, like, basically, very literally, that’s where it came from, was that we basically built an MCP server tied to the querrier, and people were doing deep research stuff inside Cursor or inside, you know, Claude, and we were just like, we should put this, without a doubt, in the product. Like, we can do this in sort of a tighter way. So, like, we will roll those out. The reality is, those do… there is cost to
229 00:47:29.240 ⇒ 00:47:51.579 Jamie Davidson: to, like, token consumption in LLM, so, like, effectively, what we’ve sort of moved to is, you know, basically, we want to bundle a whole heck of a lot of usage, and, like, I think we were, like, pretty generous in what we put in the platform. Like, I think we’ve got something like, literally, like, three customers, I think, that have been, you know, above the limit so far, so, like, sort of walking with… walking them through what it’ll sort of look like. But then.
230 00:47:51.580 ⇒ 00:48:14.860 Jamie Davidson: you can basically prepay for, basically, like, kind of token consumption, effectively, on a… at a discounted rate. So, like, if you prepay it, then customers will get sort of, like, a pretty substantial discount, and then if not, like, you just sort of have usage… like, once you go beyond, and we’ve got, like, sort of a bunch of granular tracking that we make pretty clear to the, you know, customer inside our analytics, reporting, but, like.
231 00:48:14.860 ⇒ 00:48:24.989 Jamie Davidson: I want to make it super easy, so that, like, really, like, if you find a ton of value in the deep research, or in, like, the, you know, dashboard summarization, you’re using it, like, we’ve got, like, one customer that’s basically built out, like.
232 00:48:24.990 ⇒ 00:48:47.829 Jamie Davidson: kind of… it’s kind of cool and kind of crazy, but, like, using AI to, to trigger, schedules, so, like, it looks at the dashboard and basically will summarize, say, like, hey, is this something… is anything… like, should I let a human know about any of this? And, like, I think it’s super-duper cool, but also, like, they’re doing it, you know, in a basically a programmatic way, so, like, we want to make sure…
233 00:48:47.830 ⇒ 00:48:48.450 Uttam Kumaran: Expensive.
234 00:48:48.450 ⇒ 00:48:54.459 Jamie Davidson: You know, the… it’s, like, if they’re creating value from it, it’s aligned, and we’re all… the incentives are all shared.
235 00:48:54.460 ⇒ 00:49:09.489 Uttam Kumaran: Are you guys thinking about, like, we have some clients where actually I don’t think they would ever even use the dashboard, but, like, I think they would just use chat and Slack. So, like, how are you thinking about that? Because I think that’s a whole class of users that may have never been served by the.
236 00:49:09.490 ⇒ 00:49:10.990 Jamie Davidson: Yes, very… yes.
237 00:49:10.990 ⇒ 00:49:13.959 Uttam Kumaran: It’s because the dashboard was actually the bad medium.
238 00:49:13.960 ⇒ 00:49:14.300 Jamie Davidson: X.
239 00:49:14.300 ⇒ 00:49:26.059 Uttam Kumaran: this stuff. So how are you thinking about that, or have you guys thought of it? Because there are some clients that I know where I’m like, look, we put dashboards in front of them, they never use it. They just keep asking us the questions. And I’m like.
240 00:49:26.510 ⇒ 00:49:37.020 Uttam Kumaran: I was like, we’re gonna build something to just do that, where maybe it’s an… there’s a person, but we just copy-paste their question and use AI to help us speed it up, but I… I would say we have some customers
241 00:49:37.190 ⇒ 00:49:47.089 Uttam Kumaran: we have some customers as a whole, and there are users within existing customers that would only use the MCP, like, outputs through Slack or, you know, otherwise.
