Sponsored by: Retool | The Last Mile of Enterprise AI Adoption

Summary

Reckitt—a global consumer health and hygiene company operating across 60-plus countries with brands including Lysol and Durex—built its Gen AI platform on Databricks and partnered with Retool to deliver governed marketing applications to non-technical teams, reducing insight delivery time by up to 60 percent.
Retool enables 'vibe coding' on top of Databricks by letting builders prompt production-ready applications into existence in minutes, with Unity Catalog permissions applied by default so governance is not an afterthought but built into every app from the start.
This video includes a live demo building a Lakebase grocery app from scratch on stage, showing how Retool's coding agent unifies data sources including Lakebase and Salesforce and deploys a governed, write-back application in approximately 15 minutes.

Chapters

FAQs

What is the 'last mile of enterprise AI adoption' problem?

This video defines the last mile as the gap between AI models and insights built on Databricks and the production applications that frontline business teams actually use. The challenge is not building the AI—it is shipping governed, scalable applications fast enough that business users can act on insights without waiting weeks for IT delivery.

What is vibe coding in Retool and how does it stay governed?

Vibe coding in Retool means prompting a governed application into existence using natural language, with the platform's coding agent generating the UI, data connections, and permissions layer automatically. This video demonstrates how Unity Catalog permissions are applied by default during the build process, ensuring that apps built through AI-assisted coding inherit Databricks governance without requiring manual security configuration.

What did Reckitt achieve by building its Gen AI platform on Databricks with Retool?

Reckitt reduced marketing insight delivery time by up to 60 percent by replacing manual reporting workflows with AI-powered applications that give marketing teams direct, governed access to consumer insights. According to this video, this freed marketing teams to act on data faster and enabled Reckitt to scale AI-driven personalization across its global brand portfolio.

How does Retool integrate Lakebase and Salesforce in a single application?

Retool connects to multiple data sources simultaneously, including Databricks Lakebase (the serverless Postgres offering) and Salesforce, allowing a single application to unify transactional data with CRM context. This video shows a live demo where a Lakebase grocery application is extended to pull Salesforce data alongside Databricks lakehouse data in a single governed interface.

