Supply Chain Command Center: Databricks and Sigma Self-Service Analytics at Scale

Summary

Unilever's North American supply chain teams were manually stitching together disconnected data sources in Excel and legacy reports some more than 20 years old, so the customer analytics team partnered with supply chain to build a unified command center on Databricks and Sigma.
The solution integrates orders, demand projections, forecast accuracy, and store-level inventory data into a single source of truth using data modeling, custom SQL functions, and input tables, shifting teams from reactive firefighting to proactive forecasting.
The platform serves four North American business units across beauty, personal care, home care, and foods, with a roadmap targeting AI-driven risk identification to flag supply chain issues before they materialize.

Supply Chain Command Center: Databricks and Sigma Self-Service Analytics at Scale

Unilever's supply chain teams spent hours each day juggling disconnected data sources, manual Excel stitching, and misaligned metrics between sales and operations. Legacy BI tools and spreadsheets couldn't provide the real-time, unified view needed to make fast inventory decisions across retailers and distribution centers. The month-end close process was painful, taking days and draining team capacity.
Discover how Unilever unified disparate data on Databricks using Sigma's self-service analytics platform, creating a command center that integrates orders, projections, forecast accuracy, and store-level data into a single source of truth. Learn how data modeling, custom SQL functions, and input tables enabled supply chain teams to shift from reactive firefighting to proactive forecasting and root cause analysis. See the business impact: faster decisions, accurate projections, and the roadmap to AI-driven risk identification.
🤝

Chapters

FAQs

Why did Unilever need to modernize its supply chain analytics?

Unilever's supply chain teams were relying on legacy reports some more than 20 years old and manually stitching data from multiple disconnected sources in Excel, which could not support real-time decision-making at retail scale across 185 countries. The process created misaligned metrics between sales and operations, and the month-end close was painful and time-consuming for all teams involved.

How does the Unilever supply chain command center work?

The command center uses Databricks as the data layer and Sigma as the self-service analytics interface, integrating orders, demand projections, forecast accuracy metrics, and store-level inventory data into a unified view. Custom SQL functions and Sigma's input table feature allow supply chain teams to slice data by any dimension and annotate projections directly without waiting for analyst support.

What is Sigma's role in Unilever's Databricks analytics stack?

Sigma serves as the self-service analytics layer on top of Databricks, providing a spreadsheet-like interface that supply chain teams can use without writing SQL. Its input table feature lets users add planning annotations and adjustments directly in the interface, transforming the command center from a read-only report into an interactive forecasting and root-cause-analysis tool.

What business impact did Unilever achieve and what comes next?

The platform shifted Unilever's supply chain teams from reactive firefighting to proactive forecasting, enabling faster and more accurate inventory decisions across the four North American business units. The next phase of the roadmap targets AI-driven risk identification to automatically surface supply chain risks before they escalate, building on the single source of truth now established in Databricks.