242 00:49:47.090 ⇒ 00:50:12.049 Jamie Davidson: Yeah, without a doubt. We’ve sort of heard the same message, too. Effectively, we’ve got, actually, one of our support engineers built a chatbot that we’re actually sort of in the process of productionizing, and so, like, we will, without a doubt be releasing that, too. But, like, effectively, it’s just, like, we’ve got sort of broad API coverage, we’ve got the MCP server, too. We’ve actually had a few customers connect sort of a custom chat interface that they
243 00:50:12.050 ⇒ 00:50:36.879 Jamie Davidson: they’ve built through the MCP inside Slack, as well, but, like, we’ll make it an easy native thing, where it’s just going to be, like, press a button and go. We do have a Slack integration, today with sort of our, like, you could send content into Slack and stuff, we want to be able to pull it, because, like, I think you’re exactly right. Like, actually, the thing that gets me most excited, like, coming from a skeptical on AI, which I definitely, sort of was, what gets me most excited is I think it actually does
244 00:50:36.880 ⇒ 00:51:01.809 Jamie Davidson: does dramatically remove friction. And so, like, and it’s like, it’s the support people, it’s the salespeople that are doing… it’s like, tell me about this account, tell me about this customer, tell me about their use pattern, like, help me, help me under, you know, understand, like, this marketer is asking about campaign performance, or… or, honestly, too, the other one is, executives, actually. Like, so, instead of the, you know, CRO going to the data team and being like, hey.
245 00:51:01.810 ⇒ 00:51:24.410 Jamie Davidson: what’s our pipeline look like for Europe this year? Just can ask it, you know, he or she can ask it direct in the product, and that’s, like, that’s transformative for folks, too. Basically, these are people that aren’t coming into BI tools, these are people that aren’t using data otherwise, or, like, have to wait for, you know, a 3-5 day turnaround time, and instead it’s turning it into, like, 30 seconds. It’s crazy.
246 00:51:24.550 ⇒ 00:51:28.330 Uttam Kumaran: Yeah, yeah, that’s the exact kind of use case we’re seeing as well, so…
247 00:51:28.910 ⇒ 00:51:30.130 Jamie Davidson: Yeah. Cool.
248 00:51:30.280 ⇒ 00:51:31.050 Jamie Davidson: That’s awesome.
249 00:51:32.800 ⇒ 00:51:37.389 Uttam Kumaran: Yeah, Jake, I asked a bunch of questions. That’s, like, that’s, like, a lot of what I wanted to hear about, so…
250 00:51:37.390 ⇒ 00:51:38.400 Jamie Davidson: Yeah. Oh.
251 00:51:38.400 ⇒ 00:51:42.850 Uttam Kumaran: I know we didn’t ask about some of the nitty-gritty, but we’re familiar with the product, and like…
252 00:51:43.230 ⇒ 00:51:53.629 Uttam Kumaran: this is sort of, like, when I talk to customers, they’re not… they’re not, asking me, like, okay, tell me about the topics feature. They’re telling me, like, is this going to enable AI?
253 00:51:53.630 ⇒ 00:51:54.470 Jamie Davidson: Yeah.
254 00:51:54.470 ⇒ 00:51:55.910 Uttam Kumaran: Data in the future.
255 00:51:55.910 ⇒ 00:51:58.500 Jamie Davidson: Is this, like, a win over Tableau? Like…
256 00:51:58.570 ⇒ 00:52:02.700 Uttam Kumaran: That’s the level of, like, questions that we’re getting, so…
257 00:52:03.190 ⇒ 00:52:04.919 Jamie Davidson: I think that makes a ton of sense.
258 00:52:05.960 ⇒ 00:52:14.629 Tamara John: Awesome, and if you guys, like, have additional questions and stuff, or need resources, like, I have more things, happy to… happy to help, like, review and direct things and share things.
259 00:52:15.280 ⇒ 00:52:16.120 Uttam Kumaran: Perfect.
260 00:52:16.120 ⇒ 00:52:26.739 Jake Nathan: Yeah, that sounds good. Yeah, I’m excited to, you know, go through this, and yeah, as I, start creating my draft, I’ll definitely be in the loop with you and go from there.
261 00:52:27.230 ⇒ 00:52:27.830 Jamie Davidson: Towns off.
262 00:52:27.830 ⇒ 00:52:32.020 Uttam Kumaran: Yeah, and I know, my close friend of mine, Greg, actually started at Omni.
