Mining Your Own Business Podcast

Episode 4 - Navigating Change at Advance Auto Parts with Liping Wu

Navigating Change at Advance Auto Parts

In this episode, we are joined by Liping Wu, a Senior Manager of Data Science at Advance Auto Parts.

Liping and Evan discuss how the many applications of data science that the enterprise analytics team is deploying at Advance Auto Parts, how the enterprise team achieves success with the different business verticals, and the advantages and disadvantages of having an enterprise data science team vs a team embedded within a business function.

You will hear how the the pandemic has impacted their data, the models they use, and the initiatives they work on. Liping specifically discusses the opportunities in improving supply chain demand forecasting.


In this episode you will learn:

  • Where Advance Auto Parts is applying analytics in their business
  • How the enterprise analytics team achieves success working with the different business functions
  • Advantages and disadvantages of having an enterprise data science team vs embedded teams
  • The impact of the pandemic on their data, models, and analytic initiatives
  • The opportunities for using analytics in supply chain forecasting

Learn more about why we created the Mining Your Own Business podcast.

About This Episode's Participants

Liping Wu | Guest

Liping Wu is a Senior Manager for Data Science at Advance Auto Parts in Raleigh, North Carolina. She leads company efforts in implementing analytics, from research and prototyping to modeling, visualization, and automation. Prior to leading data scientists, she served as a Senior Data Scientist, using state of the art NLP techniques to enhance the customer experience.

Liping has a Master of Science in Analytics from North Carolina State University and Bachelor’s and Master’s degrees in Literature from Renmin University.

Follow Liping on LinkedIn

Evan Wimpey | Host

Evan Wimpey is the Director of Analytics Strategy at Elder Research where he works with organizations to transform deficient data into tangible business value that advances their mission.

He is uniquely suited for this challenge by pairing his professional experience in management and economics at high-functioning organizations like the Marine Corps and Goldman Sachs with his technical prowess in data science. His analytics skillset was strengthened while earning his MS in Analytics from the Institute for Advanced Analytics at NC State University.

Evan almost always has a smile on his face, which is at it’s widest when he is helping organizations use data in innovative ways to solve complex problems. He is also, in a strictly technical sense, a “professional” comedian.

Follow Evan on LinkedIn


Key Moments From This Episode

01:50 Liping’s background
03:34 About Advance Auto Parts
06:12 How the enterprise analytics team works with the different business functions
09:33 How the Advance Auto analytics team is structured
13:27 Pros and cons of enterprise data science team vs teams embedded in verticals
19:31 The impact of the pandemic on their data science efforts
26:02 If there was a magic “analytic success” button, where would Liping use it at Advance Auto Parts?

Show Transcript

Evan: Hello everyone and welcome to the Minding Your Business podcast. I’m your host, Evan Wimpey, and I’m excited to introduce you today. Liping Wu. Liping is a senior manager for data science at Advance Auto Parts. Where the best parts are there, people? She’s been there for a little over four years, and prior to managing the data science team, she served as a senior data scientist. Liping has a masters in analytics from the Institute for Advanced Analytics at North Carolina State University. Liping welcome to the show. Thanks so much for coming.

Liping: Thanks a lot for having me. And thank you so much for remembering our company slogan so well. I really appreciate that. Whatever people are, that’s the part. Yeah.

Evan: That’s great. Yeah. Whatever commercials you guys ran in the 1990s. They stuck with me.

Liping: We do have a new jingle, if you wanted to check out.

Evan: Oh, you do? Can you share it? It’s not, like, super secretive?

Liping: It actually just came out a few weeks ago. I don’t know whether that’s public or not, but unfortunately, I cannot say it as well. Maybe give me a few more times, you know?

Evan: Okay. Okay. We’ll have you back on the show in a few months. And we’ll talk about the new jingle and all the analytics behind it.

Liping: Little bit about myself. I majored in literature in my, you know, pre-previous life and I went to IAEA in 2017 to study at Advanced Analytics and then switched to my career geared to data science. And upon graduation, I joined AP in 2018 as a data scientist in the Advanced AI team, which is an enterprise level data science team that supports the data science needs across the company. I moved to a senior manager role last year and currently manage projects while still kind of stay hands on more technically challenging projects. Some of the projects that I have worked with in the past a few years, including customer lifetime value analysis, finance forecasting, the voice of the customer, using natural language processing techniques, and the currently working on the supply chain labor forecasting project.