Full transcript

[00:07] Hi everyone. We're going to get started here. My name is uh Phil Nelson and I am the customer analytics manager uh at Unilver and I'm excited today to talk to you guys about how Unilver is optimizing their supply chain inventory with Sigma and data bricks. Just kind of go through the agenda
[00:22] quick. Uh I am just going to touch briefly on who we are as a company in case no one's familiar with who we are as Unilver. um why our inventory analytics had to evolve and ultimately why we decided to replatform this on data bricks and sigma and what that unlocked for us and ultimately where we
[00:38] want to go in the future with it. So, Unilver we are a very large CPG company. We operate approximately 50 billion in turnover uh across 185 countries and here in North America we
[00:54] make up about 12 billion in turnover roughly about 24% of that number for the entire company. And that turnover is driven there we go that turnover is driven uh by our 30 power brands. a lot of these that you guys would recognize on the
[01:10] screen here like Dove, Tresomeme, Helman's, Nor, um, all of that. And these 30 power brands make up about 70% 78% of that turnover. And as you can see, we play across a lot of different categories. And so we have a kind of a unique operating model where we're split
[01:26] up into uh multiple different business units and hubs. And for North America, uh we're actually split into the four main ones that you see there across beauty and well-being, personal care, home care, and foods. Uh the customer analytics team here engages across all
[01:42] of them. Uh and our supply chain team, while they are kind of siloed up into each one, what we started to notice is that they were all having kind of the same questions to answer, right? And so that leads us into why did our inventory analytics have to evolve? Uh on customer
[01:59] analytics, we primarily always just focused on sales and category. And about a year ago, that shifted. We decided that we wanted to engage with a supply chain team and really see if we could help them like kind of evolve their analytics into the, you know, into a new platform. Um, as we started to engage
[02:17] with this team, this question, the answer to this question became abundantly clear and it's that the legacy tools that they were using are just no longer enough. Right? Like operating at the retail scale that we operate at, it just demands a lot more than these legacy tools can offer,
[02:34] right? Some sometimes some of these reports were 20 25 years old and they were just constantly updated, maintained, somebody refreshed it, somebody added a new feature and then you just start to get sprawl, right? So one Excel report might turn into one Excel report with 20 tabs in it, right?
[02:50] But honestly, as we started to look into this more and more, it wasn't even really about scale. Like it wasn't even about scaling. it it turned into bit the basics of like just having the supply chain team answer basic operating questions sometimes required them to
[03:06] stitch data together from a lot of different systems that were just never designed to talk to each other and so that ultimately led us to the next conclusion which is that these legacy tools equals legacy limitations right so if your data has to be pulled manually from multiple different sources
[03:22] to be fed into an Excel file that could come from multiple different places and then the supply chain team What we were seeing is they had so many disasperate data sources that there was really no shared real-time view of their inventory. Sometimes they had to pull data out of a data lake that was
[03:37] connected to PowerBI. Sometimes it was through their retailer operating systems. Sometimes it was internal systems, right? But none of them were connected. And the other thing is since this was so much manualist stitching, what we were also noticing is that numbers like our plan, forecast,
[03:53] projections, all of that, it just wasn't flowing through. So unless someone was updating it daily, we weren't getting a read on it. But the biggest disconnect was actually this last point, which is backward looking versus reactive analysis. And what we were noticing is
[04:10] that it was what the sales team was looking at versus what the supply chain team was reporting. And what that kind of came down to was a hierarchy mismatch. So those business units that I mentioned earlier, the sales teams aligned to those. That's what they're constantly looking at. That's the number that they're marching
[04:26] towards. Where our supply chain team, I think I said that wrong. Sales team is looking anyways supply chain team uh because they work so closely with the retailer is always looking at external hierarchies. And so trying to mash those two together is extremely difficult,
[04:41] right? And so that just meant that there's actually kind of a funny uh story about this. I talked with a supply chain uh lead recently and he said um a lot of times towards the month end close we would just get into a a meeting room and then we would argue about whose numbers were the most correct. Like it
[04:57] would be our do we use the sales team's numbers or do we use our numbers and they would have a meeting to align until they could figure out whose numbers to use. So that way they could figure out where their gap to projection was or whatever and move forward. Um, so this just meant that our month-end close
[05:12] became a very grueling process, right? It just took up an inordinate amount of time for our supply chain team and it impacted things like our employees abilities to take time off. Uh they'd have to work late. They might scramble like all day. They might drive like dive down a rabbit hole for an entire day,
[05:29] come back and still not have an answer on why we have a volume gap related question, right? And so that's where data bricks and sigma came in. We wanted to utilize sigma um after testing out a pilot last year and determining that it was a
[05:44] platform that we thought that we could actually like use to answer some of these questions. And so the first thing we did is we unified the data together on data bricks and sigma. And luckily we started with a very strong foundation. Um I'm very blessed that I work with a really good data engineering team. Shout
[06:01] out to Kelsey Botchi for managing that team. She's amazing. Um, but luckily we already had a very strong foundation in data bricks. And so since we already had one, we were able to leverage existing data tables and data bricks where all that data was available. Where it
[06:16] wasn't, we were able to use web bots to stop having to do manual pulls out of the retailer system and landed out on data lakes or even bring them up into a delta table. Uh, and then there was a couple like, this is kind of a weird nuance, but we also had to set up some daily job runs. Um, just because there
