Sponsored by: Slalom | Scale What Works: Modernization, ROI, and Adoption on Databricks

Summary

Slalom's AI success framework—fit plus value plus organizational enablement, minus the drag of legacy systems—provides a structured approach to scaling enterprise AI on the Databricks Data and AI platform.
Slalom's accelerators include VelocityIQ for modernizing legacy data platforms, Lakefoundry for translating ideas into executable Databricks solution plans, and an AI Value Platform for quantifying use-case ROI upfront before committing engineering resources.
A Calgary Flames scouting scenario demonstrates how legacy on-premise data can be modernized, analyzed with Databricks, and surfaced through AI-generated analytics to support data-driven draft decisions.

Chapters

FAQs

What is Slalom's framework for scaling AI on Databricks?

Slalom's framework expresses AI success as fit plus value plus organizational enablement, with legacy system drag as the subtracting factor. The framework maps to three accelerators: VelocityIQ for modernization, the AI Value Platform for ROI quantification, and Lakefoundry for turning solution plans into executable Databricks implementations.

What is VelocityIQ and what does it do?

VelocityIQ is a Slalom accelerator for analyzing and migrating legacy databases to the Databricks Data and AI platform. In this video, it is used to upload and analyze legacy SQL code, automatically identify migration opportunities, and generate a governed Unity Catalog setup for a migration project.

How does the AI Value Platform help organizations justify AI investments?

The AI Value Platform helps organizations quantify the ROI of specific AI use cases before committing resources by modeling five-year value, one-time costs, and decision impact. The Calgary Flames scouting scenario demonstrates how it models the value of a data-driven player valuation system against its implementation costs.

Why is organizational enablement essential for moving AI pilots to production?

Technical fit and business value alone are not enough—organizations must also enable employees to adopt new ways of working. According to Slalom, without deliberate organizational enablement, AI initiatives stall at the pilot stage regardless of platform quality, which is why it is treated as a core component of the AI success equation.

