AI 2020: The next steps for retail supply chain operationsAI 2020: The next steps for retail supply chain operations
AI 2020: The next steps for retail supply chain operations
December 9, 2019
9 December 2019
The adoption of AI-powered automation technologies in retail is projected to leap to 80 percent in the next three years, with the most significant growth expected to fall in supply-chain planning. In this low-margin, fast-moving industry, gaining a competitive advantage is no mean feat – but machine learning technologies are allowing companies to finally take control.
Dr. Andreas Schmidt is product director at BlueYonder, a startup founded in 2010 at the Karlsbruck Institute of Technology and later acquired by American software company and consultancy JDA. Dr. Schmidt’s background lies in physics and data science; he helped build and develop the first iteration of BlueYonder’s AI-powered ordering system for fresh and perishable food. Since then, he has moved away from coding into product management, where he connects businesses with technology.
Ahead of the AI Summit NYC this week, we caught up with Dr. Schmidt to discuss the major changes to retail supply chain technology over the past 10 years, the challenges of tying up loose ends between technologists and business functions, and the shift away from AI hype towards everyday business value.
Q: You’ve been involved in retail and supply chain AI
technology for nearly a decade. What’s changed in that time?
A: “The excitement and readiness to apply AI technologies has increased significantly. In the beginning, when I first joined the company, it was less focused than today. You could’ve done very nice things with data and AI and machine learning for example, but it did not really seem there was a need for it because the conventional technology worked well enough.
“There were many fields that we were looking at. We quickly
understood that retail and supply chains are compelling fields to explore AI
because there is always the need for better solutions. Retailers were ready to
adopt a technology not because it was cool and fancy, but because it adds
Q: What have been the key challenges in terms of
communicating between data science and business functions?
A: “I realized relatively early on in the development of our system that of course, you need to get the technology right and understand the data, make the right predictions and decisions and all of that. But ultimately, successfully introducing any such technology hinges on change management and on the willingness to adopt.
“If you imagine you’re an experienced supply chain planner with a lot of past experience that you built your whole career on, and then some student comes along with a machine and says ‘I can predict fresh produce inventories better than you can’, of course you don’t believe it.
“Then this machine actually turns out to add value and take
better decisions then that change manager even feels threatened and will try to
prove that they still add value and that they can improve on what the machine
suggests. So, there is an intrinsic tension in that. There’s a huge shift in
all of the roles involved, and if that’s not managed in the right way, it will
just kill any success and the potential of any added value.”
Q: Is the hype around AI dying down?
A: “There is hype and inflated expectations as with any technology cycle. There are big stories about what AI does when it recognizes human speech or beats the world champion in chess or in Go, but these are not really the most value-adding applications.
“The actual everyday business value is created in mass decision-making and this is very close to the core of the product I work with. If you’re a retailer and you have 500 stores and 20,000 products per store these are 10 million decisions you need to make every day, because for each of these products in each of these stores you need to decide whether or not you need to order a case to replenish from the distribution center.
“A human being can do this very well for a few individual products – for those that are most visible or were in discussion yesterday - but doing that consistently for 20,000 products per store with the same accuracy is just not feasible.
“The paradigm shift is when you get a ready-to-use decision so that in 99% of the cases there's no need to touch it. Humans add value—they have unique capabilities. They know something that is not in the database, they have talked to the store manager and the supplier. So humans can connect the dots and can see exceptional things, and that’s where they add value. The highest value of AI applications, on the other hand, is when you have many decisions to take and you have lots of data to support."
Q: What are the challenges of scaling AI in retail supply
A: “First of all, data needs to be available and reasonably clean. No one ever has the cleanest data and every interested party will always tell you, ‘well, our data is not good’, but in the end its usually good enough to work with. There’s small areas you can improve but I would say this is basically a solved problem.
“Next you need machinery that is able to collect data and understand what insights it is driving. For example, what customers will buy tomorrow, or how price changes will affect demand. There comes a problem with data—the unknown quantity I want to predict. We hear a lot about what algorithms you can use, or what methods you can use to train the AI, and this is where a lot of the noise and hype is.
“But this is still not the whole story. You might have a prediction, but then you need to make an effective decision and this needs to run at scale with operational reliability. When you place the backbone of store operations or supply chain operations on a compute system you can’t have downtime, so it must basically be ready to use and fail-safe.
“More than that, the organization needs to be ready to absorb the output of it. You need to change roles of your supply planners to connect departments that were siloed before so that all of this together makes for a really good AI product. You need a use case, you need data, you need the right prediction technology, and you need to understand what decisions are taken to operate at scale and adopt it within the organization.”
About the Author(s)
You May Also Like