0:00All right, thanks everyone for joining us. Super pumped to have you here as we get to talk about how top e-commerce data teams drive revenue with LLMs. We are in a really cool new place in the world where we get to do stuff like this. I'm Jacob, I'm at MotherDuck, and I'm here with the team at silvercreek. ai. We've got Gordon, Daniel, and Phi here with us too.
0:25And I'm super pumped to take you through what we're talking about today. As a quick note, before I get into the agenda, you can either hit the QR code or the comment link that our producer will drop in here to fill out a survey and introduce yourself. Thank you for dropping the comment in there, Jerel. Anyways, what we've got on tap for you today is like, why is this happening now?
0:48What has changed, right? And then why it's hard, right? What's the reality of actually landing AI in your org in an effective way? And then of course, we're gonna jump into an awesome demo of like what it actually looks like to use your data. And then we'll tie it all together and end it with a live Q&A. And then of course, tie into an activation strategy session
1:13with our partners here at Silver Creek. So really, really pumped to get you rolling here. With that said, I think I'll hand it over to Daniel. All right, awesome. Well, great to be here, everyone. I'm Daniel, CEO of Silver Creek. And today, we'll talk through basically that we seemingly have all the answers,
1:37but we're not exactly sure what to trust. We have all the tools, we have the technology landscape is very confusing. If you talk to salespeople, they will tell you that all the tools that they have is critical for your success. And so what ends up happening is we don't know what to trust. We don't know what tools to trust
2:00or which data can be trustworthy. And so we end up in this sort of analysis paralysis that really impacts our ability to make good decisions. And we oftentimes rely on our gut. And that's driving us to slower decisions that are costing us potentially sales in e-commerce and otherwise. Let's continue on.
2:25So the old paradigm was having the data team be the bottleneck, that they are the ones that are supposed to answer the questions. Maybe you're still in this situation, but we're moving to a self-sufficient, autonomous business structure where executives can make decisions with LLMs,
2:47with empowered tools right at their fingertips. And so this is what we're gonna go through today, but we want to make sure that the data team is there to help the business and not a barrier to success or blamed for answers that are old and not useful anymore. We wanna get these on the fly and we wanna simplify our decision-making structure
3:11so that we can make answers much faster and that everyone is empowered to do so. And so that's what we'll get in today. So let's continue. We know, and we've seen this stat probably a lot, that 85% of AI projects are failing. Gartner talks about this and we have to ask why. We have the data, we have the tech,
3:36but for some reason it's stalling and everybody wants to buy into this AI hype and we're gonna cut through that today. But this really comes down to a leadership and a culture issue. We're working AI across departments. This is something that is a company initiative. If you talk to a lot of CEOs, leadership, they want to get AI in,
4:01but they're not sure exactly how to measure its success or where to start. And that's really our goal here is to identify the use cases, be specific about those, and to set up an easy lean structure to get these answers and to start implementing AI in a way that is trustworthy, reliable, and step-by-step. I think a lot of people try to do too much at once
4:25and our goal is to limit those use cases and to find success early and then spread that through the company with enablement. And we'll talk a lot about that today. So with that, I'm more excited to introduce you to Gordon, who is a stud who has been on both the customer side and your position, as well as leading major companies
4:47through their whole data journey and now AI journey. So Gordon, thank you very much. All right. Thanks, Daniel. Happy to be here. All right, I want to talk to you for a second from my perspective, right? Again, Daniel mentioned that technology works. We can get data into the database. We can throw tools on top of them. We can ask chat questions. So why aren't we done, right?
5:12And the reason is because all of that by itself is not an actual solution. You know, I think everyone here is probably experimenting with AI already. Their co-workers are experimenting with AI. And just talking to a really big firm yesterday, you know, I wouldn't recognize their name if I were to say them, and they've successfully hit the slop point, right? Everyone's doing analysis. Everyone's sending documents to each other
5:36and they're all coming up with different answers. And they're also generating a whole ton of spend. So, and the problem is that they lead heavily into enablement, right? Really enable your users. But what they're missing is that enablement also frequently leads to lack of quality, right? And that equals danger. So when we talk about solutions, we want to talk about how, you know, getting the right data in, putting the right constraints in place,
6:00making sure quality is there, having the right semantics, and having the right foundation is part of the solution. And end users can't build that by themselves. It's just not their jobs, not their experience, right? And so we're going to show you what you can do with LLMs in terms of doing really insightful analysis. But I think the point I want people to take away is that you also have to have the right bones
6:22and foundation underneath that, all right? Next slide, please. So let's talk about that a little bit more, right? Why don't we have solutions? Or why can't users, you know, just build it themselves? Well, again, it's a coordination and distribution problem, right?
