Skip to main content

In this episode, PlanningPosts talks with Rod Daugherty, VP of Product Strategy at Blue Ridge, about how AI and machine learning can be integrated into supply chain planning to drive value.

Episode 7 Show Notes

Todd: Let’s kick things off by telling us what is AI? And how has it made its way into the supply chain?

Rod: You hear a lot these days in the market about AI. Machine Learning techniques are a subset of AI sometimes called applied AI. The challenge for us is finding ways we can use AI in supply chain planning to enhance business results.

AI is prevalent in the consumer side of the market, and we all interact with artificial intelligence programs of some kind. For most of us, it is natural-language processing like Siri, Alexa or using spoken commands to make calls while in your car, for example.

We have found there is an opportunity to take the marriage of the most optimal demand planning and inventory optimization models to more intelligent levels that help organizations achieve better profitably and efficiency in your inventory investment.

Todd: Let’s set the stage a bit on artificial intelligence. Adding sophistication adds complexity, right?

Rod: Well to some degree, yes. But in all of our design and development efforts, the goal is to make the user experience as efficient as possible, while enabling the solution to do the hard optimization work. The same philosophy applies with AI. The software should do all the heavy lifting and present actionable recommendations to the user to improve the experience and make them more efficient at their job. Artificial Intelligence must add real, practical value.

Todd: Ok so how is AI different from traditional rules-based computer software processing?

Rod: Traditional rules-based technology expects input that runs through predefined parameters and models and produces an answer. In AI, the input is still data, but is also the desired objective. Along with getting to the answer or recommendation, AI / Machine Learning creates or evolves a new model based on its own feedback or learning loop. It is not just selecting a model based on rules. AI models score themselves and continuously make corrections to arrive at the optimal recommendation.

Todd: Got it. So let’s talk use cases. What are some areas where AI can add value in Supply Chain Planning?

Rod: We are now moving beyond the traditional demand planning methods that rely exclusively on demand history and beginning to incorporate outside-in, third party data.

An example of Applied AI that observes and learns is weather data. We are currently working on training models to understand and help predict how weather could impact sales for a distributor or retailer. We can:

  • Learn from observing user interaction with demand exceptions – detecting and monitoring – to automate & augment the human decision process
  • Learn from consumer behavior to drive targeted promotion and pricing recommendations – to anticipate and predict
  • Consumer search trends that lead to new product recommendations, which in turn improves the human decision process

This requires Neural Networks that can deal with classifying very large data sets which can only be accommodated through cloud-native demand planning software. The app has dynamic access to as much processing power as required for a particular analysis, which can combine and blend disparate data sources.

Todd: When we think AI, many of use thing about familiar AI apps like Cortana or Alexa. Could these be applied here?

Rod: Well Todd, yes. These apps can translate spoken commands and questions into action. One possible application would be using AI/Natural Language Processing that lets the user speak to the computer and get the analytics view they want.

  • NLP also has potential applications in supply chain planning for information retrieval and question answering.

(Privacy and appropriateness is tricky with NLP devices that have to “listen”.)

Todd: Where do you see the best opportunities to leverage AI and machine learning to benefit your business?

Rod: I’d definitely say it’s in the area of Inventory Optimization. Continuous learning that contributes to selecting and adjusting the most optimal forecasting and replenishment models for each SKU based on changing consumer purchasing patterns and multiple SKU attributes.

An example of this would be an Intelligent Optimizer solution that classifies SKUs using multiple attributes to tell a bigger, more complete picture of the forecasting and inventory optimization strategies. Knowing that bigger picture, you can blend inventory optimization strategies to produce the most optimal unified strategy for each SKU.

subscribe-to-planningpostsTodd: Ok, so by integrating AI and ML techniques, folks can get the right blend of forecasting and replenishment models to optimally address all types of SKU selling patterns and attribution?

Rod: Correct. One of those areas is demand classification. There are multiple SKU classifications defined by demand behavior and SKU attributes. Our customers face a myriad of events & complications when running their business. Inventory is their most expensive asset. So one of our primary goals is to continuously improve the inventory investment and user experience. Buyers, inventory analysts & demand planners recognize the problems associated with some of different SKUs classes, but it’s impossible for a person to accurately and consistently address strategy for every SKU.