Evan: Okay. Awesome. That is a very wide scope and it’s an exciting role to be in. Manage a team, grow a team, but also stay technical, hands on. Very cool. And you told us a little bit about the scope there, it’s wide. Can you talk? So I’m just a person who just owns a car and tries to avoid doing work on it whenever I can. But that’s like my perception of what advanced auto parts are there at the retail store where I go and buy an oil filter. You may just give us a quick background on Advance Auto Parts, sort of maybe the scale of the business. What is it all that you’re selling to people like me, the casual oil change person and sort of.

Liping: Sure, for sure. Yeah, we actually serve DIY customers as in your example as well as professional customers. So if you’re comfortable with, you know, like the oil change or changing the windshield wipers or even the batteries, we have awesome team members in store to support you, help you diagnose. What is the issue, what type of the products you need? I even help you install some products, for example, batteries or light bulbs for free. So if you, you know, like you’re not that hands-on maybe especially some very challenging jobs for the car maintenance or repair. You can also go to the mechanics and we also have a relationship with them. So a little bit about Advance Auto Parts, it’s an automotive aftermarket parts retailer. We’re headquartered in Raleigh and we serve, as I said, both professional and DIY customers. The split for our companies is roughly 60% and 40%. So talking about our business functions, we mainly have to kind of make categories. One is we call it a customer support center that includes all the supporting functions, including finance, marketing, merchandising, supply chain and inventory, etc.. We also have our field team. We have roughly 5000 stores and more than 40 distribution centers in the United States and Canada. So that’s about our company.

Evan: Wow. Yeah, that’s quite large. And I’m not the person that you need to cater to the casual DIY or.

Liping: Yes, we care for each single customer and we love to serve you if you come to our store to buy whatever that, you know, the oil filter or so. Yeah. Other stuff.

Evan: Awesome. Thanks. I’ll see you. I’ll see you once a year and then for the next few years. See you. So you’re at an enterprise level for analytics, which, you know, you’re talking about DIY and you know, professional mechanics. You mentioned some projects that you’ve done and like the customer lifetime value, the finance space within some operations. I’d imagine there are a lot of different business functions and then a lot of different focuses within there where analytics could be useful. So I’m curious from your vantage point, how do you sort of vet those? Do you go looking for projects to complete? Do people come to you, come to the enterprise analytics team and sort of how do you decide what to work on?

Liping: Yeah, that’s a very good question. It’s actually the combination of what you mentioned. I would say it’s a combination of the top down and bottom up approach. So as I said, you know, we actually have other data science and a lot of code teams embedded in to separate the business functions. But our team is a little bit different. It’s more kind of like an internal consulting group that supports, you know, the data science needs across the different teams. So usually what we do is, you know, at the Met of the Year, towards the end of the year, we have our annual operation plan and plenty of meetings. That’s kind of like the top down approach. It’s a whole process of aligned, aligning the company’s strategies and near-term goals. And then the focus in the coming year for all the different departments in the company, maybe for the coming year or even including the next 2 to 3 years, the focus across the company will be centered around this company’s key initiatives. So our projects, our teams projects that will be aligned to those kinds of things we call it tier one initiatives. And we also have another approach. We are towards the kind of later of the second half of the year, we will have a team member brainstorming session on the opportunities with our existing business partners or new business partners. So we socialize who we are, what we have done previously and or continue the previous conversation or previous project, what we have done and what we might be able to do together. And then we will come back as a team, have conversations with the partners to figure out what’s the scope, what’s the impact, what is the potential resource needed for that project? After gathering this, you know, ideas from the senior leaders and from our business partners, our team will sit together to walk through the ideas one by one. We will have a short description of what this project is about. And then we will have a short Q&A session on, you know, what is the proposal, what is a concern? Is that even feasible? What might be other areas that we should be focused on or whether we can combine some of the efforts together and we can take a vote? So the end product of this whole process is a list of the projects that our team would like to work on with ranking priority. So then, you know, our leaders in the team, they will plan accordingly as we progress through the year and then as resources are taken or freed up.

Evan: Okay. Awesome. Yeah, that sounds like a well-oiled process.

Liping: And we have improved a lot during the past few years. And I am proud to say we’re getting better and better.

