Sudip Barat
Director of Information Delivery, Belk

Rupesh Pradhan
COO / Head of Insights, Planning and Analytics, Logic

Cindy Rogers
Director of Insights, Planning and Optimization, Logic

Director of Information Delivery at Belk

We discuss enterprise data strategy and lessons learned as a result of Belk’s Data Warehouse modernization and Enterprise Reporting & Analytics Modernization Program. Sudip recently led a major technology modernization for Belk’s Data Warehouse on the Snowflake Cloud Platform (Snowflake recently made news with blockbuster success as the largest software IPO ever.)

Sudip is the Director of Information Delivery at Belk; he’s an experienced and inspiring data and analytics leader with an extensive background in leading analytics modernization initiatives in the retail and manufacturing sectors. Previously, Sudip has led analytics, master data management, and integration initiatives at The Men’s Wearhouse and TechnipFMC, a global oil & gas manufacturer.

Belk is a private department store based in Charlotte, NC with 292 stores in 16 states. Founded in 1888, the brand is known for “Modern Southern Style.” In 2015 the company went private and there is a strong focus on digital transformation as the omnichannel business grows.


Listen to our podcast interview with Sudip


Belk’s overall driving vision is to reimagine the department store and omnichannel plays an important role in achieving that goal. With that, Belk’s customer-facing offerings continue to evolve while delivering on their respected brand. Sudip, how this focus factor into your enterprise reporting analytics and modernization program?

Sudip Barat: Cindy again, thank you and Logic for inviting me to be apart of this podcast. Going back to the question you asked, omnichannel always, if you look at the last decade or so, has been a priority for retailers, and everybody is in a different place in their maturity curve. For Belk specifically, what was important was to see, “what is the path they need to go through to modernize themselves?” There is no real template in which every retailer can just take and ensure success, it doesn’t work that way.

For Belk, the few drivers that were priority: number one, Belk went private in 2015, so with that, some of the business priorities changed, and also how Belk was organized. Another aspect is that eventually the technology landscape also changed. So Belk has gone through a journey of transformation between 2015 and now, that’s half a decade of our journey. Omnichannel has always been a top goal, but I will say it is a constantly-evolving goal. When I came onboard to Belk around 2018, I think most of the business changes were kind of settling down. We had a clear vision in terms of how we can transform ourselves into a modern department store with better assortments and much more visibility online, across all the channels for our site, right?

But what was important for us to recognize was how to bridge the gap between what the business was really looking for and, what the foundational day-to-day tactical problems that the organization was facing due to technology limitations. We had to build that bridge between what’s needed to set up a foundation from a technology team perspective on-top of which omnichannel channel vision strategic vision can be built out.

During 2018 and 2019, from my perspective for Belk, the primary focus was, essentially, to get that alignment and build that phase one of the initiative around modernization, where the foundation was there for us to do to the next generation or evolution of business enhancement around omnichannel.

Cindy Rogers: Very good, thank you, Sudip. As the offerings continue to evolve, so does the expectation to keep pace with your customers and the requirements of your customers. With that, stores and roles within the stores have changed as well, including an added complexity within the supply chain. Those were all key factors in the Reporting, Analytics, and Modernization Program. Rupesh, would you want to add anything to that as being very close to this program?

Rupesh Pradhan: Yes, I think that people who are not practitioners in this area sometimes don’t realize that even though the retail industry has been talking about omnichannel commerce and sometimes unified commerce for quite a few years now, truly integrating traditional brick-and-mortar and eCommerce channels into an integrated and seamless business is still a work in progress.

If you look under the hood, especially those of us who understand systems, understand business processes and all that, you will see that there are lots of improvisations, duct tape, bandaids, or whatever you want to call it, that are sort of working together to improvise if you will, an omnichannel business experience, and really, the complexity that has been thrown into this mix because of omnichannel business is not trivial.

We have seen that there are a lot of different roles that weren’t traditionally interacting within a retail business that now have to work very close together, you know, stores have become a supply chain node, traditional marketing and customer service help desk, and now we need to look at them together to understand a holistic customer experience that customers have with retailers.

Just a few years ago, if you think about it, you could purchase an item at a store or maybe online, but today’s omnichannel journey can involve browsing online, purchasing in-store, getting part of that order shipped directly to your house, and then maybe even returning somewhere else. From an analytics and data architecture perspective, how do you model these complex activities or complex transaction models in a way that you can have an end-to-end view of what’s truly happening, right?