6:45Pulling up a chat session and just doing some, asking some questions does not actually solve problems, right? We need to be looking at business problems that have measurable outcomes, you know? Is this worth spending time on, right? And too often people don't ask that question. It's like, is my time well spent here? You know, is my company getting an ROI on my efforts, right? Are you auditing your workflow, right?
7:09Every factory, every kitchen, every food production system has QA measures along the way. So not only are you producing product, you're also making sure that product doesn't harm your end users, right? End users don't know how to do that on their own, especially not distributed end users, right? Every single person's got to build their own audit flow, right? This is where centralized solutions come into place and this is where the craftsmanship solutions come into place.
7:33No single owner accountability for outcomes, right? Who's the project sponsor? Who's the person who says, hey, let's put more money in this area so that we can generate more of that that you produced. And also conversely, hey, wait, wait, wait, who's going to tap the brakes and say, you know what? We're producing, we're creating a lot of spend here, but we're not really sure what we're getting, right?
7:57So, you know, having that, like every good project, having accountability and having sort of a single hand on the wheel is beneficial, right? Going back to data quality again, you know, again, just diving into the data, it can teach you something about the data. It probably won't give you great answers yet until you know that that is safe to use. You know, thinking like code, thinking like QA managers,
8:21anything you release, and if you're releasing a knowledge product and releasing it to other people or even to yourself, it should do more good than harm, right? And how do you ensure that happens? By leading too hard into the tech or the machinery as opposed to delighting the user, right? I've seen a lot of systems where data teams are building for themselves,
8:45they're optimizing for more reports. When I was head of data over at HubSpot, I inherited a system where I had 30,000 reports in it. I think most people from a common sense perspective can imagine that wasn't a really good system, right? We need to be building for the end user, that decision maker, right? We need to be delighting them. They should be getting knowledge products into their hands and going, oh, I know what to do now, right? And I'm really excited that you gave this to me.
9:09You know, if you're not getting that experience, then are you really getting the most potential out of your data? Next slide, please. All right, so I think Daniel, this might be back to you at this point. Yeah, B, you take it away. Thank you so much. Of course. So hi, everyone. Thank you so much for being here. So MotherDuck, they give your team the fastest path
9:34from a business question to a live answer. But now the question is, are you actually getting the value out of it? You have the data, now you need to trust it and make sure that the questions are being answered. We partner with MotherDuck on many different use cases, on strategy, use case identification, implementation. We build it with you.
9:56We're not just building it for you and handing it off. We want to make sure that you are trained and that your team is enabled to carry on the work. And we're also embedded, the training and enablement. And together we have the right technology from MotherDuck, and then you get the strategic partnership. You get the data readiness. You get figuring out what you're trying to answer and whether your data is set up to answer it.
10:21And yeah, we're just here for you. Yeah, thank you, Fi. Yeah, and just to add onto that as well, I mean, really MotherDuck is an awesome data tool, AI tool. And we know from the technology landscape that there are so many different tools out there that again, everyone is telling you
10:43that this is a critical part of your technology data stack. And these tools need a lot of work to be successful. And that's what we set up MotherDuck and our other tools to do. Largely we're agnostic, but with MotherDuck, the thing we love about them is that they're a super lean, easy tool to use,
11:03and that they enable from the very early data analysts and engineers all the way up through the executive team. So the point is to really design this in the most lean effective way, and to make sure that it's not just implementing a tool, but your team has the support to make sure that you're successful. So that's it for the team for now, but thank you so much.
11:28Okay, so now I get to do the fun part, which is a demo. So I'm gonna share my screen. Let me drop this in. And we're going to kind of just explore a data set together in a non-deterministic demo. These are always really fun. And they make me a bit nervous because we don't know what we're gonna get, but let's look.