Evan: Yeah. Yeah, that’s great. And you sort of mentioned the operation planning and then your team coming together to help scope out and right size and ask the right questions. Is your team just data science? I’m curious where maybe an i.t or data engineer would sit in that process. I’d imagine they would have a lot of input into, you know, what data is available to build your models and then what infrastructure is available to to serve your models.

Liping: Right. Right. That’s a really, really good callout. So the team structure in advance AI, we have three sub teams. One is focused on analytics. So our main support currently is the marketing team because the marketing team is relatively advanced in terms of how they use data and how they use and get insights. And we have an enterprise data science team which is kind of like our core data science. Force. And then we also have a data engineering team who helps us to connect ways, the I.T. ways, our data owners, other infrastructure in the company to and curate the data for us, migrate data for us, clean the data for us. So I think this way teams work very, very closely together on a daily basis and we do cross-pollinate a lot on the ideas, serve each other on the knowledge, and also serve as each other’s counselors on how we should approach this project. We will have some projects that we pull resources from. You know, all three teams together to form a project team. It’s more kind of our daily operations and more project based. Yeah. As you said, I feel really, really fortunate to have data engineers in our team as well as data analysis so we can work very closely together. And that really serves well for our goal, our mission of building end to end the solution. To customers. Yeah.

Evan: Yeah, absolutely. And I think that’s good. I think that’s really like the ideal for most people working in data science or most people who are trying to get data science implemented like that is the key. Can you have good communication between your business stakeholders who have some initiative? They want the people who can do the modeling and the people who can move the data and the output from models. So.

Liping: Right. I remember when we were in the way, you know, like our professors told us that, you know, like the time kind of split for a typical data science work is like 80%. You just massage the data, getting the data in the right format. Right. And then now you can do like really fancy stuff of data size modeling, you know, analyzing it. But I was not aware that prior to that 80% there was like a whole maybe 20% of the effort is even finding where the data is, right? Like how to get the data into the team’s process, to the team’s access. So it’s a whole lot of effort going on there. Yeah. So we really appreciate that we have our core data science data engineers in the team that we work very closely together.

Evan: Yeah, that’s great. I’m curious, is that the timing? Was it sort of in parallel that your team on the analytics and data science team was built up and embedded there at the Enterprise at the same time that the data engineering team was.

Liping: It was structured like that when the team started. Yeah.

Evan: Okay, awesome. You also mentioned in passing a little bit when thinking about prioritizing, you said you’re at the enterprise level, but there are embedded data science teams in business, in the different business verticals. I think that’s a pretty key question for a lot of folks that are trying to start and are trying to understand how they can organize their data science resources. And, you know, do you have that sort of federated enterprise level or do you have the embedded and it sounds like you guys sort of have a hybrid approach. I don’t know if they’re their pros and cons. If you think that’s the useful way to do it or a reason why it fits best for you guys in advance.

Liping: That’s very broad, the question is a great question. Yeah. I’m just going to try my best to share my thoughts from my perspective. I do see the advantages of having an enterprise data science team because usually compared to the folks who work embedded in the business functions, the folks who work are kind of more centralized and more. Close two teams waves. The folks who look like us who have the similar skill sets it’s easier to kind of cross-pollinate the technologies as it evolves because we all know as the design is we need to keep learning every day otherwise you know, maybe a month later the new technology is coming and coming out and you’re still using the old technology with which takes a long time. I’m not as good at performing as your new ways of doing things, especially in the deep learning field and LP feel that that’s especially true. So in terms of applying the new technologies, keep the whole team motivated to keep learning and growing the technical skills. I think that’s definitely better. And having those folks that you can cross-check the ideas with, you know, as I said, you know, with the data engineers and do analyzes and data scientists that way you are, it’s more likely you’re going to find a more optimal solution for the problems you wanted to tackle. The downside of this enterprise-like or the challenge the data science enterprise data science team faces typically are you’re not so close with the business functions so you might not find it does not come natural for the team members to understand the goal, the scope or even the day to day of those business partners. Like what do they need to deal with? What do they care about? Because we are not in their shoes. And when we talk to them, it’s more like a partner instead of one part of them. So the business buying might be like a challenge, like a potential hurdle that you need to go through. You need to win the trust instead of if you are in their team, you’re part of the team, you’re you have to trust and you have the resource that’s needed. I think the reason why we have it right now, is that we actually don’t have a data science team for every business function. It depends on really it kind of depends on the size and the analytical requirements of that business function. We also have a big data science team in merchandising. It’s because the day to day the deal weighs figuring out the inventory, figuring out the demand. There’s a lot of forecasting that needs to align and to make sure that we get the right, you know, what is the setup? Let me take my words back. What is the set of the products that’s needed that reflects the true customer’s demand in each store location? So that’s always a big task. So that’s why there needs to be a data science team supporting that forecasting needs and supporting maintenance of that forecasting. I. I don’t know where we’re going, I think compared to some kind of new tech companies. It’s much easier or more natural for them to have each data science team embedded in the business, because each business function is so advanced in terms of their analytical needs, their data science needs, and how they turn those data insights into the actions. We’re just not quite there yet. I’m also. Wondering that for a smaller size of the company or even the startup, it might be different. The pattern is just like this current pattern suits. Our business needs the best.