You need something like a unified data architecture that reflects the overall business model that retailers are trying to invent in this omnichannel world. This reimagining of how retail happens and how do you make sure that you have an underlying data architecture that can actually provide you insight and analytics to be able to see how you’re doing is not trivial. But Belk took this challenge.

Sudip Barat: Rupesh, you articulated this concept very well, and I’ll go back to one comment that you made which was around this band-aid approach. This is a rather clunky approach under the hood, right? That is exactly what ends up happening a lot of the time and I experienced it firsthand here at Belk. We are very tactically focused at Belk, every quarter or every promotion, and the teams are working in silos a majority of the time but what’s happening is that a lot of retailers who are probably the same size as Belk are not necessarily always working towards a coordinated effort, right?

Creating that Grand Omnichannel Vision, a common understanding of what, for example, is a demand margin or selling margin, is. Not having a common understanding of metrics across cross-functional groups acts as an impediment, so when you’re working in silos, everybody eventually ends up creating a nuanced, separate version of the truth. A lot of the initiative or the journey we went through was bringing cross-collaborative teams together.

We had to create a common understanding of data, metrics, and everything, because the word “Omni” is about not just how a customer is able to shop cross-channel, but it’s also within Belk, the internal stakeholders; how they have a common understanding of customers across all channels. Whether how marketing is defining success or whether we’re able to convert a customer, compared to how finance is thinking about profitability, we must build a common understanding.

Cindy Rogers: Sudip and Rupesh thank you for that background and context here, and candidly, this sounds like tackling a hairball when you think about it with the different data sources you discuss and the variances or the fragmented metrics that we are working with. Kudos Sudip, to you and your leadership team for taking this on and knowing that this approach scares a lot of retailers, but every retailer will reach a burning platform to say “we have to take action”.

Having reached that burning platform, what were the major factors for you and the Belk leadership team to say, “This is going to be complex, but we have to move forward?”

Sudip Barat: Very good question and to answer that I will revert to what we discussed earlier. At times businesses are so tactically focused on delivering near-term value, so a lot of the time getting the leadership team all aligned for a strategic program becomes difficult. In this case, as part of our journey, the leadership team did recognize that data had to be a strategic focus of the organization. For us, that journey also had certain tactical drivers and that kind of accelerated the process.

The tactical driver for us was that we used to be on a Teradata data warehouse. And even on a week-by-week basis, as the organization was delivering standard weekly reports to get certain analytics and to get specific initiatives going, we were also onboarding data scientists on the same platform. In this onboarding process, we started seeing the struggle for resources on weekly reports and data scientists doing their own work, which a lot of times is very iterative, exploratory, and a very difficult to define workload.

All this combined to create a situation where we felt that we were not making good progress either on the reporting or on the data science side. We were reaching a point where we had to renew our platform because of an end-of-life situation. The key driver at that point was, “Okay, we have a tactical issue to address; we need a better, stable platform and let’s make a conscious, deliberate strategic move,” and we did not want to be shortsighted. Can we choose a platform which is going to address the near-term issues quickly, but also can address our strategic needs? That burning platform situation was Teradata and we went through a decision-making process, and after a POC, we selected Snowflake.

Cindy Rogers: Between the hardware, the performance issues, and the upcoming data science team needing data quickly, Belk needed to be nimble and move forward. Thank you, Sudip.

As mentioned, you selected Snowflake for your Cloud Data Warehouse, which is a relative newcomer and has quickly become a leader in the market. Can you talk a little bit about that decision, and how Snowflake as a platform has been working for you?

Sudip Barat: Yes, absolutely. Snowflake has really been the key to our success at Belk from a data foundation perspective. Snowflake is actually going public tomorrow as we speak so it will be interesting to see the market’s reaction to this platform. When we originally looked at Snowflake, this was around September or October of 2018. We evaluated a few platforms and Snowflake was the obvious choice because of the recommendations we received from industry experts.

We looked into other platforms from Google, ENC, and whatnot, but eventually, a few things made us move towards Snowflake. One being for Belk, it was important to have a SQL-compliant platform, a platform that allows us to speak the common language of data that is still predominant in the market, which is structured-query language. From a compliance perspective, Snowflake was extremely close even at that point in 2018 to Oracle, Teradata, or Netezza.

The second thing was, as I mentioned before, we needed a tactical workstream and also strategic workstream to modernize the data landscape as a whole. But for the tactical, we couldn’t take the risk of rearchitecting everything, so a lift-and-shift approach from Teradata was important. Based on our proof-of-concept, we realized Snowflake does very well; you are talking about a minimal amount of code changes, and moving the whole Teradata code base to Snowflake at an acceptable cost.

