How to predict and solve supply chain problems before they happen
Predictive analytics and machine learning are helping supply chain managers become more savvy and proactive.
Sure, supply chain analytics are all the rage, but can they truly predict the future?
Yes, says Monte Zweben, CEO of Splice Machine — a San Francisco-based company that has created a platform for predictive applications that uses analytical processing and machine learning to improve over time. Online Predictive Processing (OLPP) can help manage operational processes and large-scale IoT infrastructures.
Predictive applications continuously monitor conditions to predict what could happen, and then enable a manager to proactively change plans accordingly, said Zweben, a computer scientist who has been working with artificial intelligence (AI) since starting at NASA around 30 years ago.
These applications are in a learn-predict-plan-and-act cycle, while continuously monitoring changing conditions and adapting to them. They use data to learn not only from the past, but also find patterns leading up to events, and then apply these patterns to predict an upcoming event.
"Predictive applications don’t stop at anticipating events; they enable you to be proactive," Zweben said. "They allow you to project what the world would be like if anticipated events occur, so that you can plan around them. They don’t just plan on a regular schedule; they can plan very quickly when they need to react to unexpected events. And these reactions are not limited to events that occurred, but also to events that are likely to occur."
In Frank Coker's 2014 book, Pulse: Understanding the Vital Signs of Your Business, he wrote, "Predictive models exploit patterns found in historical and transactional data to identify risks and opportunities. Models capture relationships among many factors to allow assessment of risk or potential associated with a particular set of conditions, guiding decision-making for candidate transactions."
E-commerce systems must instantly decide what to show customers. The best-in-class systems now use predictive applications to do this. The systems dynamically change their decisions in real-time based on the behavior of the customer.
Recommendations can change depending on how recently consumers visited a web page of a certain type, or whether they opened an email, or if they purchased something in a store.
As time goes on, recommendations get better and better based on reviewing and learning from previous recommendations. This closed-loop learning process learns to predict in real-time customer interactions.
A look into the crystal ball
Since it’s that time of year for predictions, here are Zweben’s thoughts on what to look for in 2018, based on the effect of predictive analytics.
- Using AI in supply chain management to predict shortages instead of buffering will save $10 billion in inventory costs
- Salespeople will demand real-time available-to-promise (ATP) systems even before orders are complete so they can promise delivery dates to customers quickly and reliably, reserving that inventory as orders are finalized
- Predictive supply chain management will emerge as a best practice, where machine learning is used to predict supply chain scenarios and planners can proactively respond to those expectations, avoiding shortages and late orders
- Streaming data, such as shop-floor data, weather data and fleet data will become commonplace and be used by supply chain systems to predict logistics and production glitches
One example of predictive analytics involves a retailer using inexpensive radio tags on SKUs to track the location of inventory. The analytics, sort of a supply chain “crystal ball,” can predict when an expected order would be late or have fewer goods than expected.
With the “crystal ball” the company can conduct a “what-if” analysis on each potential late order. It will discover, for example, what shortages would occur and which downstream consumers would be affected.
The predictive application predicts how late an order will likely be, and then simulates moving the order’s delivery date to the later time, Zweben said.
"As a result, consumers of the original order will have fewer items in inventory, possibly resulting in shortages," he said. "The predictive application’s dashboard calls out the possible shortages as well as highlighting the consumers of those items, providing a great early-warning system for planners."
The planners can then plan around these anticipated events by shipping goods from other locations, or at least warn consumers downstream so they are not surprised.
Out in the field
Quite often, the sales staff has to wait for orders to be completely processed before they can guarantee a delivery date to customers. It’s a competitive situation, and the customer may be looking at other suppliers.
"The sales person wants to look the customer in the eye and say, 'I can get your order 90% completed by X date' before the order goes through the entire process," Zweben said. "Using predictive analytics they have real-time visibility and know how much to promise."
One of Splice Machine’s customers is an electronics manufacturer that uses OLPP to take large orders and promise it in real-time to customers who are building out projects.
In another case, an oil and gas equipment manufacturer was able to predict when its customers’ field equipment might wear out or go down, creating outages for the rigs that can cost millions of dollars a day.
Predictive analytics, Zweben said, allows the manufacturer to order new parts and have them installed before any outage.
"They put the right product in the right place at the right time," he said.
Follow Barry Hochfelder on Twitter