Evan: Yeah, sure. I think that makes perfect sense. And I think it was going to ask about the teams, maybe the business functions that aren’t as mature. And if they don’t need an embedded analytics team or an analytics people, that’s fine. They can still lean on the enterprise until they sort of reach that threshold. So I think I think that makes perfect sense the way you’ve sort of tried to balance those things there. I also really like you. You’re talking about the team and working together and learning together. I think the most frustrated data scientists that I’ve ever talked to are the ones that are sort of on an island embedded with the team as the only data scientist. And it’s frustrating and you don’t get the opportunity to learn, but also there’s a lot more opportunity for mistakes or for bad practice. Even with the most seasoned data scientists.

Liping: There’s nobody to still have a group of the superheroes to think about instead of only thinking.

Evan: Right, yeah, everybody’s got their own kryptonite, but whatever Spider-Man doesn’t get impacted. I don’t know. Carried the analogy too far. Awesome. Okay. I want to ask you a couple more things. You’ve been at Advance Auto Parts, I think a little over four years. So I.

Liping: Think it’s four years.

Evan: Or close to four years. Not a surprise to anybody. About halfway through that time, there was a global pandemic that started. And I would imagine that has changed everything that every data scientist does. But I know I’ve driven my car a lot less. I would imagine people are driving their cars a lot less. I’m curious if there’s you know, what has the impact been? Have you had to retrain all new models? Have you had to throw away a bunch of models that you used to rely on?

Liping: Yeah, that’s a very, very good question. And it’s not just about cars like a pandemic. It just changed people’s lifestyles. A lot. In every aspect of that. And we can imagine. Right. Actually, that was kind of the concern, you know, when the pandemic really started, like especially in March, April 20, 20ish. Like people just stayed at home, not doing anything. And it was a big uncertainty at that time. I read some articles in Towards Data Science and then the author announced that this is the end, this will be the end of the forecasting period. But it was actually not that bad. It was not like that. But in terms of the people driving around, the way we do actually track the mobility data from Google, which provides information on the percentage and mobility change. Six categories of the places I think include retail, grocery parks, public transit and. Workplaces, I think. Scott: The last one. Residential. Yeah, residential. So it actually seems like after the initial uncertain times in March, April 2020, people started to drive again, much less, but started to drive again, especially to parks or for outdoor activities. And according to the Bureau of Transportation Statistics, the shorter distance trips are the trips. Stats that are less than 25 miles compared to 2019 had some really big decrease, like a roughly decrease to 50% in 2020 and about like to 70, 80% in 2022. But the trips with a little bit longer distance between like 50 miles to about 500 miles. So you can think of that roughly as from Raleigh to Pribaro, that’s about 70 miles from Raleigh to New York City. That’s about 500 miles. So except for the dip in 2020, like April 2020, people are taking more trips than 2019 in the past two years. Wow. So I think that’s very fascinating. Right. Like, if these are the troops that we can manage still driving and it’s safer than taking the flights.

Evan: Wow. Yeah. That’s counterintuitive for sure. But it does make perfect sense. Does serve avoiding flights. Yeah.

Liping: Right, right, right. Like so that’s theater is the power.

Evan: Right.

