I’m Afraid I Can’t Source That, Dave: Artificial Intelligence in Strategic Sourcing

Last week marked the 50^th anniversary of Stanley Kubrik’s “2001: A Space Odyssey.” One of the great scenes in this landmark movie is when astronauts Dave Bowman and Frank Poole chat “in private” about shutting down HAL 9000, the omnipresent, artificial intelligence-based control system. Unfortunately, HAL’s vision system is able to read Dave’s lips through a small window and learns the crew’s intentions. To protect itself and thereby its mission directive, HAL sets out to kill the astronauts.

Although the complex automated reasoning that HAL uses in its attempt to snuff out Bowman and Poole is a bit beyond current AI implementations, the domain-specific applications of machine learning are becoming quite real. AI-based lip readers, for example, have just surpassed human-based ones, and “1984”-class facial recognition is bringing “Minority Report” to the streets of China for social engineering at its finest.

While it’s easy to scoff at the hype built up around AI, it’s equally important to understand its components and figure out where businesses can realistically apply the technology.

For example, almost 20 years ago, there was a lot of hype surrounding web-based negotiation agents, which claimed they would not only scour the web to identify supplies but also select and negotiate with new partners on your behalf. Obviously, that didn’t pan out. It’s easy to carry that disillusionment into technology evaluation today, where even a tool like chatbots start to remind you of that annoying Microsoft paperclip:

All kidding aside, if you break down various processes used in strategic sourcing, there are numerous areas where AI can be deployed with increasing efficacy. One of the things I try to stress to business folks is to not get too hung up on the term AI and what constitutes “real AI.” In other words, does the technology in question consist only of unsupervised machine learning? Does it use any machine learning, even if simplistic and narrow? Can it include “expert systems” that use semantic modeling and rule bases? Is combinatorial optimization allowed into the AI club?

When I talked about this topic in my last post, I made my point by framing AI capabilities within the context of analytics. As you move up the analytics value evolution, you can see how AI supports the analysis within a strategic sourcing process:

• Descriptive analytics to turn data into information. What happened? Who spent how much with whom on what? How might AI help me auto-classify my massive data sets of messy spend data into my preferred spend taxonomy structure?
• Diagnostic analytics to turn information into backward-looking insights. Why did that happen? What is driving raw material prices up? Where is value leakage occurring? How might machine learning-based diagnostics find patterns in the data that help me pinpoint the root causes in my processes?
• Predictive analytics to turn information into forward-looking insights. What contract clauses are likely to create the largest business risk for me? How can I actually predict outputs like supplier bankruptcies? How can I forecast upstream commodity prices more accurately to do pro-forma product profitability planning that will then drive strategic sourcing scenarios? How can I do price forecasting and benchmarking with complex services that seemingly have no apples-to-apples comparisons? How can machine learning support these scenarios?
• Prescriptive analytics to turn insights into action. What is my best strategy to source against a particular set of requirements? Should I auction it and what format should I use? How should I construct my market basket? What scenarios should I run? How do I minimize risk and savings tradeoffs?
• Cognitive analytics that learn to interpret and reason why the predicting and prescribing is working or not — and then develop new models to increase demonstrated intelligence. Can the system analyze my upcoming bidding event; ask me questions about it; recommend RFI questions to gather data that can help me predict bidding behavior; determine an optimal a bidding strategy; and then refine any subsequent bidding rounds based on supplier feedback? Can reinforcement learning, game theory and behavioral psychology be applied to software agents/bots that can pursue reward-seeking actions on my behalf?

As you move up the analytics value stack, you begin to create higher levels of abstraction, using larger and more complex data models, more complex algorithms and lots more data to feed the system, so that the algorithms can learn from the data itself to improve the results and even uncover opportunities you hadn’t even thought of.

If analytics using AI capabilities can help you offload analyses out of your head and into software, then you can extricate yourself from lower levels in this stack to pursue other problems. For more sophisticated firms that have squeezed out basic value from the value chain, you need to start looking deeper and more broadly for opportunities. And if you want to pick the high-hanging fruit, you need to stop climbing trees and start using some automation.

After you’ve done some of the basic work on spend analysis and internal demand analysis, truly strategic sourcing is inherently an “outside-in” analytics exercise. The exciting aspect of AI in this market is all of the cloud-based tools that are honing in on specific problems and then using the power of dozens or hundreds of companies to train those systems to bring deeper levels of insights and analytical horsepower to identify opportunities (i.e., the opportunity identification step that you may have in your n-step sourcing process).

The analytics maturity levels outlined above provide some glimpses into how AI can help deliver analytics to the analysts — sorry, “data scientists” — within sourcing, from spend analytics, to market intelligence, to risk analytics, to sourcing event design and analysis, to price forecasting, to advanced contract analytics, to sourcing analytics and other use cases.