Cindy Rogers: Very good. Now Rupesh, Logic has experience with Snowflake for retailers. Do you have anything to add to that?

Rupesh Pradhan: Yes, as Sudip mentioned, Snowflake is a relative newcomer in this space, but they are also really out there from an innovation perspective when it comes to a data warehouse in the cloud. With the emergence and the rapid expansion of cloud platforms, having looked at the variety of different offerings out there, from our perspective, Snowflake has been a very stable, reliable, and mature platform compared to some of the others.

Really, the power of a cloud-based database platform, like Snowflake, is in its ability to scale as much or as little as you need on a day-to-day basis. This is something retailers appreciate as there are weekends when the demand and the activity, it’s very different than other parts of the week, or during holiday seasons, or on Black Friday. The demands of the systems are very different than other times and other parts of the year so what a cloud-based platform has allowed retailers to do is really scale-up and scale-down as your systems are demanding these resources. This has actually been a revolutionary thing when it comes to analytics.

Analytics is all about processing large volumes of data and these days the volume, the variety, and the velocity of data have increased quite a bit, so being able to scale up-and-down to be able to analyze the amount of complex data that is generated these days in the retail sector is complicated, and cloud-based technologies and cloud-based database platforms have really made that possible.

We discussed the cloud platform and the velocity and the variety of data. With that, the next step Belk decided to move forward with Robling. A retail-specific enterprise data model, Robling that addresses all the functional areas that a retailer needs today and going forward. Sudip, can you give us some insight into why you chose Robling?

Sudip Barat: Absolutely Cindy. One of the key drivers for going with the modernization route was not to just to do the technical upgrade, but also to build this whole omnichannel vision of the business on top of it. At Belk, we did not think that we needed to reinvent the wheel if there were already best-practices available. Belk already had our retail data model in-house. Based on Oracle’s model, over decades we had customized it to the point where it became unique to Belk.

We had the choice to keep building up those additional omnichannel capabilities around customer dimension, or the whole supply chain aspect. Or look into another available option, based on best-practices, more of a product, which we buy and customize for Belk. We looked at a few and based on our discussion and analysis, we felt that the Robling model was best for us.

I would say the second reason is, even though Robling is very new in this space as an organization or as a start-up, it’s backed up by two solid leaders who have been in the retail space for quite some time. All that knowledge and understanding of retail was already baked into the model when we reviewed it. But the other key thing, which was a selling point for Belk: Robling had made an effort to build a model already on Snowflake and vetted it for compatibility, performance, etc. This gave us a level of comfort, considering that strategically, we were aligning on the technical foundation of where the model is going to live and breathe, right?

Rupesh Pradhan: On that note, I wanted to add one more thing if I may. As Sudip and Cindy were saying, there is the data model part which sort of encompasses a lot of learnings that Robling has had over the years of being in the industry. There is also another part to it, that was also something that we at Logic, as a systems integrator, found very valuable, and it is around cloud technology and cloud-based database platform.

There are lots of other moving parts involved, there is a data security component of it, there’s this part of “how do you make one cloud talk to another cloud?” What happens when you have some of your systems on-premise and some of these systems on the cloud? And Robling had sort of thought through and figured it out in terms of, “how do you move data and how do you integrate systems that may be in different clouds or on-prem systems, and bring that data together to a unified data platform for analytics.” Snowflake allowed us to not have to worry about those things or need to try to reengineer or think about every little thing on our own when we’re doing implementations and integration of the system which was huge.

Sudip Barat: Rupesh you are absolutely right. Just to clarify, when I say data model it is not just a set of logical models with schema, we are talking about the whole data warehouse and the reporting layer as well. An integral part of it is how you build these layers within the data warehouse like the staging or the reporting layer, so Robling has all this incorporated. What is also important to understand is that people who are not familiar with Robling is that their primary offering is really a data-model-as-a-service, you do not necessarily need to have your own Snowflake. If you are looking for a data model, Robling provides a house and offers it as a service.

Belk’s model was more of a hybrid approach where Belk’s Snowflake is where the model was deployed so definitely from a time-to-market perspective, it is not just that data model, which is a set of tables essentially, but this whole pipeline of data, how that’s going to function, how your security will be applied on top of it, how you do error handling, how do you do data reprocessing?