Full transcript

[00:08] Thought I'd start with a question, right? Who here has tried to build applications and put them into production so people can access them? Pretty much everyone here good experience, somewhat painful,
[00:20] somewhat frictional, challenging at times. Cool. Okay. So, that's what we're going to talk about today. We're going to talk about how we help using the uh the example of Wreckit what we did with them of how we help
[00:32] with the last mile enablement of getting production applications built on data sets in into production. I'm Alex Jones. Uh I run international partners at Retool based out of London. So my first
[00:44] data bricks conference. So great to come here. And with me today is nice to meet you all. I'm Tom Kuka. I'm the solutions engineering leader here for enterprise. And I'm actually based in Toronto. This is my third conference and I'm really
[00:56] excited to demo our platform to you today.
[01:08] mission and our goal is to help people build good software, right? We spent nearly a decade helping companies do just that. Our mission today is very much ensuring around AIdriven building,
[01:20] shipping, deployment into production is fast for builders but also safe for the business. And together with data bricks, we're making vibe coding safe for everybody. We're also a database
[01:33] customer as well as a partner. So we use both lakehouse and lakebase. We've got over about 10,000 customers worldwide. people ranging from Amazon to OpenAI to Curser to more traditional businesses in
[01:46] financial services, retail, pharmaceutical like Nova, Nordisk, and Wreckit in consumer package goods. But I don't know about you guys, I'm I'm kind of a visual person. So, we're just going
[01:58] to run a click a quick vignette just to show what that experience looks like of when you're building a VIP coding application in Retool. Hopefully, this should play. Yes.
[02:15] So what you're seeing here is our interface. This is how you
[02:30] but it's allowing you to effectively build production ready applications in a moment of in a short matter of time in minutes. hear you over the music. >> Okay.
[03:08] So, as you can see in the video, it's unifying multiple data sources and now we're actually setting the permissions layer, which I'll actually demo for you shortly towards the second half of the break breakout session. So hopefully that gives you a little kind of feel of
[03:20] what it looks like from a user perspective to build application and we'll go into that in more detail with Tom. But you're not here to hear about RET, you're here to hear about Wreckit. But before we go on to Wreckit, what we thought we'd do is try and build you an
[03:33] application live while we're all in the room together. So question again for you. What kind of applications do you guys build? What kind of thing would be useful? Is it datadriven application, operational application? just if you got
[03:46] an idea, maybe just shout it out and we'll uh we'll see if we can put something together. Otherwise, Tom's got a couple of ideas to show you >> or who's had to work with a team where you had to prepare some data for for a team to actually build out some
[03:58] applications on top of data bricks. Would love to hear any use cases, any stories, things that you've had to done in the last year. I think we have one hand up over here.
[04:26] >> Okay. And the in that scenario, what what was maybe the the backdrop maybe the vertical? Was it like a marketing thing, a financial thing? >> Maybe like something like that. Can you repeat that? I don't think that
[04:38] mic is working. >> Oh, what was that? >> Grocery product. >> Grocery product. Okay. >> Is this one muted? >> I think the mic's muted, but grocery products. Okay, great. So, we're going to do a quick little uh quick little
[04:52] breakout from the breakout presentation. We're going to actually seed some data into Lakebase around grocery products. And we're doing a live demo. So, it's a breakout. It's not a keynote. We're going to keep it loose and uh we're not
[05:04] going to give you just recording. So, let me switch over to my laptop real quick. Okay, I think we have to exit the presentation.
[05:25] great. Of course, we tested it right as uh we came in and it was working fine. But let me just unplug and replplug and see what happens.
[05:40] Demo gods are not kind right off the bat. Okay, let's see now. There we go. We're up and running. So, what you see here is like the retool platform. What we're going to do is we're going to create an
[05:53] app and using that application interface, we're going to seed some data. So, grocery product reporting, you want to make an application around decisioning for that. Okay. So right now we have our effectively our agent
[06:05] builder that can tie into all of your data sources. So first we're going to ask it to seed some data into lake into lakebase. So uh let's see uh can you create
[06:17] seed data into and I will actually then search for my lakebase resource. Okay. And there we go. There's my Tom test lake base
[06:29] and prefix the tables with let's say uh SFO dBrx live and from there uh I will say
[06:43] you know I'm going to switch over to whisperflow here product plug this application will be for gross reporting data make sure there's only three tables maximum 20 column uh 20 rows at most and
[06:56] Let's get to it. Okay, we got that. Uh, this will be for let's see, I'll just make that quick little adjustment. Make sure there's only three tables maximum, 20 rows on most. Let's get to it. Okay, and uh
[07:10] let's say this will be a decisioning application that we'll make in the future. For now, just see the data. Okay, so we're going to have that uh kick off. We're going to come back to it
[07:22] in a couple minutes, but I have my lake base up and running here. We'll see if the tables come in. We'll see the seed data and then we'll build an application on top shortly after. So, that was a quick little intermission to get the use case. We're going to go back to our main