Todd: You talk a lot about having a comprehensive inventory optimization strategy. Tell us more about that.

Rod: Traditionally, the industry has thought of demand management and purchasing or inventory optimization as separate things, and sometimes an over-simplified approach of one-size-fits-all combination. At some enterprises, it makes business sense to have a division of job responsibility, like separate demand planners and inventory analysts. But the solution should still account for the best combination of demand forecasting and inventory optimization strategies to provide the most optimal recommendations for each user role.

Todd: How does this change the way we plan?

Rod: Demand occurrences sometimes form patterns, and sometimes they do not. The illustration of continuous demand tells us that, while one year does not deviate too much from previous years, there is no discernible pattern. The Continuous Repetitive and Seasonal Selling patterns illustrate discernible patterns from one year to the next. The intermittent and lumpy illustrations are very difficult to peg from a pattern analysis point of view.

If you look at graphs representing year-over-year demand, you’ll see general classifications of the broad spectrum of demand which most companies experience to some degree.

 

 

Some have a much higher percentage of SKUs that tend to be weighted toward one side of the spectrum or the other. The reality is that one combination of forecast and replenishment optimization models does not optimally fit all SKUs. Different segments of SKUs based on demand patterns and SKU attribution require different strategies.

There are opportunities to fine-tune the pairing of the forecasting model, replenishment and inventory optimization strategies.

One area, in particular, is the wide spectrum of SKUs that have intermittent demand. Intermittent Demand Forecasting is a component of CSIS that has already been implemented in the supply chain planning solution.

There are a couple of bases you have to consider for SKUs with slow moving, intermittent demand… Is there enough demand to calculate a demand forecast (using a special intermittent demand forecast model)? And how much coverage do we need in terms of Days of Supply?

But, all segments of the products you purchase have an opportunity for improvement using Machine Learning techniques.

Todd: Let’s talk about the analysis part of the solution.

Rod: A core part of any good inventory optimization solution is the analysis; it takes into account selling patterns, profitability, volume and business attribution imperatives, like how important a SKU is to the overall assortment.

We’ve been implementing the analysis as we introduce each inventory optimization enabler. There is ongoing work to continue to expand the analysis of attributes and assignment of enablers.

Data internal to supply chain planning, in addition to obvious components like demand and SKU attributes, also include things like observing how humans address demand exceptions.

Todd: Can you give us some examples?

Potential external data could include things like:

  • Customer demographics, particularly important for retail
  • Purchasing patterns & fulfillment patterns, both of which require consumer-level transaction data
  • Weather correlation is also a possibility

Todd: How does demand planning fit in?

Rod: Demand planning is one of the main areas where we are implementing the use of AI and machine learning. In addition to advanced adaptive models and intermittent demand forecasting, two model families that use AI are:

1.Auto Regression: AR is a machine learning model that can provide better forecasting precision for certain segments of SKUs.

2.Neural Networks is an AI model that’s good at identifying non-linear patterns, like where there isn’t a direct relationship between the inputs such as demand, demand variance and the new forecast.

The key to machine learning models is the learning feedback loop, which enables the model to evolve as it learns from SKU patterns.

Todd: What about replenishment?

Rod: Yes; definitely a big piece of the puzzle here. In supply chain planning, we can use AI and machine learning for multiple replenishment optimization models:

  • EOQ is a sophisticated replenishment model, but for certain segments of SKUs particularly intermittent or lumpy demand, it is not the most optimal technique for inventory coverage.
  • Our new Intelligent Bootstrapping model can be a better model to account for the Days of Supply to carry for a given segment of SKUs (sale… 30 weeks no sales…)
  • Sometimes it makes sense to aggregate demand to the supplying echelon, forecast and order from there, then allocate to the customer-facing echelons like branches or stores from the DC — for example a Multi-Echelon Inventory Optimization solution
  • But the main point is that one model does not optimally fit every SKU!

If you are interested in learning more about any of these technologies, send us a quick note.

Related article:

AI in Supply Chain Planning and Demand Forecasting: Exploring some of the ways that AI can be integrated into a supply chain solution to future-proof demand forecasting and extend the reach of DOM.