Full transcript

[00:10] All right. I'm ready to get started. I might get a little bit over excited, so I might be a little loud. If it's too loud, let me know. I'm going to try to talk a little quieter. I heard that there's a couple of folks that had a some issues with that throughout the day. My name is Sebastian
[00:22] Richter. I work for Slalom. And I get to lead our Databricks business unit as part of that. Just a couple things as usual, right? It's been prepared for information
[00:34] purposes only. Yada yada yada. We know and you've seen this all day long. Please make sure that you complete the service after as well. I'll have that on there. But today what I get to talk about is is very much our
[00:47] a little bit of a journey that that I have been doing on on behalf of what I call the Calgary Flames. A little bit of background. I love hockey. I'm a big hockey guy. I live in Calgary,
[00:59] Alberta. And the Calgary Flames have a massive draft at the end of this month. And there is I think there's about 20 draft picks that they have at the end of this month.
[01:11] And and working with organizations and sports organizations, there's still a lot of uh lot of the legacy data, the scouting data and everything else is on premise. So what I wanted to do is is show you a little bit how I would work with Slalom
[01:24] in order to actually go through modernizing it and modernizing the application that I'll be building, but also just thinking about how do I as the let's pretend I am the head of analytics
[01:36] for the Calgary Flames or the Calgary Hockey Club as I need to call it here. Um if I am the head of analytics and my GM is asking me who should I pick next, what should I do? How do I get to that and what does it look like in a Databricks architecture?
[01:48] So that's what I'm really going through. And and for us we just call this like from a scouting data to governed draft decisions. Uh just very quickly again, I just kind of went through that, but what I wanted to actually do is just very, uh,
[02:01] very much talk to you about the framework as we're thinking about AI and like how do we utilize some of our some of our formulas and some of our overall, um, approach on on the Slalom side to a
[02:14] solution like this and to an issue like this. Um, we we think of something that can be really applied to AI as AI success equals fit plus value plus organ enablement, but we have the
[02:27] drag of the leg- legacy systems underneath. So, for for us it's always been thinking about how do I actually modernize that? How do I get this legacy system out of the way so I can start running? Now, the fun part is I get to do this,
[02:40] um, a little bit more in in concert with the solutions that we've been building as part of our Databricks business. So, that's what I want to walk you through today. It's a little ambitious. I have like about 60 minutes
[02:53] left. I'm going to do three different demos. So, bear with me. If they're a little quick for you and you want to, uh, talk about this after, um, we can certainly, uh, find some time there as well. But, just to, um,
[03:05] paint the picture a little bit. Um, what you're going to see today is, uh, our AI value platform. So, really does the entire problem statement that I just talked about actually apply and will I make money off of the investment that I'm making here with a third party, um,
[03:19] to be able to, uh, to be able to run. Is then I need to think about how do I actually modernize? How do I modernize into Databricks? How do I get our SQL Server environment into Databricks? So, I'm going to show you Velocity IQ.
[03:31] And then at the end of it, how do I actually build on top of what I'm building in Databricks? So, let's say I have everything modernized in in Databricks, how does Slalom, how do we think about adding and building on top of that in a
[03:43] in a scalable MVP ready way? Everybody good with that? I got 60 minutes. Let's go. So, the first thing I need to do needed to do, and I'm just going to go straight into most of the live demo side, and I can make this a little bigger.
[03:55] Uh you before I go fully into this, the AI value platform is something that we have been working on for about 2 3 years now, and it's been you can exchange AI with data value platform or project value
[04:08] platform. That's really what this is. But, what I've been doing as part of all of this is is doing a little bit of a calculation for what we call what I of course call the Draft Day Command Center. What I needed to do is uh create some
[04:20] briefs beforehand and be able to add those into the use case overview. This use case overview really is the context behind the uh be- behind the LLM uh that's actually running a lot of the the decisions as
[04:33] part of the AI value platform. Now, there's a bunch of assumptions in here. Um number of business users and for us is a high value or not. I did some research before. Uh just so you know, a good draft pick
[04:47] is anywhere um if you get them into the NHL. Also, you make a good draft pick, they go through, they get into the NHL. That leads to about 15 to 20 million dollars in profit to an organization, just one draft pick.
[05:00] Because it isn't just the player, it is the fans that are coming to see that player. It is the sales that we're doing um on the on the jerseys, on all of the other marketing side. So, there's a massive impact if you make the right draft decision.
[05:13] There's some effort that we need to think about, and this is where the value platform helped a little bit more. This is a complex uh it complex undertaking. How ready is my data? And I I think my data was pretty
[05:26] ready, but it was just in the wrong environment and in a legacy environment. How many resources do I need in order to do this, right? What's my time to implement this? And I know that it's it's fairly narrowly scoped, so I
[05:38] thought 4 months is probably a good amount of time. What I did then is add a little bit more of the risk assessment. So, is there is there a high risk to this or not? And some of these risks as we thinking about this, if I'm making this a little
[05:50] bigger, right? Can this AI system be used to create quantitative outputs or autonomous decisions? Yes, it can. Should we think about that as a risk? Maybe. It should be part of the risk profile.
[06:02] So, I took all of that and then married it with a little bit more of the benefits. So, if I just take one of these benefits as an example, but we're to edit that.
