Ana Elsa Glaser, Supply Chain Systems Analyst - Demand Management
NXP Supply Chain Systems & Processes Organization
Country: United States
Organization: NXP Semiconductors
NXP Semiconductors N.V. is a Dutch semiconductor designer and manufacturer headquartered in Eindhoven, Netherlands. The world leader in secure connectivity solutions for embedded applications, NXP is driving innovation in the secure connected vehicle, end-to-end security, and IoT smart connected solutions markets. As the world's 5th largest chip-maker, the company has 29,000 employees in over 30 countries, with approximately $11B in annual revenue.
Over $3 billion (~30%) of NXP’s annual revenues come from 3rd party distributors who serve what we call the mass market at NXP. Accurately forecasting demand for this segment has major implications for our business. Decisions based on those forecasts determine things like, the amount of raw materials that are procured, manufactured into semi-conductor wafers, quality tested, cut into chips, packaged, shipped, and ultimately included in 10+ million electronics devices and cars every year, in end markets that expand across the entire globe.
Simply put, accurately forecasting demand is an essential part of effectively running our supply chain, managing our human resources, utilizing our capital investments, serving our customers, and impacting the bottom line for shareholders. The better we perform at forecasts, the healthier business we can run, the better we can compete, and the more value we create.
Demand for semiconductor chips can unpack into a host of leading and lagging economic indicators. Imagine, for example, all of the economic forces that can influence an automobile manufacturer’s decision on how many cars of a given model to produce and/or whether to invest in new electronic technologies that rely on NXP chips for the cars they bring to market. Everything from oil prices to geo-specific employment statistics is fair game for such analysis.
Not too long ago, however, in my work serving NXP’s supply chain operations, I recognized a major gap and an imminent need to modernize our demand forecasting methods, which at the time relied exclusively on using historical sales data to predict future sales. At the time, our supply chain analysts and demand managers were working with NXP’s business lines and marketing teams to visually assess descriptive demand trends, distributor inventory, and resale data to guide our forecasting decisions.
We were not considering a vast wealth of additional information at our disposal and therefore were missing the predictive power of the sometimes “hidden” relationships between these additional data elements that can make a world of difference in a context where incremental accuracy improvements can result in millions of dollars in impact.
In a fiercely competitive semiconductor market, our status quo way of doing things was putting our competitive advantage at risk, as new ML/AI tools coming online were available for our competitors to get ahead.
With this as the backdrop, I secured my leadership team’s nomination and sponsorship to attend NXP’s first-ever Citizen Data Scientist (CDS) up-skilling program. This is an intensive, months-long program that draws a multi-disciplinary cohort of participants from nearly every business line at NXP, with the objective to help infuse ML/AI practices into all of NXP’s business functions by teaching people the tools (including Dataiku) and data science methods used by enterprise AI practitioners.
A unique part of the program involves a requirement to “bring your own use-case” so that the foundational citizen data science concepts being taught during the classes can be applied to actual projects that participants can work on throughout the program. This way, someone like myself can learn new skills and also can bring a minimum-viable project back to their business at the conclusion of the program.
In my project, I made use of the additional data elements we had not been including in our forecasting methods to date, and also applied more sophisticated forecast techniques to modernize and improve the accuracy of our mass market distribution forecasts by factoring in multivariate relationships never before considered.
Through my studies, experiments, and hard work, I built a prototype for my project that was determined by a panel of NXP judges to be one of the top use-cases in the program. That success inspired me to continue pushing ahead with my work. With the assistance of NXP’s enterprise business intelligence enablement team, we delivered this into production as one of NXP’s first-generation operationalized use cases.
Through this experience, I began to see the transformative effects that modern analytics techniques can have not only on solving a specific business problem but also on engendering a new culture of innovation within an already mature and innovative company and creating a pathway for me to learn, grow, pursue my interests. Moreover, this gave me a platform to be a change agent at my company and inspired me to expand my data science knowledge beyond NXP’s CDS program, and I went on to register for an AI/ML Certificate Program at the University of Texas, before taking another major leap to enroll in a master's degree program to further my learning about data science and ML model building, so I could continue to take on larger responsibilities in ML projects for NXP in the future.
Armed with my newfound knowledge, I started to see new opportunities that I had not previously considered. For example, I partnered with the Eindhoven Technical University to develop an experiment that 1) corroborated that our new ML model outperformed our prior-state statistical forecasts and 2) helped determine which algorithm garnered the best results.
In this study, we found that using sales data, resale data, distributor inventory, as well as historical orders, improved forecast accuracy, and reduced forecasting bias. These improvements allowed us to better predict the peaks and valleys in our distributor demand, and ultimately improve the accuracy of our overall results.
Based on the success we’ve had deploying this ML-based solution to forecast mass market demand, I have been tasked with scaling these new modeling techniques to forecast all market segments of our multi-billion dollar company. The reusability and scalability offered by Dataiku enable us to efficiently make this transition and expand our ML footprint in the organization.
Furthermore, I’m taking on the responsibility to extend our usage of Dataiku and ML to other problem statements and applications. One such example of this is using Dataiku as a repository for operations research experiments, which we then repoint to our production data server and scale to production. On this front, the experiment tracking and reproducibility offered via Dataiku enables us to quickly adopt successful and novel techniques proposed in my partnership with Eindhoven university.
As we’re forecasting demand for NXP’s multi-billion dollar business, the incremental improvements we’re seeing in forecasting accuracy can influence millions of dollars of costs and revenues. Analyzing complex data relationships via machine learning enables our organization to produce more accurate results, in a faster time than ever before. The Dataiku workflow has also enabled some rationalization of IT services that previously extended across multiple applications.
But perhaps the greatest value for NXP is that there’s now greater interest in leveraging ML techniques in supply chain operations. The supply chain organization has since invested for the first time in recruiting classically trained data scientists. Moreover, the organization is working on finding, prioritizing, and developing a pipeline of ML use cases on high-impact issues affecting NXP.
These investments in AI have the potential to unleash 10s of millions of dollars of the returned value to NXP. Additionally, beyond our business and customers, we’re now able to contribute back to the community with this work through our joint research with Eindhoven University, with whom we’re in the process of presenting one of the experiments developed by students at an AI/ML conference in the fall of 2022.
The evolution in data science that I’ve seen in my team would not have been possible without the successes I could help deliver using the Dataiku platform in partnership with Eindhoven University and the Citizen Data Scientist program managed by NXP’s enterprise business intelligence and enablement teams.
Thanks to Dataiku, I’ve been able to pursue my dream of becoming an ML practitioner for NXP. To put my journey into context, what started as a potential use case presented when I was a business analyst in 2019 is now an operationalized ML model managed by the 2022 data scientist version of myself, with several new and exciting solutions in development and a defined roadmap of more projects to come.
Dataiku enabled me and my team to take our experiments out of people's laptops and have a structured repository where students, junior, and senior data scientists can collaborate on solutions and track experiments as they evolve. This experiment tracking and reproducibility enables us to efficiently adopt research that is beneficial to our organization.
Having convenient access from Dataiku to our enterprise data architecture, such as TeraData, Oracle, and other data repositories enables us to quickly integrate solutions into downstream enterprise value streams and serve the business. The ability to reuse and repurpose successful solutions empowers us to quickly expand our offering to other segments of the organization, increasing the value derived from selecting Dataiku as our Enterprise Data Science Platform of choice.
Finally, the visual ML makes it easier to rapidly prototype solutions and get better results, faster.
Great summary and great project, Ana! Glad to see continued long term dividends from NXP's Citizen Data Scientist Program. You're a key example of the modern workforce we're trying to create through this program.