A new tool allowed the chip maker to fully automate inventory planning. The result: Inventory targets that planners accept more than 99.5% of the time.
Inventory is a universal headache at most companies – with the potential for significant impacts, both good and bad. Too little inventory puts sales revenue at risk. Yet, having too much inventory sitting around increases costs. And, for inventory planners, getting it wrong can be a career-ending move. Through a unique project with Intel, my colleagues and I built a tool to help companies get inventory right.
Through its deployment, Intel transformed inventory management into a fully automated “lights-out” process that generates inventory targets that planners accept more than 99.5% of the time. It’s also reduced the time required to commit to customer orders from months to one day and generated over $1.3 billion in increased gross profits since 2014.
Our engagement began as a research project in 2005 that was designed to help Intel improve its forecasting and inventory processes in one division. The 9 years between the launch of the research project and the beginning of a payback period may seem like a long time. The short answer is that the math had to catch up to its use in the field; then, once we began to deploy it, we made refinements to the tool before it began to deliver results. When I present this research along with Matthew Manary, who was then at Intel, we joke that practitioners can’t believe we did the project across an organization as large as Intel in just 9 years; my academic colleagues wonder what took us so long.
At Intel, the project unrolled through a series of steps. First, we integrated the forecasting process with the fulfillment process by understanding how planners made decisions for different products, especially new products. Then, we mapped those processes to start building a mathematical tool to automate the inventory planning process.
By reducing the supply and demand problem to a set of mathematical equations, we could solve the problems to come up with the right inventory level. The tool didn’t just guess at the answer; it provided the exact inventory required for each product.
We also sought feedback from people in all parts of the company. This included providing a two-year period of time for planners to override the automated process. If they believed the tool was wrong, we asked for their reasoning and found that the problem was usually with the inputs. This allowed us to fix issues and code improvements into the tool.
Interestingly, we found that if a product was less than 23 weeks old, we needed to map the product’s forecast to a historical item. However, once we moved beyond a product’s initial 23 weeks, the tool only needed to look at the item’s own unique data. We discovered this cut-off point through trial and error and by understanding how planners made their decisions.
This process resulted in significant financial and operational improvements at Intel. Today, this lights-out process manages approximately 85% of all finished-goods inventory. That means no planner needs to manage any part of those finished goods because the process is fully automated, allowing them to work on other projects at the company.
The good news for other companies is that this tool * can work for everyone. Any company interested in automating and optimizing its planning process can benefit from it. However, there are a few questions to ask before moving forward:
Is your company math phobic?
This tool is based on math. If you’re afraid of math, this isn’t the tool for you.
Is your company ready for change?
It’s common for managers to say that their current process isn’t perfect, but works just fine, or that a particular area is too important to the bottom line to make any changes. To be successful, the company’s culture needs to be committed to driving change through automation.
Does the team represent the entire business?
The team that drives change should be comprised of analytics people to create the models and planners who deeply understand the inventory planning process. You also need the business owners to support the initiative and finance folks to validate the savings. If you only solve part of the problem because of a narrowly focused team, then you can’t fully implement the tool across the company.
Can you map new products to existing ones?
You need to adjust operations research to the realities of practice, which means mapping new generations of products to previous generations. You need to understand how planners make decisions about existing products and new products. Invest the time and effort to create models that fit the realities of an imperfect business environment. As groups use the tool and see that it works, it paves the way for a smoother rollout to other areas.
Can you retain institutional knowledge?
Employee turnover can impede implementation of new efforts. Make sure seasoned team members are retained to ensure momentum.
This process was transformational for Intel and can be a powerful tool for other companies to solve their inventory problems. It’s a big step, but worthwhile for companies that are ready for change.
* The tool described in this article was developed in conjunction with my role as a then professor at Boston University and as co-founder of Optiant, an inventory planning and management solution, along with several Ph.D.’s from Optiant. The firm was later acquired by Logility, where I am the chief scientist. Our work built on research I began at MIT in the 1990’s with MIT Professor Stephen Graves. The tool is now available as a commercial software solution from Logility. However, our work was a finalist for INFORMS’s Edelman award and was published in the INFORMS Journal on Applied Analytics. Organizations that don’t want a commercial software product can code one tailored to the needs of their organization utilizing our research. We know of several organizations that have taken this approach. The article, Analytics Makes Inventory Planning a Lights-Out Activity at Intel Corporation, is available at https://pubsonline.informs.org/doi/10.1287/inte.2018.0976.
Sean Willems is a visiting professor of operations management at the MIT Sloan School of Management and a professor of business analytics and statistics at the University of Tennessee’s Haslam College of Business. He can be reached at firstname.lastname@example.org.