Principal Statistician, Procter & Gamble
User Reference Manager, JMP
Zhiwu Liang is Principal Statistician at Procter & Gamble, based in Brussels. With nearly 20 years of experience working in the consumer goods industry, Liang is skilled in design of experiments, non-linear model and prediction, statistical modeling, decision trees, time series analysis and prediction. He holds a PhD in Mathematical Demography from the University of Groningen in the Netherlands.
In a recent conversation, Zhiwu and JMP User Reference Manager Meg Hermes discussed the course of his career, communicating data with non-statisticians and the many reasons he sees to nurture a collaborative relationship with JMP software developers.
Meg: In your role at P&G, you do a lot to promote analytics enablement through teaching and consulting as well as doing your own work with more complex modeling and experimental design. How has your advocacy for analytics transformation within P&G affected your career trajectory?
Zhiwu: Advocating for analytical methods is my job! But it has also definitely helped me to gain recognition within P&G. And a huge part of that is having a suitable tool with which I can easily communicate with non-statisticians. By teaching JMP skills at P&G, I’ve come to be known not only as a statistician but also as a JMP expert. Whenever my colleagues have questions about JMP – no matter whether it is stat-related or not – they come to me. And I can often find some issues related to their design and data analysis, figure out the problem and help them to use the correct tool for design of experiments (DOE) or modeling (data analysis). Often before they even realize they’ve used the incorrect method.
Meg: Tell me more about using JMP on a day-to-day basis. Are there ways in which the software makes your standard processes – or even your job as a whole – easier?
Zhiwu: For me, JMP is not only a tool for data analysis; it is also a communication method with which to transfer difficult statistical thinking to a simple graph or profiler and enable others – especially non-statisticians – to understand the data and models. The tool I use most often in JMP is DOE, and it’s much better than other software. At P&G, we have many statistical courses which require participants to use JMP. You can see people’s excitement and satisfaction when they get the correct results so easily for even the most complex models. In fact, my colleagues are also now using JMP on a daily basis. They like the visibility function in JMP because it makes it so easy to show the results to their manager.
Meg: One thing that is relatively unique about the JMP organization is that we see relationship-building as a way for our customers to optimize the benefits they get from their investment in JMP. Can you talk a bit about how you engage with the JMP organization and the broader network of users?
Zhiwu: There are a lot of forums in community.jmp.com that we utilize for sharing information and asking questions about JMP. But I personally prefer Discovery Summits because at Discovery, you can always learn new things – whether it’s a new method, new tool, new application. Also you can easily connect to other users and the JMP software developers. By talking to the architects of the software, you can develop an even better understanding of the methods you have access to in JMP. And you can suggest they add new features for a specific model in upcoming version releases.
Meg: That’s interesting. Why is it so valuable to be able to provide this kind of feedback to JMP software developers?
Zhiwu: It is very important for P&G to cultivate this kind of collaborative relationship. We have thousands of people using JMP on a daily basis. Different functions and categories have different needs. Sometimes we even have to combine Excel tools with JMP to solve some particularly tough questions. So we need to provide timely feedback to JMP about any existing features that do not work in specific situations. We then ask JMP developers to improve the software or provide suggestions on how P&G might build a JMP add-in tool (which, I have to say, might only work for the current version). But that’s the way we can let JMP make people’s lives better at P&G.
Meg: What is the future of JMP at P&G?
Zhiwu: With data science and machine learning booming at P&G, we expect JMP can embed more machine learning techniques, such as convolutional neural networks, LightGBM, Catboost and NPL tools – just like you did for XGboost – to make our lives easier and solve most of our problems within just one tool.