Sponsored Tiger Analytics | Data Potential to Business Value Scaling Commercial and Consumer Product

Summary

Tiger Analytics and PepsiCo's Petr Gazaryan describe the Enterprise Data Foundation (EDF) as a shared Databricks platform underpinning a global portfolio of commercial and consumer data products—spanning revenue management, store effectiveness, media mix modeling, and personalization—deployed across 25+ markets.
This video explores the engineering decisions required to scale data products at global CPG scale, including managing geospatial and hierarchical data complexity, standardizing gold layers across 30 markets, and building reusable pipeline architectures for route-to-market expansion projects.
PepsiCo operates across more than 200 countries and territories with 289 billion servings per year, providing the scale backdrop for a strategic framework that distinguishes between data potential (what a platform enables) and business value (what data products built on it deliver).

Chapters

FAQs

What is PepsiCo's Enterprise Data Foundation and what does it power?

The Enterprise Data Foundation (EDF) is PepsiCo's shared Databricks platform that provides enterprise-wide data access as the base layer for commercial and consumer data products. As described in this video, the EDF underpins programs including revenue management, store effectiveness, media mix modeling, and consumer personalization across more than 25 global markets.

How does Tiger Analytics help scale data products across global markets?

Tiger Analytics works with PepsiCo as a data and AI services partner to design, build, and operate the data product portfolio on top of the EDF. This video focuses on Tiger's approach to reusable pipeline architectures, gold layer standardization across markets, and managing the geospatial and hierarchical complexity that arises when deploying the same data product logic across countries with different data shapes.

What engineering challenges arise when standardizing data products at global CPG scale?

This video highlights several challenges: geospatial and hierarchical data varies significantly by country (different store hierarchies, distribution channels, and regional classifications), gold layer schemas must be standardized across 30 markets without losing local nuance, and new markets require adapting existing pipelines rather than rebuilding from scratch while maintaining consistency with the shared platform.

What is the difference between data potential and business value in this context?

This video frames data platforms as creating potential—the infrastructure and access layer that makes data available—while data products create actual business value by packaging that data into something decision-makers can act on. PepsiCo's experience illustrates that platform investment alone is not enough; the strategic design and operational sustainability of the data products built on top determine whether potential becomes impact.

