Kristen Daihes and Ganesh Ramakrishna
August 3rd 2017
The world’s foremost thought leader in Artificial Intelligence, Andrew Ng, said that AI is the new electricity (Andrew Ng Interview). Electricity changed how the world operated… and AI is poised to have a similar impact. This transformation is already underway in the supply chain planning space and you should be fully versed in what it would take to allow AI to transform your business. This blog highlights a few key points you should understand, like feature engineering and the role systems play, to build a foundation that enables you to jump on board.
The Transformation Is Already Underway
AI is already transforming supply chain planning through:
3 Kinds of Machine Learning Solutions
Unsupervised learning is finding hidden patterns in data. Supervised learning is predicting a number (like a forecast) or a categorical outcome (like a stock-out). Semi-supervised learning is, if you provide a labelled response, an algorithm that can learn from the labelled set to make decisions on the unsupervised set. And then there is deep learning. Deep learning is another approach of machine learning that can use unsupervised or supervised learning, it’s just more intelligent dealing with more data and lots of machine power for calculation. Humans have no difficulty in identifying someone with a fake mustache. Machines could not do this until deep learning came along. Deep learning replicates human-like thinking. Think of traditional machine learning as current state with 100 parameters. Deep learning is next generation with much more data and processing power. It can have millions of parameters and will not overfit.
There are so many possible combinations of machine learning across functional planning areas. Take demand planning. Unsupervised machine learning can help us identify a cluster of SKU’s that have stable demand, but more volatility in error, and map back to a few less experienced planners – which can highlight more training needed. Supervised machine learning can allow us to use external macro indicators to create a forecast for a product forward. And deep learning can allow us to account for even more features to create a more powerful forecast.
The Importance of Feature Engineering
A feature is a piece of information that might be useful to your machine learning algorithm. For example, if you are building a model for predicting housing prices, some features might be school ratings, median household income, house size, plot shape, exterior and roofing material. The more features to give, the better the prediction model.
If you have great feature sets, you will have great results… you are giving the machine your intelligence. People say algorithms can’t replicate human intuition, but this is what the features are. You should understand this and translate this into your business. You need to find the features that matter to your business. You don’t need to argue about the 5 most important features – you need to find the 100s of features to consider and let the algorithms figure this out.
Companies are beginning to compete on their features. Our job is to design the algorithms and the feature sets. Creating, maintaining, and expanding your feature engineering – this is your EDGE. This is why we think about feature engineering as the IP for your planning organization. Just a few years ago businesses routinely published the features that were being modeled for Kaggle competitions, like a dataset from Walmart to predict sales. Flash forward a few years and things have changed dramatically. Companies are a lot more reluctant to reveal their features – sharing only the normalized values.
A Word About Systems
Many of you likely use SAP or Oracle, and machine learning can easily fit into your existing architecture. SAP HANA has the structure to do this, but it requires custom work, it is not off the shelf. Remember the importance of feature engineering. You can build a web front-end in the SAP environment. You can use core input data sitting in SAP. But the intermediate layer to build and test and put a solution together is all custom. We need to reset the language of custom and implementation. The words from a software domain do not exactly fit with what we do here. And beware of software vendors that will pitch an easy off-the-shelf solution. They may provide some incremental improvement, but the real game changer is in the customized route. And remember, machine learning solutions should complement, not replace your current landscape. They allow customization of the right features and operationalizing value complementing your current landscape.
AI is already playing a role in changing the competitive landscape in supply chain operations. Of course, there are a few companies like Amazon who seem to be out-innovating everyone when it comes to AI, but for the most part, those leading the pack have taken bold bets to experiment with AI in a few distinct areas. Those most at risk are the ones not yet taking the leap, planning to quickly follow. It’s those late adopters who will likely find themselves made irrelevant by the Silicon Valley startups. You don’t have to boil the ocean, pick a few areas to start experimenting to learn how AI can transform your supply chain planning organization – driving top and bottom line impact across your enterprise.