Summary: Predictive analytics are increasingly important to Supply Chain Management making the process more accurate, reliable, and at reduced cost. To be at the top of your game as a supply chain manager you need to understand and utilize advanced predictive analytics.
As a large continuous process the Supply Chain has been extensively studied and is pretty well understood. It goes in well recognized steps from:
- In-Bound Logistics
- Parts Inventory
- Finished Goods Inventory
- Fulfillment (customer’s order to delivery)
- Out-Bound Logistics
As you can see these are not completely unique processes. Fulfillment for example could easily be understood to encompass all of finished goods inventory, order-to-deliver, and out-bound logistics. Manufacturing is understood by some to include all the process steps leading up to that point. But however you divide it, there is agreement that it is one continuous process and that a delay or failure at any point will ripple through the system and prevent efficient execution.
While these seven elements of the supply chain are each the focus of separate management activity, visibility over the entire supply chain is also a requirement. Particularly visibility into exceptions to the plan that might mean failure or delay.
Historically ‘visibility’ has been the key word, along with integration. In the past this meant actions taken based on observed events being closely linked with mitigating strategies up and down the chain.
For quite a long time BI and its ‘current-state’ and historical perspective served quite well. For example, using historical data we could determine that a part takes on average X days to arrive and even calculate standard deviations to make some fairly sophisticated adjustments in our procurement plan. Likewise on the demand side, we could look at historical demand data and try to extrapolate demand into the future, converting that to forecast production requirements and backwards into procurement and logistics requirements.
There are several different approaches to determining ‘supply chain maturity’ which, much like IT maturity assessments, show levels of achievement from basic through advanced. Advanced maturity typically meant having the features of lots of information sharing and even direct cooperation in product and parts design.
Increasingly though, a requirement of high maturity is the ability to better foresee the future, anticipate future events, and make optimal tradeoffs based on intentional strategic choices of top management. In short, to be at the top of the game in Supply Chain Management now requires including advanced predictive analytics.
Enter Predictive Analytics
From a predictive analytics perspective, about 90% of the problem is forecasting, starting with the demand forecast and letting that trickle back through the process to procurement and logistics planning.
There are long term forecasts that are more like broad risk assessments to try to evaluate whether our customers will continue to want our product (or will it become obsolete when they change to a wholly new solution). There is also the downwards view of this same question, are our suppliers sufficiently stable to be able to continue to provide critical resources that we need.
Of course the secret to good forecasting is to keep doing it over and over until you get it right. That is to say forecasts should be continuously updated and incorporate time frames that may be several years out (to anticipate the obsolescence issues), mid-term forecasts that drive our financial investments in plants and new products, and near term forecasts that drive actual production and procurement.
As you approach the near term forecast the details become more complex. While longer term forecast can be seen as fairly smooth, short term forecasts suitable for supply chain control need to take into account a whole host of smaller variables unique to each stage in the process. These may be weather or holiday related, allow for specific promotional campaigns, known changeovers in logistics or production, respond to anticipated increases or decreases in costs like freight, or other delays. In short, forecasts suitable for supply chain direct control are anything but simple.
If you’ve spent time in data science the one thing that should jump out at you is that across all the uses of data science, probably 80% of those have to do with predicting or influencing human behavior. What’s unique about supply chain analytics is its dependence on forecasting models.
So where modern predictive analytics begins to make inroads into supply chain management is typically in providing more accurate forecasts. This means testing any of a dozen mathematical forecasting models from ARIMA through dynamic multiple regression modeling to see which ones work best. There’s no pre-existing roadmap here, just test, retest, and finally select a champion method.
The other role for predictive analytics is contributing the mathematics of optimization. Optimization isn’t new, but any time there are two or more cost / benefit curves to compare, optimization techniques should be able to suggest the optimum tradeoff between the two, guided by whatever external business conditions you want to impose (least risk, highest ROCE, least capital intensive, fastest to achieve, and so on).
Visualization plays an important role especially at the production level. We want the models we create to be easily understood by workers at all levels. Visual displays like dashboards on tablets are increasingly a valuable medium for converting the large scale action into the specific tasks and needs of the person using it.