11:52All right, so quickly, I'm in Claude. One thing that we do that's a little bit unique at MotherDuck is we are not bringing a walled garden to you from a perspective of like, hey, you need to use our AI model on top of your data. We let you just connect it to whatever you want, and I'll show you what that looks like in Claude, for example. I'm gonna just hit Settings, and I'm gonna go to Connectors. Okay, they've moved to Customized, of course.
12:17Okay, so you'll see that we have the MotherDuck MCP. It's right here. We run this for you. It's just part of your account. You get it as part of what you pay for, for MotherDuck. It is not an added fee. So you can just find it. You can hit the plus there, and you hit MotherDuck, and away you go. You click this, and then you can add it. You can add it to your deployment. And so as soon as that's in your, is part of your Claude, this also works in,
12:43this also works in ChatGPT as well. My question went away. Okay, well, that's fine. Let me go grab it. We're just gonna grab this question. New chat. Let's try this again. So we can see here that I have a question, and then just so you can see, here's my connector. I've just got the MotherDuck one turned on right now. And I'm just gonna hit Enter. And we'll see how smart Claude is at, A, finding the right tools, and looking for it.
13:08So it immediately identifies, hey, that sounds like it's in the MotherDuck database. Let's go see what's in there. And so it's doing some queries on things. You know, it's listing tables. It's telling us what's in this dataset, right? So this is an e-commerce curated home goods company that we have in here. And it's got, you know, some core entity tables, some marketing tables, and some operational behavioral tables as well.
13:33So we also have a net revenue definition here, which is quite nice. So, all right. So the first thing I'm doing here, actually, which is really interesting when you're thinking about interacting with LLMs and your data, is I'm actually asking a query here that's just about the data, not about a specific question that I have. And this is actually really important because what you want to do is hydrate the context of your model before you start interrogating the questions, right?
13:58It needs to know kind of the map before you ask the question so it can go to the right place in the map. Otherwise, it's just going to make assumptions about your data. And this is a really important step kind of as you think about working with this, right? So we're going to now ask another question. Let's see. Let's ask it about revenue over time. Let's see.
14:21And again, I'm asking like, you know, very, very basic questions here. I'm not like being too verbose. You know, I'm talking with it like I would maybe talk with an analyst, right? I'm not being too specific. You will find that, you know, depending on the model you're interacting with, that your specificity requirements may vary. Great. It's charting it. Lining up the dollars, it says.
14:47Thank you. Best month was July. Interesting. Maybe we sell like outdoor products or something. Who knows? All right, great. Got a nice chart. Quite a bit of variation month over month. Doesn't look seasonal. Okay. Very, very cool. All right, great. So do we want to drop into a daily, weekly view?
15:10No, I don't actually think so. I kind of want to drill into Q4 revenue. Let's look at Q4 revenue. Split it by category and collection, let's say, right? These are dimensions that we use to kind of track product attributes. I'm just telling it,
15:32just giving it a kind of general way to look at this. Again, not over specifying it because what we want to do is be able to kind of like find something interesting in here. All right. Cool, cool. It is visualizing it for us, which is fine. It's a little bit slower than I would like, for the sake of a demo, but that's how it goes sometimes.
15:58Let's go. Let's go, Claude. Come on. Furniture, dominant category. Okay, rug, second. Interesting. All right, very cool, right? So we can see this. Okay, few things worth noting. Outdoors, only two collections. Long tail categories. Okay, interesting. Do you want to compare this to see, dig into margins by collections? Yeah, good idea. Good idea.
16:23Let's look into margins by collection. Sure, let's do it. All right, I think I've actually hydrated it enough and I can start maybe asking more specific questions about my dataset here. I think margins, the margins is interesting here. It may help us find something interesting in the dataset.
16:49Ultimately, again, what we're trying to get to is not just what happened, but why it happened and what we can do about it, right? I think existing analytics really let us get into what happened and it gets really hard to kind of get into the next step in LLMs, help us kind of move up the value chain for analytics for sure. Okay. All right, it's still rendering its own charts.
17:14That's a fun demo. Let's see here. All right, I have a hunch that something's going on with discounts in this dataset. I'm gonna ask a question here. Yeah, it's a mixed bag. Sure, sure. Oh, this is interesting. Why is the margin bad on this?
17:38Hang on, this is surprising to me. Why is the margin low on, I'm just gonna copy this since I don't wanna type it. Look at discounts, quantities, et cetera. Okay, let's see what it found. Why is our margin lower here?
