Unifying Retail Data for Smarter Planning and Store Execution

Summary

This video demonstrates a unified retail application built on the Databricks Data and AI platform that serves multiple personas — executives, marketers, category planners, and store associates — with a single source of truth for sales, inventory, supply chain, media spend, and store operations data.
The application showcases granular demand and sales forecasting with what-if planning scenarios, audience builder and campaign activation workflows, and in-store task assignment — all connected through Genie and autonomous agent orchestration.
A closed-loop ROI measurement capability tracks campaign performance from audience creation through in-store execution, with a multi-agent supervisor coordinating Genie agents to automate retail workflows from insight to action end-to-end.

Unifying Retail Data for Smarter Planning and Store Execution

Retail and consumer goods teams often work across fragmented signals from sales, inventory, supply chain, media spend, and store operations. This demo, presented by Pavi Singh, Senior Solutions Architect for Retail Consumer Goods at Databricks, showcases a unified application built on three foundations: data unification to create a single source of truth, governance to control data access, and intelligence to support recommendations and automated decisions.
You’ll see how multiple personas like executives, marketers, planners, and store associates land on a unified application interface powered by Databricks to create granular demand/sales forecasts, model planning decisions, build audiences, launch campaigns, assign in-store tasks, and measure campaign performance. The demo also shows how Genie and autonomous agents can orchestrate these workflows end-to-end.
What you’ll learn from watching:
→ How to review sales, inventory, traffic, and delivery KPIs in one application
→ How unified data and governed access support cross-functional retail decisions
→ How to compare forecasts with actuals and test planning decisions using what-if scenarios
→ How to build audiences, activate campaigns, and track store-level execution
→ How Genie and autonomous agents connect retail workflows from insight to execution
Learn more by watching our virtual Retail Forum: https://events.databricks.com/FY27Q2-WB-VirtRetailForum

Chapters

FAQs

What three pillars underpin the unified retail application?

The application is built on three foundations: data unification to create a single source of truth by bringing together point-of-sale, loyalty, supply chain, and media spend data; governance to apply fine-grained access controls so different roles see only the data they are authorized to access; and intelligence to deliver AI-powered recommendations and autonomous workflows through Genie and multi-agent orchestration.

How does the application support demand planning and forecasting?

The platform compares actual sales against forecast data to surface variance by category and subcategory, and supports what-if planning scenarios so planners can model the impact of different inventory and pricing decisions before committing to them. This shifts planning from reactive analysis to proactive, data-driven decision-making.

How does the audience builder and campaign activation workflow work?

Marketers can create audience segments using the unified dataset, generate campaign briefs, and activate those campaigns — all within the same Databricks application. The video shows a closed-loop measurement capability that tracks store-level campaign performance and calculates ROI from audience creation through in-store execution.

What role do Genie and multi-agent orchestration play in this solution?

Genie enables natural language queries across retail data for any user persona, while a multi-agent supervisor coordinates autonomous Genie agents to execute complex retail workflows end-to-end with minimal manual intervention. This architecture allows the system to move from insight generation to automated action — such as assigning in-store tasks or launching campaigns — in a single orchestrated workflow.