Full transcript

[00:08] So good evening ladies and gentlemen. Uh first of all thank you so much for coming out here on a on a evening especially on the tail end of the day. I'm sure people are ready to jump to happy hours. Uh and I hope you all are having exciting day and a great time at
[00:21] DS 2026. Myself Paraka I'm a part of Tiger Analytics. I'm a client partner here. I focus on our retail and CPG business. With me, we have Peter Gazarian. He is
[00:33] senior data senior director at PepsiCo. He leads data engineering for consumer and commercial applications.
[00:49] So before we jump into the data products uh talk about PepsiCo story, I would just spend like a minute talking about Tiger Analytics. First of all, who all who all know about Tiger here? We get some hands raised. Okay, very few. Great. Uh, Tiger
[01:03] Analytics, we are a 15year-old company. We are an end-to-end data and AI services provider. Uh, we are about 6200 professionals focusing on delivering data and AI services day
[01:15] in and out. As you see, we have recently been recognized as the RCT partner of the year award by data
[01:27] bricks. So data bricks basically thinks that we are the best in the industry to provide data and AI solutions specifically for retail, CPG, travel and hospitality industry. What I say is that
[01:40] I think they are right about 50% of our revenue comes from retail and CPG business and out of that uh if you see the number here 90% of our customers uh that work
[01:53] with us or the revenue that comes to us is from the repeat business. What it means is the customers who engage with us once they don't want to leave us. They work with us for a very very long time. Uh the other fact uh why we are
[02:06] able to actually deliver great services to our customers is is primarily because we have a great talent. We are we are a place who is able to attract great talent who is also able to make sure that we provide them enough
[02:19] opportunities and challenges to keep going on and that resonates in our attrition rate which is lowest in the industry. Right. uh with the data bricks award uh I think we are set uh to focus
[02:31] more and more on on RCD business and and sorry with that I'll jump to the next slide uh talk a little bit about our tiger and PepsiCo partnership
[02:45] I'm sure you all heard about Pepsico today at the the keynote uh fall calling some some call outs from Magesh Pepsico operates their business in more than 200 countries and territories uh they have the largest fleet in North America but
[02:58] how do they manage that kind of scale right and it's very complex business as well and that is where we are proud to partner with with PepsiCo in their transformation journey which started 6 years ago we are not just doing some
[03:11] point programs and solutions we have partnered with them on building their data foundation which is actually the industry's largest data foundation that is built on data bricks techstack uh we
[03:23] are partnered with them on building AI platform platforms, computer vision platforms, very cutting edge solutions. On top of that, these platforms do not deliver value, right? What delivers value is a data products which are sitting on top of that. And we are proud
[03:36] to partner with with PepsiCo for a for last 6 years delivering value across the value chain. May it be plan, make, move, sell. We focus on delivering value whether it is on the financial side of
[03:48] the house, sustainability, business planning, supply chain, consumer, commercial, name it and we are there. And with that I'll hand over to Peter who's going to focus a little bit on consumer and commercial applications. We
[04:01] have a great platform but how do we unleash value out of it? Let's hear from Peter about that. A round of applause ladies and gentlemen for Peter. >> Thank you. Thank you Parak. Um so as per mentioned my name is Peter Gazerian. I
[04:14] lead the um commercial and consumer data engineering organization within Peps. So um my plan today was to walk you through briefly though to some of the
[04:27] decisions we've made and some of the principles we try to follow to um scale our data engineering workflow workflows in the context of Pepsica.
[04:41] So I'm sure many of you have been listening the keynotes today and um our SVP of DNA Magesh Pavati was presenting u you know was was speaking uh briefly
[04:54] in the keynotes um my guess my guesses was mentioning uh some of the numbers I wrote them down uh because they go nicely with my you know the rest of my talk. So what one of the numbers which
[05:06] caught my own attention was 1.3 billion occasions a day which our consumers and customers do uh with the with the products across the globe which are
[05:18] produced by Pepsica. So that that just this one number shows the shoe scale of of the operations we are dealing with. Um he was also stressing um on the focus
[05:32] we make on being faster for our um consumers and um customers. Uh so Pepski indeed is a massive uh
[05:44] massive organization. Uh so Mesh was also mentioning EDF briefly. EDF is the enterprise data foundation
[05:56] basically in the in terms of the datab bricks medallion architecture. EDF covers for bronze and silver layers. It serves as the um single source of truth
[06:08] for the business focused programs such as commercial and consumer programs. But there are other programs which cover various aspects of the PepsiCo complex business such as uh supply chain uh
[06:20] business planning, financial um reporting a business etc etc. Some of them parag mentioning as well. So and EDF is indeed a massive
[06:34] undertaking right. So some of the numbers I can present here and it just shows the you know this the
[06:46] size of the data EDF operates with some of the um aspects of the data quality and the data coverage which EDF provides. Uh so but basically EDF is a platform
[06:59] all by itself. doesn't enable the business value but it it really uh actually it doesn't realize the business value but it enables the programs such as commercial consumer programs to realize this business value and bring
[07:12] this value to to uh to our um um you know stakeholders inside that and outside of of of outside of company and so we we consider internally we consider
[07:24] the the business programs are the gold layer
[07:42] um the question let's see what's the next yeah the the question which comes to the data engineers data engineering teams are not only the data engineering teams data science product management data
[07:54] governance you name it uh the the question is very uh actually simple so how do We unlock the value for the business at scale at that scale.
[08:08] uh and this scale as you may imagine can come in various flavors right so for some of the programs comes in the flavor of let's say data size they they manage for some of the programs this flavor is
[08:20] near realtime processing uh for some of them some specific um regulatory or security considerations for us for commercial consumer again I'm
[08:32] going back to this number of 1.3 billion and this is not just 1.3 billion of the same interactions which just come with this stream of you know
[08:46] whatever signals these are very diverse set of interactions and basically that is kind of the first layer of complexity we are dealing with um you may think of
[08:59] you know PepsiCo operates in about 200 markets and territories so all of them are very very different uh you may think of let's say USA versus Mexico where USA is a prevalence of the
[09:11] big box retail you know organized trade Mexico is a lot of mom and pop shops traditional trade and so on just this just one of the aspects we're dealing it with but there are many more right so on