[07:35] presentation and check in with me in a little bit.
[07:49] >> Perfect. So, what we want to talk to you guys is about Wreckit, who's a consumer package good company, customer of ours. Um, if you're not familiar with them, they're in global uh consumer health hygiene company. You probably know some of their brands, Lysol, Durex, Mucinx.
[08:03] Um, and like most enterprises, they're heavily invested in the data platform in this case with with data bricks, but they hit what we see as a very familiar kind of problem or wall within the enterprise in that the data platform is
[08:16] very modern. It's very flexible but the operational workflows around it were where they were getting stuck where they were having challenges and they were very intentional about the way they went around this. They worked with Boston
[08:28] Consulting Group to go and consult with the business understand kind of where the challenges were uh go function by function and they decided that they wanted to focus on marketing as the primary use case. They want to test out
[08:40] how they would solve for this problem how they would solve that last mile of of AI adoption. So when we look closely at the marketing organization um there's really three kind of
[08:52] challenges that they were seeing. One was around technology right even with data bricks unifying a lot of the data there were some data sets that lived outside of that that they needed to connect to as well but most of it was in
[09:05] data bricks the insights lived in dashboards but they didn't automatically really drive the execution of what the marketeteers were trying to do. Secondly, the workflows around that the process were extremely manual. So the
[09:19] process looks something like you know we want to generate insights from data bricks put them into Excel emails then go around. You're probably familiar with these sorts of processes within your business and then approvals eventually
[09:32] get made. It was creating a very long drawn out process. A lot of manual steps in there and created weeks long before they could really get a weekl long process before they could really get the insights out. And the third thing was really around the people. You know, they
[09:45] had highly paid, highly skilled marketeers who were really getting into their hands dirty on data, looking at extracting data into the Excelss, etc. rather than focusing on what they were
[09:57] paid to do, which is more around creativity and strategy. So, how do we free them up from some of those workflows and how do we kind of solve for that problem? So, this is kind of what the the existing process looked
[10:09] like. As you can see, you know, it's very much founded in data bricks, but you can see all the steps, siloed teams, inflexible output, repetitive analysis, etc. So, we looked at how we analyze that process, where the gaps were, and
[10:22] then how we could reimagine that within retool to help the marketing organization, and really provide the close the gap between insights and action, that last mile problem we'll
[10:34] talk about. And this is this is what we reimagined with them in terms of the process. We started to uh extend the value of data bricks into that operational layer for the marketing teams. And this is now rolled out across
[10:47] uh thousand or plus thousand plus marketeteers within wreckit. Started with an initial group and then it been rolled out gradually over that. We're including things like data uh delta lake for unified data unity catalog for
[11:00] governance. We're integrated to all of these things. And the reason that we wanted to do this within marketing is because that's really where all the consumer insights, the data was living within that marketing organization.
[11:12] We're effectively the on top of that data foundation, we're effectively the application layer that allows the marketeers to integrate with that in interface with the data that's in data bricks and we're freeing up a huge
[11:24] amount of time for them. They're getting about 40% of their time back from where they were doing very manual processes to actually focus on more of the strategic stuff, more of the marketing side of things that they wanted to do, which is their day job. So, two days a week,
[11:36] we're kind of freeing up the time. This is the metrics that they've shared with us directly. Tom, what would you add in terms of the architecture and what we're doing with >> Yeah, you know what? So, really with
[11:48] data bricks as being a source of truth platform, there's so much value. There's so many insights that can be harvested out of that. But the reality is in the enterprise and in large organizations
[12:00] even big and small, it really is that last mile problem, right? The data is there, the insights are there. How do I actually do this at scale? Who can I partner with? What can I actually do to deploy these applications that I'm
[12:13] creating on top of this rich data uh into production? Right? So all of you uh to many degrees are using data bricks today. And from there uh you have your data layer right you have also your intelligence and government uh and
[12:26] governance with unity catalog and mosaic AI but you you're still need to actually you know build uh useful items on top applications workflows agents and there's a whole ecosystem out there
[12:39] within data bricks with data bricks partners like retool how can you actually do that at scale well with the retool platform we can actually connect to all of your data within data bicks
[12:51] and from there embed uh have arbback SSO permission cross team collaboration what that actually allows you to do is to do that security review once and then build applications many so we're here to
[13:05] accelerate our platform helps bring those insights into actual applications that are deployed at scale within minutes and as we're going to see with uh with my demo how we first seated some data we'll see how we can quick quickly
[13:18] create that application so With that, I'm going to switch over back to my laptop. Let's see how we did with the seated data and continue building from there.
[13:34] Okay. So, let's go to the application and where are we here? There we go. So, uh let me see how's that coming up. Maybe I'll zoom in one or two levels. So, we uh wanted to create the the seed data and we wanted to prefix the tables
[13:47] with SFO DBRX live. It went through the thinking process. It listed the tables. It created the grocery reporting tables. Verified it, right? And see the grocery tables and also related them through
[14:00] foreign keys. Good stuff. Let's actually go into Lakebase real quick. Validate that they're in there. And yes, there we go. So, we can see uh we can see the data. Looks like it had some uh sample
[14:12] grocery products that I put in, categories, unit price, uh sales, also uh some store uh and some utilization uh records as well. So from here, we're ready to actually build the application.
[14:26] So grocery reporting, I think that was one of the the use cases there
[14:47] So what we'll do is
[15:00] decisions, maybe even edit some data after doing an audit, right? Let's how about let's do that. So make an application for a grocery district manager to go into this data, have a dashboard, make some changes to
[15:13] the data after doing an audit. So what I'm looking for is a dashboard with KPIs, maybe a form to edit some of this data and submit it. Let's do that. All right, we got that typed in. Let's
[15:25] go ahead and get that started. So we're going to come back again after around 4 or 5 minutes and see what it comes up with. So, as you can see, if you were to extend this out, uh you can get really really creative. Uh there's many ways to
[15:38] connect into our platform through MCP, through APIs, etc. So, let's let's come back to this after it gives a couple minutes uh to cook here. So, with that, I'll bring it back to Josie.
[15:51] Perfect. Thank you. So you Tom's been showing you how you can build that application demo but we also want to bring it back to what Wreckit found. So with us our our
[16:06] champion S mean he was able to say that you know they were able to deliver consu customer insights to innovations team 50 to 60% faster. But rather than hearing
[16:18] it from me why don't we hear it from Wreckit as well. So,
[16:33] Rekit is the brand house name that is behind some of the world leading brands like Lizo, Detl, Finish, Vanish, Durex just to name a few. When the AI evolution started, we feel we could really leverage this capability to help
[16:47] serve our consumers better, to help build stronger, more relevant brands. We decided to be very intentional, avoid the the scatter gun approach, the sea of proof of concepts, select an area, and
[17:00] go for a reinvention, an automation of workflows uh and business processes. We knew that we were going to have to build a lot of bespoke applications. We historically have done bespoke
[17:12] development, but we don't have a huge internal engineering team. So the ability to build at pace and enterprise uh grade applications without compromising on you know security. We
[17:25] were looking for strong partners who we can test, learn and scale at an enterprise level very quickly. Retool came along with the right functional tech fit for purpose but also the innovation mindset and the strong focus
[17:39] on customer success for us. It was just the right fit. Pace was a key criteria. The ability to build minimum viable product very quickly security. We are a 15 billion pound organization and to
[17:53] seamlessly integrate with our datab bricks data uh platform >> with marketing. We ran a very deep study to understand where AI can actually augment and support. We are now able to
[18:05] deliver the consumer insights to our innovations team 50 to 60% faster. We have hundreds of marketeteers across the world using the applications on the
[18:17] retool platform on a daily basis.
[18:30] marketeers to do what they enjoy most and are are good at and were hired for. We are uh developing the next set of AI applications in the R&D function. The
[18:42] objective is to roll that out to pretty much every single function in the organization. The way Rekit has scaled the solutions was only possible through partners like Retool. Retool has been
[18:54] the accelerator for us to drive not just deployment actually the adoption in the market. This is a partnership for innovation for value at scale and impact.
[19:28] some of the takeaways that we want to share in terms of uh of what they said. So you know I think as Simeine said it's like look at those high value impactful use cases identify those do the deep
[19:40] dive with the teams to understand how you can reinvent how you can reimagine the processes the applications that they're using. Embed insights into the workflows as well so that we can get the not just help operationalize the data
[19:54] but also understand the impact that that's having. Governance is critical for every organization. So the tie-in with Unity catalog work we're doing with data bricks around there making sure that everything is secure make sure
[20:06] everything is governed managed in the right way people are only accessing data that they're allowed to they've got permission for they're not exposing anything outside of that and then it's measuring that adoption. So how are
[20:18] people finding this impact? What is the the value that they're getting? How are they getting as they as Record said you know 40% of their time back? How are we providing the insights 50 to 60% quicker? How can you measure that impact
[20:30] on the organization? So before we uh before we move on, how are we looking Tom? What's uh what's the situation?
[20:42] >> We're still cooking. Let's do a quick little status update. Looks like it's actually now creating the dashboard page. And if we scroll to the top, we saw that it's like building a front end. Uh all the functions are works. So all the functions are written are written
[20:55] right now. So yeah, we're we're still cooking. I think we should probably have something in the next two minutes. >> Perfect. Okay.
[21:09] So, as I mentioned earlier, we're also a data bricks customer as well as a partner with them. And later on today, if you guys are interested, we have our chief product officer, Abasha Gupta, talking about how we've scaled on Lake Base as well. So in the southside uh
[21:22] level 3 room 301 he'll be talking about how we've scaled to a million databases on lake base. We're one of the largest uh customers with uh with uh data bricks in that space.
[21:37] Also uh if you'd like to come and and investigate a bit more what we're doing there's QR code up on here that you can scan that will take more details of retool. We have a booth here at uh in the expo at 212 in the uh zone 2. We're
[21:51] running a competition at the moment to get some customized Nikes. So, if you want to come along and uh uh and enter that, you'd be more than welcome. >> And I think we're we're pretty much cooking on the application now. >> So, yeah, let's switch over. What I'll
[22:04] do here in that original video, uh this should be just about done. Let's actually go to one of the actual completed applications. So this was the global brand and campaign command center that you just saw in the intro video itself. Now to actually double click on
[22:18] some of the unique items from a build perspective. Uh we see that this uh this actual uh brand and campaign command center uh the resources it's connected to is uh your standard data bricks also
[22:30] lakebase and in this case Salesforce. So from here you're actually able to unify data sources across the enterprise across your organization very quickly and even create joint functions to
[22:42] actually uh unify that data and context between it. So as a quick example you can actually let me now zoom out here. You can actually see the live queries. There's nothing proprietary. It's all
[22:55] react SQL everything underneath. We expose it all. And then from there, you can also uh see a diagram of how this actually connected together. So let's actually check in with our app right now. And it looks like it actually just
[23:08] finished. So let's take a quick look on what it built. So all exports check out. We got a dashboard with five KPI cards, total revenue sold, transaction stores, daily revenue, trend chart, tab,
[23:20] product, and store tables. So you know what? I like the look of this app. Why don't we go ahead and publish it? So I'm going to go ahead and publish this application. This can also be tied into your source control with GitHub. Uh so it looks like we have to review some
[23:32] functions because they are not yet approved. So that means that we as builders need to first validate all of these queries to make sure they're actually okay and you're at the end of
[23:44] the day touching production data. So after a quick review, we like we like the query itself. We're going to go ahead and approve it. And now we should be able to publish. Let's try one more time. Oh, there's
[23:56] even more functions to approve. Let's go. Let's go ahead. Let's approve this one as well. Looks like we'll have to approve all of them in this case. There we
[24:08] Okay, we are good. Let's select the published folder. After that, we'll get the link and then we'll try out the application. Now, it's publishing and while while it's actually publishing, let's actually
[24:20] quickly check it out. Uh, so we have our grocery district manager console. We got our dashboard. We have the KPIs, revenue, etc. We got some graphing charts over here. Revenue by region, by
[24:32] category. And then we also have some uh uh top stores by revenue. Very quickly was able to create this application with minimal context uh just based on the actual data itself. So we have uh you
[24:45] know our audit manager here, our data audit. We can do it by stores or by product. We see organic bananas, produce, fresh farm co. You know, I think uh with inflation a little bit these days, maybe they are $169 instead
[24:58] of just 69. So, let's go ahead and save those changes. And updated organic bananas, $169. So, let's actually see if it actually did that change. We'll go to the
[25:12] products themselves. Organic bananas, $169. So, you saw we are able to surface data and also write back to Lakebase very, very quickly. So, we have our our link. It's been now been published and now it's ready for distribution. So even
[25:26] within a presentation from idea to build to production all it took was 15 minutes. And this is something that you can now further extend to all of your data bricks resources and beyond to
[25:38] create a very unique rich application and get those application insights for your team or teams that you actually work with. >> So thanks Tom. Great. Uh I think the demo gods were with us. Yeah, definitely
[25:51] friendly today. >> Definitely friendly today. The uh the Wi-Fi connection held up. So hopefully that's given you an introduction into what Retool does as a business, how we help organizations with that last mile of connectivity between the data and the
[26:04] users, reimagining workflows, reimagining applications. You know, as Tom showed, it's relatively quick, relatively easy to build those those lightweight in applications that you can then industrialize and push out to
[26:16] production as well. So it's very much about how we help the business go quickly but with the safe secure governance that the IT organization need so that they know that live data or
[26:28] production data isn't being exposed to people that it shouldn't be or outside the organization. As I said we do have a booth here at 212 in the expo. More than happy to uh to talk to people there. But
[26:40] if there's any questions that people have live, we're happy to do some Q&A as well. So, I think Null's got the uh the microphone, but if uh if anybody has any questions, please feel free to ask.

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.