Rise of the ML Engineer: Chick-fil-A’s Korri Jones
Korri Jones has a call to action for senior leaders: take Data Science and Machine Learning seriously and build the right teams and systems to adapt to unforeseen and unexpected business and market changes.
“Drive by any Chick-fil-A across the country during business hours and you’ll almost always see a fast-moving line. Maintaining that scale from an operations standpoint, while simultaneously focusing on a top-notch customer experience is far from easy. Our data science team also shares in this responsibility and attempts to support our restaurants and overall business as best as we can.”
We had a virtual sit down with Jones to discuss how the third-largest restaurant chain in the U.S. uses AI, the evolving role of ML engineers and how the company thinks about ethics and bias in AI.
Q: Korri, what is your organization’s role in helping Chick-fil-A maintain its steady growth while adhering to its high standards?
Jones: My department, Enterprise Analytics, is mission-driven to build and scale analytic capabilities that enable enterprise business strategies. Meaning, that for our more than 2,400 restaurants, which are open 6 days a week, we exist to serve and support the owner/operators and the team members that represent the Chick-fil-A brand each day. While my team doesn’t sell chicken, we support those that do and our mandate as an organization is to ensure that we’re using the right data and tools to help deliver a fast, high-quality food experience that keeps customers coming back.
Q: How do machine learning and AI fit into the mix?
Jones: Not to sound cliche, but we look at these technologies as a means to an end, which is to deliver continuous value for the business over the short and long term. Our organization is responsible for delivering new ways to achieve business success. We are encouraged to innovate and think outside of the box when we are trying to solve business problems.
If you look at many of our top projects across both our IT and other organizations, a representative number of them have AI/ML components, either directly or indirectly. Examples include our mobile app, which can predict users’ preferences based on their previous purchase history and recommend items to use their accumulated rewards points as well as identifying emerging food-borne illnesses by analyzing social media posts. Throughout the pandemic, these technologies have played a huge role in helping us adapt to make sure our customers receive the highest level of care, and that our team members and owner/operators have as many tools as possible to provide the type of experience that many associate with Chick-fil-A.
Everyone from the C-suite to the frontline in our stores is focused on using data to support healthy sales growth and to boost the overall customer experience.
Q: With this in mind, what’s the optimal team structure to achieve your goals?
Jones: At Chick-fil-A, our approach is to hire people that are tenacious thinkers with big hearts and a natural curiosity. Whether they are software engineers, data scientists, or the analyst that engages with all verticals of the business, the above characteristics are the building blocks for some truly awesome teams!
We approach MLOps with a shared ownership approach across the organization. We have a data science team that builds models and data engineers that operationalize them. Our first line of defense once models are in production are our systems analysts. They use model monitoring and observability tools to track performance and drift – and if these tools observe anomalies, the analysts work with the appropriate engineering teams to either fix broken data pipelines or with our data scientists to help optimize model performance.
As senior lead machine learning engineer, alongside other ML engineers, I play a strategic role in helping orchestrate the overall ML pipeline, making sure it’s resilient, reproducible, and scalable in order to ultimately optimize the data science lifecycle. At Chick-fil-A, ML engineers interface with business experts, data scientists and systems analysts. We are tasked with diving deep into all aspects of the data science life-cycle and making sure that our processes and tooling enable our AI/ML products to solve the right business problems in ways that enable speed to value through a combination of engineering & data science excellence. A key measure of success for our ML Engineering teams involves reducing complexity as a mechanism to allow data scientists across our organization to move focus on their craft and not the infrastructure, which is a cornerstone of a robust MLOps practice as we see it. Also, with the inherent complexities of productionizing models at scale, we are committed to co-learning, co-building, and defining the right processes in order to ensure that team members across the organization aren’t wasting cycles on recreating the wheel each and every time they attempt to productionize AI/ML models.
Q: Do you find it difficult to bridge the gap between data scientists and data engineers that come from different backgrounds and may not speak the same language when it comes to models across the build, train, deploy continuum?
Jones: Not every person in our organization thinks about data in the same way, and that’s OK. Our job is to stitch together the right people, tools and technology to achieve scale and performance without losing velocity. That’s the essence of the ML engineer’s role.
Using a football metaphor, our data scientists are my Quarterbacks and I’m part of the O-Line. If we’re having an issue with a model in production (aka the “real world”), our scientists shouldn’t have to become experts in production environments, myself and our other ML Engineers can stand in that gap.
Q: So part of your role is to align all the systems and processes to make sure MLOps are optimized for velocity, scale and effectiveness?
Jones: At the end of the day, as an ML engineer, I don’t build or write models, I support the process to make sure it gets to production with the greatest likelihood of achieving the goal it was designed to achieve.
I want to get to a point where we can have a model pushed into production within 5 minutes of development being completed and the data science team giving the mark of approval, but to accomplish that, we need to have the right processes in place to provide the necessary clarity & context while also making sure we are committed to not overcomplicating things unnecessarily. If we can achieve that, then we are delivering unique value that will help Chick-fil-A continue to innovate, care for our customers, care for our phenomenal restaurant team members, and empower our owner/operators to do what they do best, which is make an impact in their communities and in the lives of those they serve.
Q: How do you think about AI ethics and bias in your role?
Jones: Bias and AI Ethics are very critical conversations that we have internally. Frankly, everyone in the data science space should be thinking about these topics, and if they aren’t, then those are some major red flags in my opinion. As our organization grows our usage of ML/AI, we also grow our awareness and intentionality when it comes to the topic of AI Ethics. Just because we can use our data in certain ways, doesn’t mean we should. Those types of discussions are one of the reasons that I enjoy doing what I do, because it kicks off the right discussions across multiple levels in the organization while also enabling more diverse voices to be heard.
Also, as a representative of an ethnic group that hasn’t fared well due to poorly developed models generating biased predictions that were built using biased data, I am personally very in-tuned to this overall topic. We, as a community of data science & engineering practitioners must challenge ourselves and our leadership to always keep the bigger picture in mind as it deals with all things AI Ethics & bias related.
Q: We were intrigued by some comments you made recently about how it’s important that executives have a better understanding of ML initiatives, can you elaborate?
Jones: No matter how talented your data science & engineering teams are, the core projects that they work on will ultimately be defined by the needs of your respective business. Meaning, without leadership awareness, support and understanding of the overall data science life-cycle, odds are that getting the necessary funding or headcount will be an epic journey. Imagine being a world-renowned data science practitioner but to not have any dedicated engineering support to push your models into production and keep them healthy? Why does this happen, probably because leadership isn’t made aware of the needs required to effectively support data science at scale. We as data scientists, ML Engineers and analysts have to lean-in and educate our immediate and extended leadership in ways that are valuable to them. Without this, no matter how awesome your team is, the work will be far more cumbersome than it should.
One thing I am so very appreciative of at Chick-fil-A is that we have leadership that is bought into the promise of data science. Without that level of support, I assure you, that we would not be where we are today, especially with the rapid pace of change in our industry.
Q: What’s next?
Jones: I spend a lot of time trying to think of the best strategy for operationalizing data science workflows to help build smarter restaurants. We want to connect, analyze and support our owner/operators and their respective businesses as much as possible. Because the better we can support them, their teams, and ultimately their business, the better experience they can provide to their customers and their respective communities. To see how that plays out, you’ll just have to visit a Chick-fil-A soon.
This is the third piece in a series profiling The Rise of the ML Engineer. See also: Lyft’s Alex Zamoshchin.