Liping: That being said, we did see some kind of big change in the sales patterns, in the customers behavior during the past two years. Actually believe it or not. And this had some positive impact on our company on modernizing the forecasting methods because in some businesses a function that people used to use like old fashioned forecasting methods such as just a weight in a moving average. And they started to realize when the historical patterns don’t hold true, they need to switch to more sophisticated forecasting models instead of just applying the average and applying the weight, the weights on the average. So the auto parts industry had record high DIY sales after the dip in March and April. In 2020, we actually had a really good sales year in 2020 and 2021 because counterintuitively might seem. But actually it would make sense because we all know that people have more time at home and they take care of their houses. They also take care of their cars or those jobs, those maintenance jobs that probably are long overdue. Right. And then also, the pandemic accelerated the shift to e-commerce. So some industries closed out, you know, on average by ten years. They kind of pushed e-commerce. Forward by ten years, but it might vary depending on which industry you’re in. But our e-commerce also benefited from the purchase pattern shifting in 2020 and 2021. Okay, so why don’t you talk a little bit more about the modeling challenge? Because these are symptoms, the brain challenges modeling, because the previous patterns now don’t help you. What should we do? We make two aspects of the adjustments. Mainly, one is adding more features to capture the macro factors. So, for example, the computer case, data mobility data, consumer confidence index, a stimulus check, we do have those kinds of more features in. Retraining some of our models. And we also try to tune the models to focus more on the recent data and rely more on near-term patterns instead of the longer term patterns from 19 or 18 previously. So I would also like to add here that we did find that simplicity is the virtue in these changing times. The complex models usually require retraining more frequently because the features might also change drastically and distort the predictions. While, for example, the Sarah Max model that we built in 2020 still it’s still generating reliable forecasting now. Yeah

Evan: All right. Now, simplicity. Simplicity is certainly a virtue. Yeah, that’s great. And it’s incredible how counterintuitive some of those things are like auto parts, retail auto parts is one of the things I thought, well, that’ll be hugely impacted. But yeah, you, you make a, you make a very reasonable case that well, the data doesn’t support that. And I think it’s seen across a lot of industries that e-com has accelerated greatly in the past couple of years. Super. So I want to ask you one more question, Liping. You walked us through sort of the enterprise level planning process that advanced auto parts and whatnot. Let’s just hypothesize. We scrapped that whole planning process and it is the Liping show. And you get to set the priorities for, you know, the next six months or even for just one singular project. If you’ve got one thing that you and your team can work on and you’ve got full buy-in from the top, from the CEO level down to the business, who’s going to be acting on the decisions you make and you get to work on whatever you want? Where do you point your team? Where do you point your efforts?

Liping: In the heavens? Right. Well, I think this idea for us, that is a very, very bold question. I think I have. Everything said, I don’t like the stores, stakeholders alignment or the buying that I need and the resources I need. I would like to continue my work on the supply chain side. I’m currently working on a supply chain labor forecasting project. I have come to the realization how complicated and dynamic the process is. When I started to work on this project, we all knew from direct experience how the global supply chain has been pushed to the brink over the last two years. Many aspects of this complicated system have been heavily challenged and the impact is everywhere from the computer chips that we read from the article to toilet paper that we go to, the grocery store, the supermarket and the weights that we theorize ourselves. So there are a lot of opportunities in the journey of modernizing the traditional supply chain. I think the project going on our supply chain side will be especially exciting because A.P. is expanding our business. There is a huge business impact in supporting the ongoing replenishment it needs to. Our DC still needs to take the huge volume in on a daily basis from the suppliers across the world as well as they need to make sure order replenishment going to the stores according to the plan so our customers can have what they need in our stores, keeping this kind of a huge chunk of work going on while optimizing under the constraints of the supply. We’re still facing that challenge now. The transportation, logistics and the labor shortage that we are experiencing now, United States is a super challenge, but I think it would be super fascinating.

Evan: Yeah, that’s a big problem. A challenging problem. Right. And a fun one to work on from analytics. Yeah, I think I would think especially at advanced auto parts, but just the scale of what you do, dozens of distribution centers, thousands of stores and all of the constraints that go into it. So I hope we get to continue working on it. I’m sure you and your team will experience a lot of success. Liping, Thanks so much for joining us on the show today. It’s been super enlightening.

Liping: Thank you so much for having me. I enjoyed talking with you.

Evan: All right. That’s it for the show today. Make sure to, like, subscribe, multiply and divide and we’ll see you next time on the Mining Your Own Business podcast.