Full transcript

[00:04] All in all, I'm Paddy Singh, part of the UK and Ireland retail team here at Databricks. And I'm here today to showcase how can retail and consumer goods organizations can streamline
[00:17] um their sales insights, demand planning, audience building, in-store operations, as well as measuring the ROI of the entire loop through a single
[00:29] unified application within the Databricks platform. Before we jump into understanding the application demo, let's understand what the three pillars
[00:42] on which this capability is built on. The first pillar being data unification. We believe having a singular version of truth, aka reducing silos, is the most
[00:54] important thing for an organization. Obviously, here the assumption is your point of sale, loyalty, supply chain, media spend, amongst other data, is all flowing through the lakehouse platform.
[01:08] The second pillar is governance, making sure you have the ability to one, not only own your data and set your rules, the foundational and compliance aspects
[01:22] of the data are check-boxed by giving you fine-grained access controls, allowing you to control and make sure that different types of data
[01:36] can only be seen by the relevant people and parties needed. The third bit is the intelligence in the era of AI and GenAI that we are in,
[01:50] having the ability to get recommendations and move beyond reactive analytics to proactive is really key. Databricks
[02:02] embeds AI across every layer from natural language query with our Genie capability to agentic workflows that can get orchestrated with one click.
[02:15] Allowing you to automate decisions end-to-end. And that is definitely a big plus point for any business. Let's jump into the demo and start off
[02:28] with view for personas in marketing, your CMO, your marketing director, category planner. On a Monday morning, you would want to assess your overall KPIs across sales,
[02:43] inventory, transactional count, traffic, uh how much are we delivering on time, so on and so forth, monthly return on sales, sales by category, and you can drill down into different categories,
[02:55] subcategories, understand what the top movers are, what the underperformers are. For executives, you might want to create an executive brief. So, I'll click If I
[03:07] go back up, the executive brief has been created. You can see total sales was 2.4 million as you can see on obviously the dashboard. Um food storage leading all categories as
[03:22] the top category. Uh and four out of five categories are growing month on month, which is a fantastic thing. If I jump into the detail breakdown section, you can obviously assess
[03:34] categories and subcategories, the KPIs associated with those. Um jump into a map view where you can click on one and understand the KPIs by store level.
[03:47] Um the stock that's available in hand, the units that have been sold, the revenue that has been generated, uh as well as have uh an ability to assess what the
[04:00] top performing stores up for the demand planners and merchandisers within your organization. As you've identified a particular category, you would want to assess how
[04:12] are we tracking to the demand that we'd laid out at the start of the season. You can see here for this particular footwear category, we are slightly off track by 5 and 1/2% compared to last
[04:25] month. And the actuals are trending below forecast by 30%. You have the ability to view forecast versus actuals rolling window
[04:37] across 4 months. You can also drill down into different categories and assess forecast and actuals, as well as going into specific SKUs that are identifying what the
[04:51] um KPIs across that particular SKU is as it loads in a few seconds. I'll go back and let that load for a second. Jumping into advanced analytics, now you
[05:05] want to move beyond understanding what my forecast and actuals were and are, but I also want to drill down into what the daily divergence rates are for that particular category or at an overall
[05:18] level. I can understand what the uh volatility across different days were, what was influencing that what volatility as well. What is my gap across different days
[05:31] within a particular month or a season. I can jump into optimizing and it gives me recommended uh units that I should
[05:43] reorder to make sure that I'm meeting the demand of the particular SKU or category or subcategory across a specific store and that's all outlined
[05:55] here. The good thing I like about what-if scenarios is a simulator that based on me as a merchandiser or planner, if I play around with certain levers, aka
[06:08] discounts, if I give let's say 15% discount um I have 72% replenishment rate within that particular store
[06:20] and I have let's say 76% audience activation reach I'm expected to hit a projected uplift in units by 43% and a revenue impact of close to 30% amongst other KPIs that I
[06:34] can observe. Gives me a good segue now to launch the audience visit. As you can see here, the audience visit essentially is aimed towards our CRM
[06:46] marketing uh teams based on signals that we've seen in demand and sales um and much more. It starts recommending
[06:58] what could be the campaigns that are likely to be most successful in moving the needle. At the top here, we're seeing obviously 4th of July um conscious that 4th of July just went um
[07:12] but for the purpose of demo we'll stick to 4th of July as a campaign but you can see other signals recommending certain other campaigns. You've got food storage recovery um as well as a custom campaign that you
[07:25] might create and multitude of other signals that are there. I'll click on launch campaign and you can see here the audience number is specified. What is the average ROI that that campaign might generate
[07:39] um the window that we want to run this particular audience activation. I'll click on launch campaign wizard. Now, what this essentially does is a six-step process to essentially