18:11Okay, revenue-weighted discount is higher. Why would that be? Heavier discounting is landing the higher revenue skews specifically, let's look. Okay. Claude is so funny. Whoa, every single product is this one product name. It's the one collection, that's so crazy. Interesting. I do appreciate this.
18:35All right. Cost-effectiveness. All right. Cost ratio is fine. So what's causing the margin gap? Let's check the discount distribution. Let's see if it can discover what's going on here. There it is. Okay. Interesting. Okay, so only 43% of our sales are at full price on this one, right? So we're discounting this one and the discount's actually working to drive revenue, right?
18:58Meaningfully enough for us to detect a margin change on this product. This is really interesting, right? Okay, let's keep going. All right, it wants to just like chart it for us, which is totally fine.
19:21I'm just gonna say, stop charting, use tables. We're doing a demo. Okay, let's see. Can you look at discount rate versus quantity sold? For this item. Maybe there's a bundle. Maybe we have a bundle or something. I don't know.
19:46Let's find out. All right. Good call, switching to tables. Thank you. Thank you, thank you. All right, I think we're getting pretty clear on an understanding of what's happening. All right, here we go. Look at this. the road for quantity equals five. Cool. Cool. This is almost a buy five, get 50% off bundle promo. Oh yeah, that does make sense. Confirmed. Okay, great.
20:10And it checked against some other collections just to see if this bundle existed. So it doesn't. Okay, great. Let's see here. Fully real and traceable, deliberate bundle, plumb, promo, whatever. Um, let's see. You want to look at either? Let's see. No, I don't really want to look at that. What I think I want to do is, um, uh, let's just validate. Let's just validate that this, okay.
20:35So this is the source of our gross margin change, right? Gross margin change. Correct. Okay. Let's just make sure. Hold on. Oh, check the bundle is new or has always been there. That's a good point. That's actually not the question I wanted. There's no margin change to explain.
20:59Okay. Stop that. The question that I wanted to ask was, um, sorry, what I meant was this explains why the margin is lower in this category.
21:20Okay. Yeah, that's fine. All right. Great. Yes. Yes. Uh, I appreciate you, Claude. Um, let's see here how this shapes out. All right. Interesting. Higher sourcing cost ratio. Yeah. Yep. Yep. Yep. Higher revenue weighted discounts. Yep. That makes sense. Okay. Cool. Um, this part makes sense. So, so now we kind of know like what happened,
21:42which is we had lower margin. Why did it happen? Okay. Well it happened because, um, uh, it, it happened because there was a discount, right? There was a bundle discount that gave a, uh, you know, um, a better discount on the product and it actually did drive units enough that we detected it. Right. So now the question is like, well, what should we do about this? We can be like pretty naive here. Uh,
22:05what should we do about this? Perhaps I'll suggest a couple of things. Perhaps we can like simulate how this lift may affect other products. Um, right. Like we only see discounting kind of this steep on this one product. Maybe we could run it somewhere else, right? Maybe we can make more money.
22:26Let's see. Risky without a baseline. Sure. Um, I agree with this, right? Okay. So how do we test it? How do we test this hypothesis? Um, and see if it works. Great. Great. All right. And it's validating it against a campaign attribution. This is cool. Um, you know, are the, is there a campaign here? Yeah. The bundle orders are messy.
22:50There's not a lot on what to do this. Okay. Okay. Uh, let's see. Uh, I think this lines up to option two. All right. So I'm just going to say, Hey, like, let's build a test, right? Good option. Do the right call. How do we test it?
23:13Basic mechanic. Okay. So it gives us a, you know, an obvious candidate. It says, you know, obviously do your QED, do your own research on this, uh, as it relates to your products and how to think about this. But again, we've gotten from something where we're able to explore a dataset, identify an outlier, right. Identify what triggered it or and why it occurred and then able to design and get
23:35some recommendations. Right. For me, like as someone who, you know, before I worked at MotherDuck was working in the data every day, like I would spend days on this type of research and it's just compressed all of that for me into, you know, what is this? A 15 minute demo. Um, this is a crazy, this is like, what's possible here. Right. Now, obviously there's a lot of things you need to do to get to the point where you can ask questions this way. Right.
