This article is an excerpt from the report:
AI/Machine Learning for the Supply Chain — Practical and Visionary Use Cases
A copy of the full report can be downloaded here.
In Part Two of this series, we examined the use of AI/ML for demand planning, forecasting, and inventory management. Here in the third and final part of the series, we look at how AI/ML changes the way we get results from our supply chain systems.
Use Cases Abound in AI/Machine Learning
The above are just a few instances of supply-chain use cases and how AI/machine learning can help and is being deployed to address them. These are offered as examples to illustrate the ongoing issues of process, systems, and data obstacles that have persisted in spite of great system developments in the past.1 There are solutions to problems and also queries that have no past traditions beyond the flat file, fixed B2B data we have been processing for decades. In particular, a more systemic engagement with consumers is a new world and we are just beginning to enter into this realm.
There is a lot more to consider such as how users actually access solutions and what might be the differences — and there are — in implementing a machine learning/AI system.
Rethinking the Process of Getting Results from Systems
AI/machine learning is a rapidly evolving area within the supply chain planning world. Major solution providers are working to blend the new algorithms or are modifying tried and true methods of the past to fit new data opportunities with broader visibility and types of data, data rationalization, and curated data, and making the data accurate and relevant to their customers’ situations.2 They provide a flexible, highly dynamic platform environment where users have a range of options — from very hands-on to semi-autonomous.
This is unlike traditional systems implementations where we identified a needed capability such as a seasonal forecast, and selected some programs from a stock list of algorithms to run our forecast and inventory plan. Then we sought out and cleansed data sources, often manually. After some machinations in an implementation, we turned it on. Post implementation, our “questioning/learning” technology was based on perpetually writing new reports.
Machine learning and artificial intelligence stake out a different position — creating a different process — a process of evolution in learning and self-automating as we and the system get smarter.
The key to modern supply chain management systems implementation is to look at the data first, see the patterns that emerge, and then pick the right method and algorithms.
Often, algorithms are used in combinations to coax out the nuances in the data and create the right forecast. Through machine learning, the cycle of continuous learning, we can see buying patterns, supplier sources, production conditions, channels, and delivery methods change over time and we adjust methods as circumstances change.3
With added extendable computer power, we can apply that automated learning to every single product and understand what the individual disruptions, changes, and causals are at every single supply point and at every selling/consumption point.4 That change method is a best-fit approach: what fits the data — now.
From a systems implementation point of view, that means establishing a hierarchy of data stores — data lakes, databases, data warehouses — and, as we mentioned earlier, a richer data resource management toolset appropriate for an AI/ML empowered supply chain.5
Best fit is a dynamic capability that is applied at initial systems implementation and then becomes part of the overall operational capability within the planning system.
Best fit should be a standard that organizations rely on in their planning system, since there is a tremendous amount of variability in plans as products move along the lifecycle from design and evolving features/attributes, to sourcing, manufacturing (and ingredient or components), logistics, sales, customer use/valuation, realization, and maintenance, through end-of-life. Additionally, there are the associated costs, discounts, promotions and pricing, as well as the varying objectives of planning (revenue or supply chain operating). That is a mouthful, for sure. And users know it is really complex if you start to dive into how even one data element’s value can be influenced and can change over time.
At each stage, and under various conditions, demand or supply characteristics change. That is just too much for a human to handle if there are a lot of products and activities on one’s plate. However, this is where machine intelligence and learning really shine!
Conclusion: We Have to Face Change
Seeking stability was the name of the game in the supply chain for a long time. We used to say, “I don’t want too much nervousness in my supply chain.” This was knowing full well that change could happen at any time.
At some level, we were blissfully unaware, since we simply did not have as much information thrust upon us. Now that we do, we are forced to deal with the reality brought by all that visibility and detail, as well as with all the forces we discussed earlier.
With this knowledge, we need to modify our approach to supply chain management. Change won’t happen by piling on more people, and, anyway, it’s difficult to find and hire the required analytical types.6 In manufacturing, the high-labor models of the past are gone. The obvious conclusion is there will be more automation based on data science.
The path forward will require our full attention. It will require commitment to allow people to learn and grow. It will require investment. And with all this, as we have learned, there still will be some unseen twists and turns on the way. But, if change is a constant, why be a victim of it? Why not make it a positive change, taking us on the road to supply chain transformation, smarter solutions, and a better future?
1 For more examples, read: AI for Supply Chain Definitions — Return to article text above
2 Read: It’s All About the Data. — Return to article text above
3 Relying on what we learned many years ago as “the” forecast method may not fit now. And what we choose today may not fit next year. — Return to article text above
4 This data can be stored in data lakes and used as a reference point when evaluating the history of forecasts; and actuals, which would be the forecasting system’s database. — Return to article text above
5 The changes to the development/implementation systems process how they will impact supply chain teams can be read here: The Supply Chain Department of the Future. — Return to article text above
6 Read: The Emergence of the Supply Chain Scientist and The Data Scientists, Software Engineers and Data Managers. — Return to article text above
To view other articles from this issue of the brief, click here.