All of these are very key when the deliverable is business metrics which is the accuracy of the data. Accuracy of data is extremely critical in any data project and I will say Robling has pretty much got all those aspects covered from an end-to-end offering perspective.

Cindy Rogers: As you both are aware, as new data sources surface and are introduced to your teams every day, it sounds like is Robling does provide that blueprint or that framework so that the teams can quickly gain access, or it could be integrated and quickly gain access to those new data sources in a logical manner.

Sudip Barat: Absolutely Cindy, and going back to one of your earlier questions, what was Belk’s driver to get Robling? One of the things we wanted to consciously do is, as heavy users of Manhattan Suite around order management, warehouse management, and on the ERP side, we are Oracle RMS users. The closer we could stay to those core source data model, the better it would be for us. Every company must diverge a little bit and with Robling having an in-built connector to those data models of those products, it simplified our time-to-market as well. We also tried to stay close to the actual product as much as possible but there will always be gaps.

Earlier you spoke about it saying, “we had a lot of metrics and we had, maybe different definitions or calculation for the metrics.” One area that Belk placed a lot of energy on was data governance, and looking at where they are at today and how we are going to provide governance air coverage going forward. What level of importance did this have for you, and how did it factor into the decision-making process throughout the program?

Sudip Barat: It is extremely important, without governance, what we would end up having is what we already had. Two years down the road, we will again have ungoverned, multiple-versions of the truth residing in a cloud, instead of being on-premise. That is what would have happened so having that governance and leveraging it is extremely critical.

As I mentioned earlier, omnichannel vision is not about giving a customer a seamless experience around delivering their product. It is also about a seamless experience around how each group within Belk collaborates around data and metrics. Everybody must agree on a common, shared definition of these metrics, right? So, what we have really been focusing on this year is essentially around two aspects. One is standardizing the metrics, so that yeah, every single analyst is not creating their own set of metrics and the other part of it is governing those metrics through a concept of Center of Analytics.

After completing this journey, we have reduced ad hoc metrics by 40% which the business users were building just to serve their own reporting needs. Overall, we have reduced overall metrics by 25%. What that means is now we have more metrics at the enterprise level and fewer metrics at the departmental silos. So people are talking about more metrics in a cross-collaboration environment which is a good thing. The Center of Analytics was originally set up with the key SMEs across each business unit. Our goal was any change of metrics or any introduction of new metrics was to be driven by these business data stewards or SMEs, only if they all agree. Over time it has become part of our DNA at Belk.

We do not necessarily have a formal process of creating new metrics or going through an approval process, but it is part of the overall way we are cross-collaborating now. The moment we start talking about a new metric, we kind of instinctively panic, asking ourselves, “why do we need a new metric? We need to probably bring this group together, the set of users to see, do we really need it?” And that is a behavior change, right? I think that’s it’s not necessarily a tangible outcome or deliverable of the initiative, but it’s something we have evolved, and kind of ingrained our nature now, and I think that’s a big, big, positive.

Cindy Rogers: Such a huge win, a win that I think would help every retailer. I understand that the Monday morning business recaps, rather than just sparring over the performance metric and how it was derived, you are now more aligned on that performance metric and you’re more focused on the action planning than on the data. Is that fair to say?

Sudip Barat: Absolutely.

Cindy Rogers: Another thing around data governance and the alignment of it was how seamless it is now for the organization. Your Vice-President of HR made a wonderful point. “This isn’t just a reporting and analytics modernization program, it’s also seen as an associate engagement program” because instead of the heavy-lifting on a Sunday or early Monday morning to prep and gather performance insights for these business recaps, there is much more ease behind it now.

Sudip Barat: Absolutely and as you are aware, any time there is an initiative that needs a certain amount of resources to be allocated, there are always these ROI calculations we end up doing. What question this points to is “did the thing that you just talked about, the productivity or efficiency to deliver the value?” It is hard to measure that ROI but that is the biggest piece of value to come out of this program.

Rupesh Pradhan: One more thing I wanted to add on the whole topic of data governance. We hear in the industry these days, many retailers, not just retailers, other organizations too, want to be a data-driven organization. We hear a lot about data analysis, data science, and all these tools that are out there these days to allow you to do data visualization. Through all of this, the challenge for an organization to be a data-driven organization is really data. The problem with data analytics or the challenges with data analytics generally is the availability of data.