[06:16] If I if I put one of these benefits up here as a drive decision value improvement, right? And for us it is a revenue revenue increase just like I talked about. Going back and forth, doing some research and everything else, we think about per player it could be up to $3 million with 70% confident in that. What
[06:30] I'm going to do as part of this decision-making is actually just think about pulling this across 5 years. And I added a bunch more in there as well. Again, I want to talk about more of the technology once we actually get
[06:43] going. And then of course one time costs. This is a swag, this is a kind of stab in the dark, but what what is the third-party cost as part of that, right? In order to make the full calculation going. Needless to say,
[06:55] um investing in this is a great idea. If you think about the 15 or 20 million dollars that each player will get give us, right? And the initial investment maybe being 400 or 500k in third-party
[07:08] um third-party costs, why wouldn't we do this? Why wouldn't we undertake this? So, if you if you wanted to do this and compare more than one use case, this is really the way that Slalom does that.
[07:20] And we have we have really cool workshops as part of that where we can we can we can help you out with. Um but what I what I love doing with this is kind of taking this also and making this part and and the basis of the brief that I'm giving some of the
[07:33] tools um that I'll show you just a second to to um to build the next piece as part of this. Okay. So, now I've made the decision, yes, I'm going to go.
[07:45] Let's figure out what to do next. Now, the next thing is is I have a Scout DB database that is sitting um on my SQL server, and I need to get that into Databricks. So, the next thing that uh Slalom and I
[07:58] would do together is utilize our tool is called Velocity IQ. Velocity IQ, and if you need to learn a little bit more about that, the individual who helped build it sits right at the front here, Arun. Uh Velocity IQ is um our migration uh
[08:11] migration factory. And as we think about migration factory, there's a couple of different things, right? We think about um uh agents, teams of agents. We think about the best practice on the Databricks side. How do we get there? How do we modernize?
[08:24] And when we think about Velocity IQ for us, there's also an analysis piece as well. What am I actually moving? What does it look like? And I'm getting a little excited, so hopefully I'm not too loud. Um
[08:36] the first thing I need to do in Velocity IQ, of course, is set up the project, and you'll see a couple different projects here. I'm not going to bore you on that. But, what I think is very important to just think about is we need to we need to set some
[08:49] parameters, and we need to set some uh foundations in Unity Catalog to be able to govern govern this entire migration. Then I need to have a lake lake base back end for us because not just are we saving a lot of these uh pieces of
[09:01] information in and UC volumes, but we also need to have them readily available for the agents to to grab and go along. Then the other thing I need to do is kind of like I mentioned, uh make sure
[09:13] that we have permissions. Do you have the permissions set up? Am I Am I actually good to go on my Databricks environment with this tool? And then last but not least, and very, very important, I need to update uh
[09:25] upload my legacy code. And the way that I'm doing this is I'm actually doing an export of of the database, um but including DDLs and a couple other pieces in there. Um and then I actually fed this beforehand, a lot of the context
[09:39] documents around like this is what this database is. This is where how it's used, all of those different pieces. Next thing I did is ran a current state analysis. So, if you think about the database that I'm looking at, um it's
[09:52] fairly small in the way that the Scout Scouting site works. Um but there's about 28 tables that I needed to move that I have in here. And just maybe
[10:05] on on an overall view, it also does a lot of analysis of what the overall uh sequencing is from source to target. In this case, I don't even know if this works on here, probably not. But if you see on the top there, this is Scout Pro
[10:18] operational database. Right? Because I'm not moving the operational database. I am moving the analytical database and the data warehouse into into Databricks, right? So, of course I need to consider where all of this data comes
[10:30] from. And of course there's some CSVs because in the sports environment, not everybody has access to that. We will have CSVs in there as well. Now, here is the way that generally everything comes together. So, we bulk
[10:42] load, we have some ETL master pipelines and everything else. And then as you see and as you're probably used to, there's a little bit of back and forth between tables, between uh mechanisms, and between objects.
[10:56] We kind of end up at this Scout Pro data warehouse. All of this analysis is done within Velocity IQ, right? We look at the dependencies. We look at we look at the uh context documents and
[11:08] everything else that we're bringing. And we have the ability to show this to you fairly quick. And as we as we've been doing this with clients and and doing this with organizations, this is kind of the eye-opener. I'm like, "Oh my god, I know what my entire picture looks like."
[11:23] So, uh to shortcut this cuz I got 8 minutes because I wanted really talk about Lake Foundry as well. I went through and I did the entire migration. So, there's a lot of migration planning that I did Um and the velocity IQ does for us and we decided
[11:37] there's two groups, two waves to migrate into the Databricks system. So, I went through that and did that. Now, the next thing I did is utilize the Lake Foundry tool
[11:49] that we have at at Slalom. We Cameron did a tool talk on this yesterday. Um Lake Foundry is really if you if you think about the building foundation on
[12:01] Databricks and being able to interact with um with a product owner to be able to build something on Databricks, this is what Lake Foundry has in the like easiest sense possible. Again, Cam is here.
[12:14] If you if you wanted to have more conversations about that, we can actually do that together. But, what I did is I actually fed fed it the a brief. So, I build a brief together based on what what I found in Velocity IQ and what the table structure is going
[12:27] to look like and where it sits. And you see that it starts making decisions on as part of this Hey, this these are kind of the bronze level table set I will have. Here's the silver
[12:40] level silver tables and everything else. All I did is give that to Velocity IQ and Velocity IQ then sorry, Lake Foundry and Lake Foundry is like cool, I can build on top of this.