23:57You need to have a really nice dataset that you can look at, but like the, the amount of noise that Claude, as you noticed was filtering out kind of as we were going here is actually really high and it's writing really good SQL. And like, that's the crazy part about how all of this works. Right. So it's, it's giving us, it's giving us a full plan. Right. Um, you know, and if I want to justify it, I could probably even be like, you know, okay, uh,
24:21help, help me understand what the lift could be, what the lift could be so I can pitch this to my boss. Um, so now we have kind of like some, some, some sort of way to know, Hey, like, how do we, how do we say, Hey, we've discovered something. We have an idea on what to do about it. How do I get this moving in my business so that I can actually start beginning
24:45to experiment with, you know, Hey, is this discounting actually working? What is the actual upside? You know, you know, do we have, do we have a lot of money that we could drive from this? Um, I don't think we need to see the kind of the, the rest of the answers. I think that the pattern here is pretty clear on how to use, you know, your data with MotherDuck and, and an MCP and the MCP with something like Claude or ChatGPT to really drill into this stuff. Um,
25:08so I think we're going to stop sharing here. I'm going to, um, stop sharing and then we will jump into the next slide. Thank you for pulling that up. All right. Um, all right. Nice. Thank you, Jacob. That was awesome. Yeah.
25:32It's really cool to see you just kind of work through that natively like we normally do in LLMs. Like you just start asking questions as they come to you. And not only that, but Claude is prompting you to maybe ask other questions that are important. And that's the workflow that we're used to now. I think everything is going to this LLM approach,
25:54but how do you make it so you can actually trust the data? Uh, we mentioned a number of times, it's about the data set. It's about the data integrity and being sure that you have the right data and set up properly so that MotherDuck can do its job to give you those answers. Because of course, if you're going through this LLM and you don't have good data,
26:17it's going to be spitting out information and bad recommendations that could really affect your business. Right? And so what we're looking at is set up MotherDuck to be successful. Look at your overall technology stack and eliminate a lot of these tools that aren't necessarily needed in your business for where you are right now.
26:38And again, talking about the sales approach and the technology landscape, there are so many tools that you are being pitched every day on LinkedIn and otherwise that are just over complicating your stack to a large degree. And what I love about MotherDuck is it's the center of your data universe, so to speak. If you can get things in there of the applications and the data sets that you
27:02use and you can set that up in a solid data model with a team like silver creek to help you get there or your own. But as long as that data is good, you can now avoid having to use a lot of these other tools. And I think if you zoom out from the overall perspective of the industry, we're seeing a lot of consolidation of these tools, these, these data warehouses and others.
27:26And a lot of these loose players are just kind of disappearing because they're not as effective anymore. And on the visualization layer, you saw Claude spit out graphs and charts just on demand real quick. Those two week old month old static reports that nobody looks at that the data team took a month to build that is now obsolete. And so you're seeing all of the visualization layers.
27:50They are embedding this LLM approach as well, but MotherDuck is already doing that. And if you're looking at Pareto principle, you're going to be able to solve 80% of your analysis just strictly within this tool. And then for your extended use cases, sure, you know, add on as necessary, but that's the really cool thing that MotherDuck is doing.
28:10And as a kind of a newer product that is coming out here comparatively to the others, it's building natively like this, not retroactively consolidating to kind of catch up to MotherDuck. So I love how it's leading the way, but again, we want to look at the big picture and how to get that data and the right use case and to set you up for success. Because I know when you look at this, it can be overwhelming,
28:35but really it's a step by step approach to get you successful. And if you start small, which means fast speed and low costs, now you can start to expand based on the success. And it's not this huge undertaking. So let's go into the steps on the next slide. There you go. So we've tried to narrow it down just to a simple five.
28:58But you guys kind of already know this, right? We want to understand what's the use case. We want to have a clear understanding of your current state analysis, where you're at in the business today, the current goals, challenges, and what are the right data sets that solve that business problem. So don't think about the data first. Think about what are you really trying to solve as a business? And then what connects to that use case? Where's that data?
29:23And then what's the stack that you really need? What applications are housing that data? And how do we just move that into MotherDuck? And there's so many great lean, efficient, autonomous ways to do that that can really free up the time of your data engineers and get business users activated quickly so that the team, the data team can spend their time on solving business challenges as well,
29:46helping the company grow, making more sales versus, Hey, can you create this report? I probably won't look at it. That type of thing. So we want to build and activate and that's what we're here to do. So that's what this paddle session is about. You know, we just went into a dive. We're talking kind of jokingly like the paddle before you dive. This is a level set.
