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.
This is the first in a three-part series on supply chain use cases of AI/ML.
Introduction
Although we have been talking about AI in business applications and specifically supply chains for a few years, the application is still new and novel to many practitioners.
AI technology and techniques in supply chain technology are bringing opportunities for needed change in how we serve our customers and design, manage, and measure our supply chains. It may be that some of our guiding principles and approaches are a bit dated — maybe even obsolete. Surely, the last few years have put an exclamation point on that. Limitations in data and technology bounded past methods and therefore used to be the only practical way to measure and process data. But richer data sources and more advances in technology — and a lot of learning — have opened out vista to greater opportunity.
Stability was the name of the game in process, products, and data, controlled by policies enacted across trading networks and industries. And there is no getting around that it has helped make supply chains more efficient — if not more reliable. Because when we look at the last twenty years or so, rather than seeing a stable environment, we see constant change. And in the last decade, dramatically. So, why are some of us clinging to the stability game?
Maybe, then, it is not so bold to say obsolete because as planners, we know we have been working with inexactitude for a really long time. Outside the door, there are major forces at work: global warming, sustainable supply chains,1 and culture wars (based on demographics and gender, for example). Then there are various crises such as political and international divisions and trade wars, volatile economics, and the latest, the pandemic. Depending on our industry, these forces have made us confront our values and our management norms. When we talk to many executives, they admit this is the new normal and we won’t be going back. In our 24/7 global marketplace, we have needed to rethink some of our values.
Technology has not taken a back seat in dealing with these issues, providing solutions to enable more visibility, sustainability, compliance, fair trade, and responsiveness to change. Right now, AI and machine learning developers are taking a leadership role in helping the supply chain deal with all the uncertainty.
AI/machine learning, big data, IoT, clouds, last mile, and end-to-end visibility are all part of the technology mix. But we are only recently really learning and grappling with these technologies and capabilities. AI/ML queries are topping the list of what users want to hear about. They are learning that AI and machine learning are not some esoteric algorithmic black box applied to far-out demand problems, but are used to address day-to-day tasks.
Things have changed dramatically in our technology. And accompanying this is the dramatic change in our perspective about our world and our supply chains. Things are happening so fast and unpredictably that we do need methods to help us keep up and, yes, survive and thrive in such a dynamic world. Given all that, how are AI and machine learning helping? What are some of the applications and use cases that real users are doing today?
Practical and Visionary Use Cases
In spite of the great strides we made in mathematical systems over the decades, we still have many challenges with accurately planning, pricing, and achieving profitability. AI/ML algorithms and the big data they leverage can open the door to new possibilities.
AI/ML, though, is most often applied to fix a lot of pesky, pernicious issues in planners’ day-to-day work, allowing them to shift from manually fixing bad data to automatically cleansing data and constantly scanning the data for important issues that might impact plans.
Let’s look at a few of the new possibilities and the day-to-day use cases and see how AI/ML is helping upgrade our processes and tasks.
Pricing
Pricing is one of the diciest areas of the supply chain. Though often considered the domain of marketing and product planning, pricing has a huge impact on how supply chains are funded — determining budgets for production and logistics, and inventory investments. The supply chain, as well, is responsible for analyzing and implementing price changes from product launch through EOL. Price has its own lifecycle which is impacted cross-functionally, with different stakeholders having a greater say as the product’s life goes on. The issue is how can we manage this in an auditable way?2
Part of the problem is that data and methods reside all over the place, within our channels as well as within functional groups — product marketing, sales, supply chain, and finance. These groups often use different data sources, systems, formats, and metrics. Data may be locked up in spreadsheets or as web data, e.g., capturing dynamic pricing from various websites, and customer sentiment from social systems. That does not lead to the ability to create holistic, auditable plans.
How AI/Machine Learning Helps:
AI/machine learning can translate many of these sources and formats (structured and unstructured data) into a coherent format. Importantly, it can validate the quality and accuracy of these data and determine which are valid inputs to be included in future planning and execution tasks.
Pricing can become a lot more scientific and granular by assessing prices by many channels, customer demographics, product attributes, and so on, as well as demand over the life of the product. This accuracy can lead to greater sales and profit, and, we often hear, much better communication between the people involved in this work! On the web, with dynamic pricing, AI-based content search can help find and identify competitor pricing, and machine learning can find useful patterns in the data to help inform when these price dynamics could impact our pricing.
Promotions
Promotions are not just about pricing. The goal should also be to attract new customers, beat the competition, and, of course, move products. However, the lofty goals often get lost in the challenges of execution and in some hard facts about promotions — sometimes they just don’t work. That is, they impact pre-and post-sales, often reducing sales in their run-up and post-promotional periods,3 which reduces the value of the promotion. In creating promotions, companies consider a host of ideas such as price, product bundling, special promotional packaging and sizing, special ordering or fulfillment terms, channel-specific deals (trade promotions), and so on.
In addition, promotions cost money to execute, from building additional inventory to support the promotion, to channel coordination, advertising, special logistics, packaging and display costs, and so on. When you consider these factors, you quickly realize that the typical approach to measuring promotions may be limited. And we do know that promotions, though they may temporarily draw customers, tend to fail at any broader goal.4
How AI/Machine Learning Helps:
AI/ML analyzes past promotional performance across all the variable factors and then helps to develop insights to improve the currently planned promotion. Promotions, even in the same product category, are not the same. Other factors such as the product attributes, packing in a specific channel, target customer (and their lifestyle, demographic, and geography) and so on5 can now be included and compared to previous promotions. Or, as is often the case in new AI/machine learning implementations, determining which variables are important for your products or the markets you service and then setting up a data store to make these evaluations in the future.6
AI/machine learning can set parameters within the planning engine to find and include other data or ranges of the data, such as promotions of the past and any associated factors. For example: the weather the week the promotion ran, competitive move or countermove, or late execution. From there machine learning may select or modulate the algorithms that might be used for a specific promotion or product.
From these exercises, machine learning can evaluate the best planning algorithm at that point to fine-tune this specific promotion, drawing parallels and contradictions between this plan and what happened in similar circumstances in the past.
In Part Two of this series, we examine the use of AI/ML for demand planning, forecasting, and inventory management.
_________________________________________________________________________________
1 Greener products, fair trade — Return to article text above
2 Note: We will use the term “auditable” within this paper to mean consistent, compliant, repeatable, verifiable, and able to be used by multiple systems. — Return to article text above
3 When many users first systematically implement pricing and promotion systems that have all the relevant costs, they are shocked to find out they are consistently losing. Some organizations know this, but need to conform to their channels’ requirements. In both cases, learning to actually manage promotions to your benefit is the key. — Return to article text above
4 such as acquiring new customers who stay — Return to article text above
5 AI/machine learning could also include and rationalize cross-functional data such as logistics that support or impact promotion (extra warehouse or transport costs that dampen margin). — Return to article text above
6 Given the ability to capture the raw data and past plans, data scientists can sort through the history and apply learning exercises to see what methods and algorithms might have fit to achieve a different (better) result. Read: It’s All About The Data. — Return to article text above
To view other articles from this issue of the brief, click here.