People don’t necessarily think about it, but if you have “data” where it’s all jumbled up, everything is there, it’s not harmonized, it is not curated, and it is not governed, then it’s actually very difficult to make use of. Most of the time, data analysis, data science, analytics, data-driven organizations, they suffer, or they, fail at what they’re trying to do really because they don’t have access to well-designed, curated data. That discipline to make sure that you have data governance, so that the data that you have available in your organization is actually consumable by various groups within your organization, various data scientists, data analysts, it is very important.

Sudip Barat: Totally, you are probably familiar with this phrase, “in data science, 80% of the time is spent on data preparation, and 20% in actual value creation. This is primarily because data scientists do struggle to really rationalize and curate any form of data set, which then they can use for their algorithms, right? The key thing is the availability aspect which we have addressed as part of the program, creating a data lake where the source data models are available as is, without really putting any additional business logic behind it. Now, our internal team is really able to find how the data is in the source systems but more importantly, it allows us to find the gaps in the business processes or the data uniformity or issues with data quality. These are the things we’ve learned that have allowed us to spend less time as we are launching a data science engagement since are already aware of upfront issues so we try to either address those separately or to try to work around them so that the data scientists don’t get stuck.

Cindy Rogers: Sudip, this was such a large-scale initiative that included a data migration to the cloud and also included taking all these data sets and those relationships and making it consumable to a variety of different roles within the organization so they could make business decisions based on their responsibilities. It was also about the rationalization of what metrics are truly going to report on our performance. A huge, successful initiative here that Belk’s leadership team took on.

What are the lessons learned for other retail leaders who are considering making a similar move?

Sudip Barat: From my perspective, now that we have the luxury to look back and evaluate and to reflect on what went well, I think that always thinking strategically was a huge part of our success. Even with a burning platform, as mentioned, we had a lot of tactical heat we had to deal with but we never lost our strategic vision or sight of the end goal and what it was going to look like three years down the road. Also, any tactical decision around data must tie to that broader scheme of things, that’s number one.

Secondly, as part of the IT data organization, we always worked with the business as they are the true owners of the vision around data and what business that data can drive. Our business leaders owned aspects of the subject area or aspects of metrics or aspects of functions so the accountability piece was very clear throughout the whole process. The accountability piece is key because otherwise, we would end up with a lot of confusion around the data. Going back to Rupesh was saying this whole governance area becomes very murky which so making the key business stakeholders apart of the process early on is extremely important.

The other thing I will say that is a tad bit tactical, but at the end of the day if the deliverable of the initiative is a set of business dashboards for executives to consume, care and thought has to taken by the architects of the program that those high-level dashboards or KPIs map all the way down to the granular-level metrics or data points which the day-to-day operations analyst is really using to report up.

For us, we felt that we probably could have done a better job when defining these high-level dashboards and the reports and metrics and KPIs, and how it ties to the lowest-level of a report from an ad-hoc metrics and reporting perspective. That plumbing must be clearly understood and designed upfront to avoid a lot of operational issues. Those would be my recommendations that working with business leaders is important but also do not forget about the lower level. The operational analysts or operational workers who have a day job. I mean, they are much closer to the data than an executive is, executives are looking at a weekly view of the business or a quarterly view of the business. But the guys who are working with the data on a day-to-day basis, it’s important to get their feedback and make sure there are no design flaws in the solution that we are building.

Cindy Rogers: Rupesh, you were close to this program, as an executive partner from the outside looking in and even as a part of the core team, do you have any takeaways from your perspective?

Rupesh Pradhan: I actually want to second what Sudip just said, which is, having that overall vision of what you’re trying to achieve in let’s say one year, two years, three years, that’s great, but having the discipline and the drive to make sure that this system also works for those who are trying to work on a day-to-day basis operationally. Having that discipline to make sure that this entire architecture and this entire infrastructure that you are putting in works for them is very important. Otherwise, you are going to have the same problem of where different people are looking at different sets of data, different versions of the truth, as Sudip says, and you’re always going to struggle with it in the future.

Just having that discipline while you’re pursuing your long term vision, trying to make sure that on a day-to-day basis, making sure that the system also works for the rank-and-file people in the organization who need to consume that data and make day-to-day decisions like, transfers or allocations or, returns or purchase orders, is very important. We had a very, great experience in this project with our partner being very disciplined about it.

Cindy Rogers: Very good. Thank you both so much for your time, your good counsel, and Sudip, thank you for allowing Logic to be a partner in this journey.

Sudip Barat: Thank you so much, Cindy and Rupesh.


Credits

Music: www.bensound.com
Cover Photo: Dan LeFebvre on Unsplash

Tagged with: