Self-Service Assortment and Space Analytics at Walmart Scale

Summary

• Retail optimization: Assortment and space analytics optimizes product selection and shelf allocation to boost sales, improve inventory management and enhance customer experience on Databricks Data and AI platform.
• Genie-powered self-service: Walmart's use of AI/BI Genie interface enables non-technical store managers and merchants to execute self-service analytics without SQL expertise.
• Quantified impact: Walmart achieved 90% reduction in time-to-value and saved $5.6M annually in FTE hours by implementing unified platform analytics on Databricks Data and AI platform.

Self-Service Assortment and Space Analytics at Walmart Scale

Talk by Alexandro Arreola-Garcia, Senior Manager, Assortment & Space , Walmart ; Nikit Shah, Sr. Delivery Solutions Architect, Databricks
Assortment and space analytics optimizes product selection and shelf allocation to boost sales, improve inventory management and enhance customer experience. However, challenges like evolving demand, data accuracy and operational alignment hinder success. Older approaches struggled due to siloed tools, slow performance and poor governance. Databricks unified platform resolved these issues, enabling seamless data integration, high-performance analytics and governed sharing. The innovative AI/BI Genie interface empowered self-service analytics, driving non-technical user adoption. This solution helped Walmart cut time to value by 90% and saved $5.6M annually in FTE hours leading to increased productivity. Looking ahead, AI agents will let store managers and merchants execute decisions via conversational interfaces, streamlining operations and enhancing accessibility. This transformation positions retailers to thrive in a competitive, customer-centric market.
Here's more to explore:
The best data warehouse is a lakehouse: https://www.databricks.com/product/databricks-sql
How to migrate legacy data warehouses: https://www.databricks.com/resources/ebook/migrate-your-legacy-data-warehouse-databricks
See all the product announcements from Data + AI Summit: https://www.databricks.com/events/dataaisummit-2025-announcements

Chapters

• 0:00 Introduction and agenda
• 1:03 Defining assortment analytics
• 1:49 Challenges in retail analytics
• 6:27 Technical solution with Databricks
• 11:39 Genie interface demo
• 18:04 Business value and outcomes
• 19:23 Future with AI agents

FAQs

What is assortment and space analytics?

Assortment and space analytics optimizes product selection and shelf allocation to boost sales, improve inventory management, and enhance customer experience by selecting from thousands of items to match customer demand.

How did Walmart use the Databricks Data and AI platform?

Walmart leveraged the unified Databricks platform with seamless data integration, high-performance analytics, and governed sharing to replace siloed tools and enable data-driven decisions across the retail operation.

What AI/BI features empowered users?

The AI/BI Genie interface enabled non-technical store managers and merchants to perform self-service analytics without traditional BI complexity, driving adoption among business users.

What measurable results did Walmart achieve?

This video demonstrates that Walmart achieved a 90% reduction in time-to-value and saved $5.6M annually in FTE hours through faster analytics and better operational efficiency.

Full transcript

0:00 Introduction and agenda

All right, good morning everyone. Hope you had a good time at the keynote. So we had some really exciting announcements this year. Uh let's get started with this session. Uh just before that some forward-looking statements. Uh all of this what we're going to present is more forformational purposes. uh should it it's not a guarantee of future performance but we would love to work with you and your customer teams uh and your other teams uh to uh replicate these type of results for your businesses.
Uh as always uh provide us feedback. We would love to hear from you and how we can improve and what all you would like to hear uh next year. All right. So the topic for this uh today's session is self-service assortment and space analytics uh at Walmart scale. Um so our agenda is going to be speakers unfortunately it's just me today. uh assortment what is assortment space analytics what are the existing challenges how we solved them using data bricks uh we'll get into a demo and the business value that Walmart reaped out of this and future uh with AI agents that Walmart is thinking about um so a little bit about me I'm uh a senior delivery solutions architect at data bricks been with data bricks for around three years uh I am based out of New York new jersey area and uh my colleague who unfortunately fell sick two days back.

1:03 Defining assortment analytics

He's a senior manager of the assortment and space analytics team at Walmart uh who we collaborated this with. So he was not able to make it today but he sent us the demo. So we'll hear uh the demo in his voice. All right. So let's start with what is assortment and space analytics. uh and to help everyone understand some of you might probably be aware of this space uh but for some most of us who are not some of us who are not aware of this space uh or are not deep into this let me start with like I want everyone of you to imagine you are like a store manager of your local Walmart store you can imagine uh that store right now you've been there multiple times you generally know where what items are placed now you have about 140,000 items items give or take that you have to choose from to place at the right shelf at the right time so that it meets the customer demand.

1:49 Challenges in retail analytics

Customers find what they need when they enter your store. Second, that in turn helps maximize your sales and third, it helps with improved inventory management. Now to choose from those 140,000 items you could use one way could be to use your judgment could be based on your past experience based on your local understanding of the demographics and the demand place the right things at the right shelves but the right right way to go about it would be to analyze all the data and then based of that take decisions on which product to place at which shelf.
Um so essentially I mean uh essentially like analyzing millions of records from your store sales, your store inventory, uh your pick rates, the rate at which your staff is able to get stuff from your store warehouse into the shelves, modular plan of those uh shelves, the width of those shelves that is available, uh shipments, returns, all of that will give you the right answer in terms of how to place those 140,000 items. which by the way will also change with the time of the year, latest trends and all those uh kind of changes and seasonal factors.
Um so lucky for you Walmart essentially knows that there is a lot on you as a store manager. You have to manage customers, your staff and all those things. Uh there is this team which is part of the assort and space analytics team which analyzes all those millions of records and gives you the recommendations. uh well they don't directly give it to you as a store manager but they give it to one layer above you called as merchants merchants think of them as brand managers uh so there are all these brand managers for Hershey chocolates or oral be brushes and so on their aim is to maximize the sales for those products and based on them you can then in turn choose for your store how do you want to place them so that's pretty much like assortment space analytics is analysis of the vast product selection that's available uh that you can uh allocate in the space that is available in that shelves of your store so that it leads to enhanced customer experience, maximize sales based on the customer demand and inventory management.
So let's understand a little bit deeper now why uh this can be hard. Well, first like we just uh kind of saw it requires a continuous analysis of customer demand. Um just to give you some numbers, Walmart has like 255 million customers visiting its 10,750 stores uh and across its numerous e-commerce websites. So one sizefits-all solution is not going to work. U they their supply chain like global supply chain moves like 100 billion items uh each year.
So again overstocking or stockouts uh need to be avoided and for that you need to really know which stores in which region need what kind of products. Second uh data accuracy and completeness all that analysis that uh team does uh needs to be kind of based on trustworthy data because you are going to make decisions based off of that. Any wrong, incomplete, inaccurate data can lead to millions of dollars of losses for your store and Walmart.
And lastly, right product at the right shelf at the right time. It's not enough to just analyze all this customer demand, but it's also important to know which items can fit in that shelf because you have some limited shelf there and there are different sizes. So, it needs to be right uh right product, right place at the right time. So now again peeling the onion layer a little bit more let's get into the data side of the things um what are the technical challenges and how we solved using data bricks so the challenges that team faced um with the existing and older solutions one of them was siloed products so they had all these uh technical tools like tech uh tools and platforms to do ETL a separate tool tool to do BI, separate tool to do ML, separate tool to do like web apps and data governance and so on.

6:27 Technical solution with Databricks

So it was like dealing with multiple architects and teams of these different disparate systems. Uh all of which that had their own opinions by the way and it was just really a lot of friction to use all of these together to put together this solution which requires massive like tens of TBS of data to be processed every uh almost every day or so. uh so it was essentially leading to higher total cost of ownership of using these multiple tools uh to ultimately meet the business SLAs.
Second issue was slow performance uh for these large workloads. Uh like I said tens of TBs of data like we are talking about 90 billion I think in inventory and 35 billion uh store sales uh records that need to be processed. um these older solutions had a lot of performance issues, frequent job failures and also because there were all these multiple disparate systems uh team is like a business team they didn't have a large tech team so they had to rely on the central IT team to do most of these stuff so that led to a lot of further delays in collaborating with the central IT team and their tech developers and so on and lastly data sharing u Alandro's team needs to kind of work with multiple sister teams within Walmart to be able to provide these recommendations.
Uh so it is important that they securely share and receive data internally. Uh so there there needs to be a central governance layer to do that. So with all these challenges they they approached uh data bricks and and let's see how data bricks essentially solved these issues. So the biggest advantage that uh uh the team got essentially was because of the unified layer that data bricks brings. It's one tool, one platform, one ecosystem to do your ETL, your BI, uh your ML data governance through Unity catalog, delta sharing and even now web apps is something they're looking into like data bricks apps that you just saw in the keynote. uh once they got access to this platform uh they sort of became self-sufficient because it was all one tool they needed one architect instead of now relying on multiple teams and central IT team uh their tech teams uh were able to develop these pipelines and workloads because there's this assistant again which gossy talked about it generates all these SQL queries so they were able to create all these pipelines using uh delta live tables which is now called as lakeflow delt uh declarative pipelines through this to ultim ultimately land data into the raw layer uh cleanse it transform it and ultimately the gold layer. um performance-wise because data bricks has uh photon and it can h handle all this massive volumes of data easily uh they were able to meet these performance SLAs's uh and then finally delta sharing because uh delta sharing is controlled by through unity catalog it is one sec uh central governance layer it easily allows you to share receive data and then finally the gamechanging thing was Genie which we're going to get into the demo today genie essentially allows you to have like a uh interface for business users.
You don't have to even know SQL. You can do this ad hoc analysis using uh just plain English statements. So we'll get in the demo of that. So just to summarize, you have all this store data sets uh store salesmentary model plan all gets into this raw uh bronze layer we call it and then from there it's cleansed transformed. there were uh multiple product ids that need to needed to be standardized and those things and ultimately aggregated and then into the Genie layer and they're using apps through shiny today but they are now le they're exploring data bricks apps uh to see if they can essentially use the integrations that data bricks has built in.
Cool. So just before we get in the demo just a quick brief preview on what Genie is. uh some of you may already have used it uh and some of you may have already seen in today's uh keynote but Genie essentially provides this uh uh it's called as data room or genie space you can add your data sets from your indie catalog uh it can you can then train it uh provide these context for the role your jar business jarens which we're going to see today uh clarifications that users might ask and so on and so it really acts like an AI agent or AI assistant sorry And then you can add some of the sample queries so for it to learn and it all is governance through Unity catalog.
So it helps with um just overall uh unified like security. So, let's look at the demo. And this is where you'll see voice. And let me start here. Walking through a detailed user interaction, illustrating how Genie significantly streamlines and enriches the process of assortment analytics. Traditionally, merchants and analysts navigated comprehensive dashboards, selecting multiple hierarchical levels to precisely filter their item sets.

11:39 Genie interface demo

Although this manual method was effective and reliable, Genie provides an even simpler and more intuitive experience, Genie leverages sophistic sophisticated metadata capabilities to automatically interpret and filter the item hierarchy. Rather than manually traversing each hierarchical level, users simply message Genie their hierarchy. Genie immediately identifies the appropriate hierarchy level, streamlining analytics and enabling quicker, more intuitive data access.
Let's illustrate this by posing a common mo merchant query. What are my top five performing items in golf balls? Genie evaluates the hierarchy contextually, automatically identifying and presenting the top five items based on performance. Significantly, Genie's understanding of performance is adaptive and contextsensitive, dynamically selecting the most relevant metric depending on a specific category or area under analysis, ensuring insights remain precise and meaningful.
Further, let's effortlessly enhance this analysis by incorporating additional metrics. For example, we can easily request Genie to include metrics like AUR or average unit retail and growth percent. Genie can interpret acron acronyms like AUR and updates the analysis to include these new metrics providing deeper insights into item performance. The tables are also customizable so you can format them to your liking. Moving forward, let's explore Genie's visualization capabilities.
I'll ask Genie to show performance versus currently modularized stores. Genie generates a clear inform in informative plot instantly. This visualization offers richer insights by displaying displaying item performance relative to store exposure quickly highlighting strategic opportunities or pinpointing areas require requiring attention. The plots are also customizable so you can customize them to your liking. Genie also excels at enhancing productivity by allowing users to save common analyses as reusable instructions.
These instructions can be labeled with specific descriptive keywords enabling easy and rapid future retrieval. significantly improving consistency and efficiency across analytics efforts. Finally, Genie integrates powerful feedback mechanisms to support continuous improvement. Users can directly and provide feedback regarding analytics accuracy, usability, or any other concerns. Administrators then monitor this feedback closely, responding promptly to user suggestions or required corrections, ensuring ongoing enhancement of analytics quality and usability.
Cool. So that was essentially uh what we heard explaining. So just to kind of summarize like give you like a two-minut uh summary of this entire thing this is how I mean he obviously they cannot show their entire systems uh so they sent a screenshot of what merchants and uh had in terms of user interface they had to drop down select uh they know which item to filter and how to get to this specific item they need and they would get a data set which they would then have to maybe download in Excel do some analysis.
So it was like really kind of just difficult to maneuver from this. Uh getting genie interface really just kind of changed the game by uh them being able to ask these questions in uh just plain English and it will generate all these analysis charts and graphs which they can then download and send it to other teams. Um and what you also saw was uh it could also just understand uh some of the acronyms. So you see there average AUR average unit retail essentially that means uh total sales upon number of items sold uh it just understands because has been trained on this.
So all of this essentially helps kind of make it easy for you know business users to not uh worry about these complex SQL statements. uh they can just use this genie to uh to perform analysis and um kind of make their job simpler, focus more on business results. Um cool. So let's switch back to the demo. So next we will get into uh hold while this comes up. we look into essentially what was the business value that uh Walmart reaped out of this.
Um so like I said this was like a unified tool to do uh ETL, BI, governance, ML, all of this together. They they suddenly kind of realized how one platform was able to do get the job done and not rely on these multiple teams uh to do that one to get to pull together to get into one project. uh which essentially helped them save like tons of hours. So that's where it was like 90% faster time to value because of the unified ecosystem that data bricks provides and then the actual business value of using this data in the business.

18:04 Business value and outcomes

Uh Walmart achieved last year around 7% adoption rate of this uh continuous assortment planning models which have generated additional $630 million for Walmart. uh and by this year the target is to reach additional 25%. So they they're just going to keep on increasing the adoption rate for this. Um this is a quote from team uh one of the senior managers there. Essentially the main advantage they thought was uh being able to do all of the things that we just saw at scale without needing to ask multiple teams.
And uh finally what they see the future is with AI agents. So very similar to something what we saw in keynote today. Um they have something developed called as Wall-E which is their agentic framework. Uh which is going to we are going we are now currently working to use Genie APIs to be integrated into these W-ally agents. Uh so merchants can directly have um this type of conversational u genie like interface within this agents but those agents can do much many other things as well.

19:23 Future with AI agents

Um so that's something later this year we are going to work with them. Uh and so with that thank you very much. Um if you have any questions I can take some right now or uh down the stage as well. I'll be around here. Thank you.

Learn more about the Databricks Data and AI platform.