[06:32] was like some files that were being landed by a different team that wasn't connected anywhere, we needed to pull it into our environment so that way we could read it in. And in Sigma, we were able to utilize their data modeling and we were able to build out data models very similar to how you'd build out a
[06:48] fact table inside of PowerBI. So fact tables, dimension tables, all that stuff. We could unify that across in different data models. Um, one thing that we were able to do with our last point about bringing in just specific files is we could write a custom SQL function to actually bring it into Sigma
[07:06] read straight off of our data lake as a CSV file and land it in a data model. So that way we capture that data and have it available to be able to connect to other data sources, right? And excuse me, the other thing is that Sigma makes it very easy for us to aggregate data at
[07:22] different levels. And so we were able to align and match all these hierarchies from external to internal very quickly, right? And so all of this just meant that we had a good foundation to start with, but really kind of the last step that we noticed was that sometimes
[07:38] there's special buy items. There's items that um you know they're a special like they might be a seasonal promotional item, a special case pack, something like that. They're not in our retailer system, but we still need to account for the volume somehow, right? And so that's where Sigma came in where they have
[07:54] input tables that allow us to be able to like let the end user go and select the uh hierarchy level that those items need to be attributed to so they can flow through into the system, right? And if you've ever worked in PowerBI, that last
[08:09] point that I just mentioned isn't really possible. Like PowerBI doesn't have a way for the end user to input data, right? The only way that I know how to do it is you have to put a Excel file out on a SharePoint. Um, have that picked up by a logic app processed overnight, all that stuff. And while we
[08:25] could go that route, it just wasn't as like lifetime and responsive as we wanted it to be. And so, Sigma was uh kind of the way that like unified all that together. So, saying all that, being able to utilize both of these platforms together,
[08:42] uh, to us felt like I can't get my thing to go to for us felt like a superpowered team up, right? Um, by the way, I was asked recently uh why there's not a picture of me on this, and it's because if anyone knows where this is from, the only other person I
[08:58] could put on here is Robin, and I didn't really feel like being Robin in this slide. Um, don't worry. I'm sure one of my co-workers will photoshop me into it later. So now what this unlocked for us was our
[09:14] real-time decision-m and we started with that one report. We started with the the biggest report in there which was our daily order sales tracking and that was the one that everyone was using, right? That helped with our month end close and everything like that. But from there it expanded
[09:30] out into approximately 10 15 different reports. And ultimately where we ended up at was with our command center. Right? This is our supply chain analytics application. This has a daily view of risk and opportunity. There is
[09:46] real-time data in here. Orders versus projections. Uh gaps to targets and early signals. Everyone uses this now. and it has our daily order sales tracking, store sell for out, store sellout forecasting, forecast accuracy reporting, everything. I've had to
[10:02] update this slide about five or six times since I started building this presentation earlier this year because that's the speed that we've been moving with and how many different reports we've been adding to it. So, this is their one-stop shop. They come in here, they get highle views right away. um
[10:17] they can see early reads on stuff and we included links to external applications on this as well just so that way it's this is it this is where they go right and ultimately the shift here because we launched this a couple months ago and the shift that we saw with this was
[10:34] extraordinary right like our team went from firefighting to forecasting right this these new capabilities now and what it's unlocked uh has just kind of brided the gap between what sales and supply chain team really should be, which is a partner. And our old methodology and
[10:52] kind of how we were doing stuff in the past was just kind of antiquated. It didn't really uh like take into account back loing and frontloading. It didn't have like forecasting projections, different stuff like that. So, when we were looking at how we were going to close the month out, that was kind of being ignored. And so, it was really
[11:07] hard for us to get a read on it. Our new method though, since it's using forecasting and projection models inside of there, um this is now a lot more accurate and it's explainable, right? We also added things in like projection bias and um like dispatch rates,
[11:24] different things like that to help explain things. All of this just means that our supply chain team is no longer reactive, but they're proactive in finding incremental order opportunities.
[11:39] Yeah. So, if there's one thing I want you to take away from this entire presentation, it's this. Is that we didn't just build that first report. We built that first report and we align the business around one version of the truth.
[11:56] There we go. And when we look at like what this unlocks for us as a team, all of these capabilities and the impact that they have on us just means that we can make faster in the- moment decisions. We now have one single source of truth that we're all rallied around. And that's probably the biggest unlock
[12:12] that we had there. And we can get down to our root cause analysis much more accurate. And we're also being very proactive in how we're planning things out, right?
[12:27] And when we look at what's next here, like for our future use cases, like I said, I've had to update that that command center slide five, six times now. So, some of the stuff we've already started doing, but we're already starting to scale this into our sales team. We're starting to scale this into category management team. Um, we're expanding the scope. We want to do
[12:43] scenario modeling, promotional planning. We're adding in additional supply chain optimizations every day. Um, but our long-term goal is to move from the the left column to the right column to go from our BI reporting and we want to move from insight to action and enable
[12:58] AI. That's the next step that we haven't done yet. And what we're looking to do is just be able to identify risks automatically. um it just gets emailed out to the end user so that way they have a starting point to go with, you know, model these scenarios in real time and ultimately
[13:14] just like lead to where we're all going to, which is to have extremely accurate data and drive these decisions a lot faster. Again, guys, thank you so much for your time.

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.