This means drilling down to the atomic level of activities, inventory, procurement, shipping units, and customer orders. Big data architectures are clearly a plus, but even more so the new Hybrid Transactions Analytic Platforms (HTAPs) like SAP HANA. These new platforms are completely in-memory, hold ‘big data’ volumes of data, and remarkably can process both transactional and analytic queries simultaneously. These new tools mean near zero delay in interpreting the inflow of customer orders, current inventory positions, and any manufacturing or external delays into near instantaneous updates to supply chains forecasts and plans at all levels of detail.
Appropriate Targets for Predictive Analytics in the Supply Chain
Here’s a high-level list of activities that could be improved with the application of predictive analytics:
Demand Analytics – How is my forecast tracking with actual sales.
- Detailed demand forecasting at the level of point of sale (store level, retailer, distribution channel roll-up)
- Deviation analysis of forecast versus actual at the SKU level.
- Forecast integration with promotional events and holidays to fine tune the forecast.
Impacts: forecast accuracy, in-store availability, lost sales.
Finished Inventory Optimization – What stock should I hold and where should I position it.
- Inventory budget optimization
- Safety stock level recommendations
- Segment inventory for tailored and customized fulfillment strategies by customer type.
Impacts: inventory cost, customer service levels.
Replenishment Planning Analytics – What, when, and where should I ship.
- Integrated planning at the retailer, distributor, and channel level.
- Optimize fulfillment logistics to account for handling, storage or warehouse constraints.
Impacts: in-store availability, customer service levels.
Network Planning and Optimization – Do I have the right network of manufacturing and warehousing facilities.
- Number of physical plants for manufacture and warehouse.
- Optimized flow paths to fulfill different segments of customer demand at the lowest total cost.
Impacts: Fixed and variable costs of operations.
Transportation Analytics – Optimizing transportation routes and loads including contract compliance.
- Optimizing routes including backhaul.
- Optimizing shipment schedules.
- Maintaining compliance with transportation contracts.
Impacts: freight costs, equipment utilization, contract compliance.
Procurement Analytics – How to achieve lowest landed cost and secure long-term high quality supplier partners.
- Scoring models for vendor quality, cost, and stability.
Impacts: total landed cost, security and quality of the supplier chain.
Predictive Analytics on the Factory Floor
While the factory floor is often not considered part of the supply chain, delays here can obviously impact the overall supply chain performance. At least one technique from predictive analytics is achieving wide acceptance and that is predictive maintenance.
In short, predictive maintenance utilizes different types of sensors on critical, capital intensive production machinery to detect breakdowns before they occur. This requires a data architecture based on streaming data (see our previous articles on Event Stream Processing – Who Needs It, and How Does It Work). The sensor data is initially analyzed by data scientists to prepare predictive models of different failure conditions. Those predictive models are then used to evaluate the incoming streaming data from the equipment and if a potential fault is detected, depending on the type, a message can be sent to the operator and maintenance staff, or an action can be created to immediately shutdown the machine to avoid damaging the capital asset and further disrupting production.
Geospatial Analytics in Network Planning and Optimization
In addition to the forecast and optimization mathematics that are standard for this task, we would add geo-spatial analytics. Several advanced analytic platforms now allow the addition of detailed location information that can be evaluated for travel time between nodes and especially against local customer density and time-of-day traffic patterns.
Strategy and Tactics
Supply chain improvements need to happen from both the bottom up and from the top down. Tackling one problem at a time, the bottom up approach, captures near term value. But the supply chain should also be a topic for strategic top down evaluation that’s guided by overall business goals. The data and insight that predictive analytics provides for both perspectives lets you address some of the really difficult questions with greater accuracy.
- How fast will the supply chain recover from external shocks?
- How do I plan for those external shocks and protect against them?
- Where are the biggest opportunities for additional profits from the supply chain?
- How can you protect margins when demand falls?
- How can you plan to protect profitability at the product level if a major supplier fails?
To be at the top of your game as a supply chain manager today, you need to understand and utilize advanced predictive analytics.
November 25, 2015
Bill Vorhies, President & Chief Data Scientist – Data-Magnum – © 2015, all rights reserved.
About the author: Bill Vorhies is President & Chief Data Scientist at Data-Magnum and has practiced as a data scientist and commercial predictive modeler since 2001. Bill is also Editorial Director for Data Science Central. He can be reached at:
Tags: supply chain