[12:52] As part of this, I also wanted to say, "Hey, I need an app in order to give to my executives to start making decisions on the draft." So, Velocity IQ went Velocity IQ provided it to us. Lake Foundry then built it for us. On the
[13:04] Lake Foundry side, I need to do some interactions with the agent the LLM judge back and forth to be able to say, "Yes, this is a good idea. This is not a good idea." And that's kind of the interaction that you see here. Once I get to a point of having a spec
[13:16] ready environment, I actually can go through here and and show this back off and say, "Hey, how good is the overall quality of my spec?" Right? If you're going to going to go below 70 kind of gets a little bit dicey, right?
[13:29] The uh uh the agents will make up some of the missing pieces that we're not providing. Anything about 70, it's it's it's a pretty decent score for us to start building the first iteration. And Lake Foundry also, as part of this,
[13:41] puts this into more manageable engineering phases, right? So, if you see here, we've got the phase one, which is data engineering, phase two, which is MLIR, and then the Databricks app that we need
[13:53] to build on top of that. So, three distinct phases. So, I kicked that off as well. And little shout out as as well as part of this, as I was building this entire environment, um
[14:05] I was using our Lake Foundry UI, and then we as a team have also been working as making Lake Foundry available within Codex, which is which is what we're using internally at uh at Slalom. So, I
[14:17] actually worked on it in in both sides, but really what what we have the ability to do within uh with Lake Foundry is utilize the AI dev kit, marry it with our standards on the Slalom side, let's go build Databricks MVP ready products
[14:31] fairly quick, right? So, I came up with my full Calgary Hockey Club draft uh ap- application, right? There's a couple different things. Again, I said
[14:44] we have a lot of draft picks. Um just 10% in the first four rounds go to Calgary, which is a crazy amount of drafts. Um and really what I wanted to show off is going to be hey, round one,
[14:57] pick six, who should I who should I go for, right? And there's a recommended move, which is trade down and target uh Victor Bjork. The reason why it's actually doing that um is there's a couple different things.
[15:10] Calgary has a lot of defenders right now. We need wingers. We need uh we need centers. So, it's starting to think about those um those pieces as part of the overall decision-making.
[15:22] What I can also do on the side and in my in my aspect that I built for Lake Foundry I said I also need to have Genie integration as part of this. I want to be able to interact with the data. So, for every every single pick here I can say I can interact with
[15:35] Genie. In this case I can now say who do we think will go immediately before uh before pick six. So, this is a Genie call in there as well. So, as we start thinking about how do I very quickly get
[15:47] to that MVP? How do I very quickly get to something that my executive can look at? We've We've been able to do this at scale, at velocity. Oh, velocity is a little bit of a fun there. But, uh
[16:00] at velocity, um especially in the last three to four months. Um and the Wi-Fi is kind of No, this wasn't helping me out earlier. Um very very quickly. So,
[16:12] I I've enjoyed the entire way of thinking about like just Think about what I've just done, right? Like I've gone from a legacy environment into a Databricks environment utilizing Agentyc tools. And then, yeah, we finally have
[16:24] some have some stuff back here and some of the answers The players projected to go immediately before are K'Von Fairweather and Mr. Rudolph. Which is exactly what I would
[16:37] what I was expect, right? I I'm pretty deep in this stuff in in my mind, so I can actually fact-check that pretty quick. The other thing that I thought would be a good idea as part of my overall
[16:55] draft um draft application is can I compare players to each other, right? Can I just kind of put them next to each other? And a lot of this like UI interface and everything else, that's actually the last mile.
[17:07] Right? The data is there. I just need to go back and forth. And this is where some of the plugins and everything else that we can see on the Codex side have been very very helpful. Databricks as a tool um and and a platform
[17:21] is is quite nimble on the on the application side and we can actually ask to do it a lot like API calls going out and everything else is part of the overall application. So, that's happening in the background here as well. By the way, all of this is powered
[17:33] by Lake Base. So, I've got Lake Base running in the back end. In the way that we are going and have been promoting applications within Slalom, and going forward has always been based
[17:46] on Lake Base simply because the quick API behind it makes this feel like a real website. If you do a serverless SQL server, it's always have the spinning running thing for a little bit on Streamlit app kind of thing.
[17:58] Unless you beef that up to crazy costs. So, the Lake Base side has actually been a little bit more performant for us as we go in the background. The one more thing that I I'm a little proud of, but I also like because of course the
[18:11] individuals that I have to give all of this to really like a printout. So, I've just been thinking about how do I make this as easy as possible as part of this. This been a bunch of iterations in here,
[18:24] but I like the what if analysis that it also puts on here. So, what if the board shifts as part of the entire thing? Give it to them as well. This is for the for the executives that don't really know how to use AI yet. So, I also
[18:36] needed to have that as part of this as well. Um To summarize and just to um just leave it with one thing, I have loved in the last 6 months thinking about how
[18:48] agents, teams of agents are can start thinking can start accelerating all of the creativity that was always within us, right? Our entire team every week is like, I can do something cooler now
[19:00] because I don't have to worry about this cuz it's being done by a team of agents. So, that creativity is coming to our team and what I'm showing off here today is is just part of the things that we've been building.
[19:13] And I got to talk about hockey for a little bit as well. So, um with that, thank you very much. Thank um and I hope you have a safe travel home either today or tomorrow. Thank you.

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.