[09:25] top of that inside the Pepski we have various personas various business users who look at the same thing differently good example I really like is uh geography
[09:37] Right. So you can chunk the the map of the let's say United States or any other countries to the pieces or to the dots very very differently depending
[09:51] of what your business unit is is trying to accomplish. Marketing you know covers it with let's say tiles of DMAs. Retailers are covering with the dots of their own taxonomy and stores and uh I
[10:05] can continue it right. So the the supply chain operates in the structure of geographical ge geographies um centered around the data warehouses and the routes and so I can can continue this
[10:18] but again this this shows some you know deeper level of of complexity and then even deeper is the data itself. Yes, as I mentioned and as
[10:30] Mesh mentioned other leads in Pepsica, DF covers maybe at this point 85 90% of the of our data needs but there is always a long tail and especially in the convention consumer space where we
[10:42] always on board the new data sources usually external we deal especially in the smaller markets we deal with the data which comes not from the systems but rather manually just because of the
[10:56] uh the nature of the business process. Yes, it's 2026, but it happens all the time. Um
[11:10] um so yes, so and that essentially that goes to a question of uh when the business comes to us with with the new ask or they ask to deploy the
[11:23] existing product to the new market. The question which we are trying to answer is how do we do it faster? Meaning how do how do we build how what what set of
[11:36] decisions we have to make what architectures which we you know we have to have to build the new products faster to the level of let's say to write as less amount of code as we can to
[11:50] generate as less amount of artifacts in GitHub or or in in in the delta delta lake as we can and um um for this uh We basically let me go to
[12:04] the next to the next slide. Basically I called it strategic directions not the decisions because again it's a very big and diverse business and it's really hard to set something in stone and go
[12:17] with it. So I called it the directions. Those are very simple but fundamentally they they help us to frame the conversation with our stakeholders and partners in the right way. Right? So for
[12:30] instance, we when we say we need to build something, we want to build the product, the data product. As much cliche it sounds as it is, it's still true. We
[12:43] don't want to deliver 5K or or 500 or thousand new KPIs. We want to deliver the product at scale which would produce these KPIs or the the data for the um
[12:58] downstream consumers. let's say machine learning or data science or or or AI workloads which basically are directed in the same way right so and with this
[13:10] we so we we built the the gold layers you know being go uh um product oriented we strive to standardize as many things as we can from the level of the
[13:23] individual notebooks all the all the way uh to the um to the orchestration to the access to the data to the image catalog and um uh you know access provided to
[13:36] the agentic workloads and and such. So another one is when we make a decision of deploying or building something we always strive to build the system which is configurable rather than customizable
[13:51] and that is very important because again it it helps us to write as less code as we can. Of course, there are always um um you
[14:03] know some some individual uh markets which have very specific use cases which we know would not ever repeat on other markets. So we still have some room to customization
[14:17] but in general we have a direction to you know we go towards configuration versus customization. Uh so all of it right now is of course
[14:30] uh backed by the um AI embedded uh no I need I need to go back a embedded um at every stage of delivery. Uh so AI is you
[14:42] know helps us to achieve this goal of course. Um so this slide is just illustration how do we think on um a automate the delivery life cycle. We do
[14:55] a lot of work in this space have a lot of of of good results over there. So um one project I wanted to go briefly we have just three minutes is uh store DNA.
[15:12] So the store DNA is basically the data product.
[15:24] so bas basically this is the program which creates a datadriven representation of the physical space on the level of the individual store on the
[15:36] various markets. So think about the use cases of rolling out the new assortment let's say what are those sets of stores in conjunction with our understanding of the audience and potential consumers.
[15:49] you need to roll this product out first or the effectiveness of the individual stores or the collection of the stores. Um so the the use case or or another
[16:03] good use case is wide space and the um the those points of interest which are not necessarily the stores which are around the stores which kind of help us again to understand the profiling of the
[16:15] stores u you know understand the the consumers living around this and who who who can reach this this individual store right so this program was originated I
[16:27] don't know how many years ago um as a very small local product somewhere in Europe but at some point when it started to bring the results the the decision was made to scale it up to
[16:40] multiple markets I think at this point it's about 30 markets so you you you hear this number 30 many times today because you know essentially that you know the scale for us is is um you know
[16:53] spinning around this number of you know dimensions we scale up to right So, so this this product going back to this principles I was mentioning was completely redesigned. It worked
[17:05] especially was small and now it's big. Um, so at the heart of this codebase is the framework built on the datab bricks bundles. Uh, data ingestion mostly
[17:18] happens you know the bulk of the big big bulk of it happens from ADF. Uh so if if I go to the gold layer again inside of it I can say there's there is a harmonization layer which basically
[17:30] ingest if it needs to ingest the data harmonizes it according to the business use case then there is a transformation layer where all the data is being merged and um put into the uh star schema which
[17:43] is very um you know which is which is the same for all the markets right so and then the delivery uh layer uh um helps to propagate the data to downstream
[17:55] consumers. uh it done it is done through the UC um uh with all the um you know access management uh going and and um I mean data management going with this
[18:07] right so I can go more about the SDNA but essentially that product um implements those decisions and direct
[18:19] directions we've been trying to follow and we are trying to follow right so um that basically that helps us to deliver that many right so again I
[18:33] mentioned 30 programs right now right now I think we have about 25 26 maybe 27 active programs but of course we have some programs which we don't develop anymore they're just being refreshed on
[18:45] the continuous basis uh we have thousands of pipelines um have 500 about that plus data sources
[18:57] integrated into our various goal layers and automated uh we have uh platform you know and our approach helps us to maintain it 24/7 um and uh I think that's it I think I
[19:12] think we we wanted to say together thank you so much

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.