263 00:52:32.640 ⇒ 00:52:33.599 Jamie Davidson: Oh, you’re yay, dirty.
264 00:52:33.600 ⇒ 00:52:34.980 Uttam Kumaran: Yeah, so I.
265 00:52:34.980 ⇒ 00:52:37.170 Jamie Davidson: Yeah, that’s awesome, we’re super excited to have him.
266 00:52:37.170 ⇒ 00:52:51.969 Uttam Kumaran: Yeah, he’s awesome, by the way, like, great person, also great… has a long career in data, but yeah, he messaged me, I know he was, like, thinking about leaving Fivetran, and then he messaged me, he’s like, hey, I’m going to Omni, and I’m like, dude, why didn’t you tell me? He’s like, what do you think? I’m like, fire product, great product.
267 00:52:51.970 ⇒ 00:52:53.039 Jamie Davidson: Love it. Crush it.
268 00:52:53.040 ⇒ 00:52:57.789 Uttam Kumaran: So, yeah, I was super, super happy, and I was like, get into our Slack channel, so…
269 00:52:58.240 ⇒ 00:53:01.089 Uttam Kumaran: I’ve been slacking him through Slack Connect, so, yeah.
270 00:53:01.090 ⇒ 00:53:01.510 Jamie Davidson: Yeah, that’s.
271 00:53:01.510 ⇒ 00:53:02.000 Uttam Kumaran: Really nice.
272 00:53:02.000 ⇒ 00:53:19.330 Jamie Davidson: I love it. I love it. Yeah, no, I’ll say that he’s great. I’m super stoked. Yeah, we’ve got, we’ve got, like, a kind of a small Fivetran contingent now, actually, too, which is fun. Like, I… I love it. We sort of are slowly but surely grabbing kind of everyone from the, the ecosystem, but it’s, it’s good, and we, like, I love the Fivetran team, too, so it’s… it’s good.
273 00:53:19.330 ⇒ 00:53:19.750 Uttam Kumaran: Yeah.
274 00:53:19.750 ⇒ 00:53:22.540 Jamie Davidson: We didn’t poach him, he was, he was looking, you know…
275 00:53:22.540 ⇒ 00:53:26.690 Uttam Kumaran: I don’t know, I know. I mean, I’m not gonna play for it, I don’t play for any of…
276 00:53:27.340 ⇒ 00:53:36.059 Uttam Kumaran: So, but no, he’s just a friend, he’s just a friend of mine, so I was really happy, because I was like, dude, the product is really great. I just wish you guys, more people know about you, so we’re.
277 00:53:36.060 ⇒ 00:53:36.570 Jamie Davidson: pitching.
278 00:53:36.570 ⇒ 00:53:38.029 Uttam Kumaran: Left, right, and center.
279 00:53:38.030 ⇒ 00:53:43.769 Jamie Davidson: Yeah, we hear me both. We’re trying, but we appreciate all the help you can give.
280 00:53:43.770 ⇒ 00:53:48.569 Uttam Kumaran: And we’re excited. I think we’re trying to get more of our team access to Omni and trained up.
281 00:53:48.570 ⇒ 00:53:50.360 Jamie Davidson: You know, of course, for some of the.
282 00:53:50.360 ⇒ 00:54:01.969 Uttam Kumaran: some of the non-data users, and some of our AI team also is wanting to get into data, so they’re learning about, like, traditional data modeling within there and things like that, but, yeah, it’s really exciting. So yeah, appreciate the time today.
283 00:54:02.150 ⇒ 00:54:04.700 Jamie Davidson: Yeah, no, totally, likewise. Yeah, it’s good to meet you guys.
284 00:54:05.010 ⇒ 00:54:05.440 Uttam Kumaran: Cool.
285 00:54:05.440 ⇒ 00:54:07.370 Tamara John: Awesome. Thanks, everyone.
286 00:54:07.830 ⇒ 00:54:08.220 Uttam Kumaran: Bye.