[07:53] describe your intent, clarify obviously some KPIs that you want. It recommends an audience, then you can specify a spend, um add a bit of creative bits, and
[08:06] ultimately activate and send that particular campaign to the different stores that you want that campaign to run. This campaign wizard is built on top of our customer lake
[08:18] platform, which essentially lets us define the audience, configure how we want these digital or in-store campaigns to run, and then activate across multiple channels, which is stores, it
[08:31] could be e-com website, it could be online media. So, let's jump into it. I've got the campaign brief here, which has already been preset. I can obviously edit, modify, change on it, but because we've
[08:43] given a very decent campaign brief for the 4th of July, I'll click on pass intent. Fairly certain no clarification would be needed in this particular scenario.
[08:55] Give it a few seconds. And here you go. It starts recommending me the top six segments that I could choose from. You've got deal-driven refill hunters, you've got holiday stock up loyalists, you've got bulk
[09:07] multi-category um folks uh that are interested in home cleaning. Bear in mind, this is again the food recovery uh scenario. Um I can choose whichever one I want.
[09:22] Let's say we stick to these three, and I click on continue. You have the ability here to specify the budget that you want. In this case, the default is set to 250, but I have the
[09:35] ability to scale that up to let's say 1 million.
[09:50] Um and obviously the impact that is expected to generate is already getting pre-calculated. Here you go, the campaign is activated. Now once the campaign gets activated, you then segue into what is going to
[10:02] happen within the stores. Uh as a store manager or a store associate, I think the the most likely view would be viewing things on an iPad. Uh as let's say
[10:14] an in-store execution tab. Uh as a store manager, where you can log in into your particular store, you can identify um and see what restock units needs to
[10:28] be um delivered across different SKUs. I can assign it to particular store associates within that particular store. If I drill down, uh you can see here yeah, you've got
[10:41] actions like updating price tags for certain SKUs, assigning it to specific individuals, so on and so forth. You've got people within the team that you have. Click on promos, you can see the promos
[10:55] applicable to this store based on obviously the campaign that we just launched. Click on that. You can see for that particular campaign the different stores um that were recommended. So the store
[11:09] obviously was recommended, which is why it's showing there. Um and you have the ability to ask via Genie. If I change uh my persona to store
[11:22] associate and go back to my task view, it should now give me the ability to be obviously on the receiving side as a store manager assigned certain task. So, let's say in
[11:35] this scenario as Beatrice, I have the ability ability to get a checklist of the work that I need to do. And I have the ability to once I've done it,
[11:47] keep a track of the things that I've done. Um and it's also giving me certain recommendations around the promotions or replenishments on what if I do that, how much it might
[12:01] generate. I think lastly uh is once we've executed that in store, going back to the marketing director,
[12:13] category planner, or a CMO, I need to have the ability to understand all the campaigns that I'm running and drilling to specific categories on how they're
[12:26] performing. Um I'll jump into the food storage recovery category. And as you can see here, the KPIs associated with
[12:38] what that particular campaign were cuz it was run previously. Um the velocity of recovery that it caused, the um ROAS by tactic KPIs,
[12:52] certain insights across the featured store, um as well as obviously understanding and getting a view on um the different metrics.
[13:06] Now, as we've now seen the closed loop of measuring a particular campaign that we ran and how much effect that it might cause on sales,
[13:19] irrespective of all this being within one application, you still have to go by persona into different um aspects and do things. The multi-agent Genie gives you the
[13:33] ability to orchestrate different agents to do that for you. Now, if I simply ask it plan a full storage category recovery end-to-end
[13:45] and in the interest of time, just go to the one that I ran earlier, what it gives me is a snapshot of sales, margin, and stock rate, certain critical issues that it's
[13:58] identified. Starts giving me portfolio actions based on the recommendations behind the scenes of certain SKUs and what you should be doing.
[14:11] Similarly, inventory and supply chain, replenishment, operational excellence, and a plethora of things that it's pretty much automated through every single step of the way by orchestrating
[14:26] different agents specific to your sales insights, your demand planning, your audience building, your in-store operational management, as well as measurement. Now, um
[14:39] for people from a technological background, you might wonder how is that playing around. So, let's jump into a very high-level view of what the architecture looks like in this
[14:52] scenario. This architecture slide pretty much showcases the different components within Databricks that have been used to build this application, um including Genie, including the
[15:05] supervisor agent. You can see the data is powering every single Genie agent. And how the multi-agent supervisor is pretty much stitching across all five. And hopefully, that
[15:18] gives you um a food for thought to think of what your organization could do. Uh if you've got any further questions, uh feel free to reach out to
[15:31] your account teams or myself. And um I look forward to collaborating um with you all in the near future. 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.