30:10And if you hit that square code and we have this at the beginning of the conversation, but the square code will lead you into this paddle session. And this is really cool because this is something where it's not just MotherDuck and it's not just Silver Creek. It's us working together MotherDuck and Silver Creek from a technology perspective, as well as a sales perspective, or I mean, a product or services, excuse me,
30:35to help you generate this strategy and use case. So let's go to the next slide and we'll kind of talk about this paddle session in detail. But what this is is it's a 45 minute session with both of us. And this is senior data people, people on the phone, MotherDuck people to help really just explore your strategy, your specific business. And what are you currently doing today?
30:59This is not a sales call. This is really just a strategy session. Of course, if you want to buy from us later, we're not going to say no, but we want to help you identify a use case and set you on the right path. So you can decide what resources do you need to be successful with that? And you might have a fully enabled team already and you'll be on a good path, but at least you can leverage experts from MotherDuck specifically,
31:23as well as our team to help you get through and generate the best case scenario and also minimize a lot of mistakes. You know, I'll say this, when talking with customers with great teams, they are naturally confused about exactly what we talked about today, the technology landscape, how do we put it all together? What's really needed? What's not? And if you can eliminate a lot of those things,
31:47you're now saving a lot of time because you're not experimenting with things that we've already kind of proven that won't work or high unlikelihood to work. And so let's work together, collaborate, take advantage of the session. You know, it's 45 minutes for you guys to kind of bounce ideas and answer any questions that you might have. And we're happy to do that with zero obligation of anything else. We just want to help out. So that's the survey.
32:12It's just a quick qualification, five questions, and we'll be in contact to set that up. So thanks so much. I'm sure this will be sent out after as well, but that's all I got. Some more resources. Okay. And then yeah, let's jump into some questions. Okay, great. I'm going to pull up the first one here.
32:35I think this is a great question actually for both sides of the team here. So I'm going to just pull this one up. How do you present, prevent hallucinations using this setup? And what about David data privacy? I'll start on the data side and then actually let's do this. Daniel, Daniel or Gordon, do you want to, do you want to start and then like throw it to me from there?
33:00Sure. I can, let me start from there. And so I have this thing, FEMA, I'd like to talk about what, you know, what's old is it's new again. So we keep talking about hallucinations, like there's something that's isolated to only LLMs. You know, it turns out humans hallucinate as well, right? I know it's come up with results that are incorrect because it was based on top of all the foundations, previous analyses, the wrong context, so on and so on.
33:24Can't tell you no more time. Someone's gotten LTV wrong, or even just measuring revenue. You know, I had a CFO once who used to include sales tax and revenue, right? Very unusual. And so hallucinations happens and the solution is the same in each case. So it's building successive layers of trusted foundations. So you were minimizing the blast radius of doing this kind of analysis, right? If we're abusing the arms against, you know,
33:48a reliable MotherDuck database that has defined metrics in a symmetric layers, so on and so on, star schemas, defined dimensions, no, no duplicates, what have you, you're building on a much more solid foundation, right? And you have less opportunity for a hallucination. Also Jacob mentioned hydrating your session. Now that's something he did on the fly during that particular session, but that's something that can be applied at a systemic level, right?
34:12Your agents can be, can have a controlled experience where they're given instructions, they know how to do certain things and they're working with you. The third thing is when you take down with people, you know, you also minimize hallucinations by bringing in subject matter experts. What we're doing now with LLMs is now allowing people who are subject matter experts in the business to ask questions as opposed to someone like me who,
34:37you know, I'm pretty good technically, but I'm not a marketing analyst. I'm not a finance analyst. Previously I was building reports for marketing people and I'm pretending I'm a marketing person. I'm going to hallucinate, right? So the very nature of who's asking the question, who's engaging with the agent is going to help reduce this. No, any of course you have all the normal tactics like bringing vector databases, bringing in rag and so on. And then that's all something that gets done as part of the solution.
35:01But really thinking about it from a topography perspective, you're what you're doing is building trusted layers. So you're bringing that trusted foundation to the user. So Dan, you got anything to add to that? No, I mean you said it perfectly. And I think really in its simplest form, we're just getting the right data foundation into MotherDuck.
35:22And so part of that data integration is going to be the semantics and all the things that we need to make sure that we have trusted data definitions and that everything is secure within your environment. And so moving on to the data privacy and maybe Jacob, you have some things technically to speak about on this, but this is a question that we get from clients all the time because you don't want to just take your normal data and throw it into Claude.
35:47This is a different setup because it's pulling from MotherDuck via the MCP and that is a key integration that's going to keep your data secure and not just send it out to, you know, OpenAI and Claude. So it's really that integration and setting up the foundation on a secure setup. Jacob, you agree? Yeah, I think I agree. I mean, I think like, again,
36:10part of the reason why we're thinking about, you know, data privacy or like our integration with LLMs a little bit differently is like data privacy is, you know, that part of it in terms of what LLMs sees what and trains on what, like that part of it is governed in your relationship with your LLM vendor, right? We don't want to get in the middle of that. We want to empower you to use whatever you have from your, you know,
36:33enterprise agreement standpoint. So, you know, I think, I think that's a little bit of a different perspective versus like, I don't know, someone else who might be having their own LLM with their own, you know, maybe they're, they're self hosting it. Maybe they're, you know, using Fireworks or somebody else who knows, you know, we don't want to add that complication. I think the second part is, you know, the trusted layers of trust of trust is like the key thing, right? And, and on the MotherDuck side, we're building some abstractions.
36:58We haven't launched them yet, but one of them is like a context layer so that you can write in the MCP. Actually you may, if you were paying really close attention to the demo, you saw that it did pull context about some definitions, actually the definition of revenue right into the chat and say, Oh, Hey, I know how to calculate revenue already. Great. I need to exclude, you know, returns and I need to, you know, subtract discounts. Right.
37:20So new that already. So we're thinking about, you know, how to build those things out and the cool thing I'll call, I'll call it two things. So the first thing is we want to make MCP tools so you can use the MCP to do it. The second thing that we're doing, which I think is, which is unique or is something we carry as a first principle that maybe not every other vendor does is everything that you can access with the MCP is also a
37:45SQL function. And so if you want to, for example, govern your context in some other system, and then maybe it's, and when I say that, I mean like get, for example, and then you want to just run a SQL query that like post that into into your database. Well now it's, it's all in sync. It's kept in sync. You have a central place that your engineers know how to use. It's very easy to collaborate on and you can just ship that,
38:10ship that code into prod with a very, with a deterministic function. Right? So everything that is non-deterministic, maybe in an LLM interface, right? Via MCP. We also have the deterministic version for once you've discovered the thing and need to repeat it. And I think that's a key thing that we're always thinking about is like, okay, like LLMs are really good for discovery, but not repeatability. So we want to make it so you can discover it and then repeat it. And so,
38:34you know, we want to build the infrastructure to let you do that. We're not trying to build a SAS application that does that for you. We want to build the infrastructure so you can build the thing that you want on top of that. And so I think that's, that's kind of how we're thinking about it. Yeah. Really great, great question. I don't see, oops, I don't see any more questions. So I think we can wrap up for now. Daniel,
38:59Gordon Fee, any closing comments before we wrap up? Anything from you, Gordon Fee? Oh, I think I've, I think I've said my piece. I think I wanted to iterate one thing on the previous slide that I forgot. I do want to just keep saying that we're a small international consultancy on purpose.
39:22We all know that business is personal and that a small team means faster everything, faster communication, faster implementation, faster turnaround. And even with MotherDuck, we want you to get everything that you need as fast as possible. There won't be any delays, any excuses. You need anything, come to us, both MotherDuck and Silver Creek Insights and we will help you solve your
39:44problems. Thank you, Fee. It's always great work. Our customers love Fee. I don't know why, but no, I'm just kidding. She's awesome. So, and last remarks from me, I mean, we talked about a lot today and obviously there's going to be some more questions specifically about. business, and that's what we'll take a look at in that paddle session. So hopefully we
40:05can schedule that with the attendees here and look forward to talking with you guys on that. But thanks so much for being here. Thank you, Jacob. Thanks, everyone. All right, we'll wrap up. I'm going to hit the button, Jerel.
