Artificial intelligence (AI) is the hottest technology topic right now, and it’s certainly going to be the most disruptive business trend going forward. To be sure, there are other technology topics that are also hot right now. Blockchain, for example, will certainly be disruptive, but no CPOs I know are clamoring for distributed ledgers as a must-have enabler to meet their goals.
But as big as AI will eventually be, it’s still overhyped right now, even within “cognitive procurement.” If I see one more graphic of a cyborg cognitively fingering a smart phone, I may just scream. (On a related note, this procurement AI avatar from Accenture is requisitely both terrifying and humorous.) Nearly all of the robotic process automation (RPA) implementations out there are delivering good old-fashioned labor automation stemming from glorified process scripting and screen scraping — and not a lot of “real AI.”
It’s easy to get jaded and assume that there’s not really much going on, that all of this AI stuff is not worth wasting a lot of time on. But that’s a mistake.
Much of the problem with AI is that it’s misunderstood as a technology area that’s somehow different and separate from day-to-day processes and technologies. Certainly, a deep learning system employing multilayer neural networks is not something that practitioners themselves will be spinning up anytime soon. That said, it’s important to understand generally how the technology works, what problems it’s solving out in the trenches and what types of providers can be used to do this.
I’ll be talking in-person about this topic at a free event next week in Atlanta. From a provider standpoint, most are at least evaluating AI within their technology stacks, and a few are actively pursuing it within product development. We would caution providers who are not seeing AI on their “Feature 500” lists to not ignore the area. Why? All of your users are valuable machine learning “supervisors” you can use to train your knowledge bases to improve usability, customer self-service, fuzzy search and, most important, analytics.
We are currently working on an AI in procurement research series that we’ll be releasing over the next few months. For now, though, check out this webcast that we did with KPMG where we spelled out 23 different procurement AI areas that are being deployed out in the field. In it, KPMG had a few nice graphics that overlaid various AI technologies onto different procurement business processes. For example, machine learning can be helpful in classifying spend-related text (including unstructured and semi-structured) to any target spend taxonomy within spend analysis for strategic sourcing, but it can also be used to perform real-time auto classification within an e-procurement “guided buying” scenario.
One of the key points we’ll be emphasizing in the forthcoming research is that AI system capabilities are about creating software-based analytics that augment the analysis used in our everyday thinking (i.e., cognition) and then scales it up collectively and in the cloud beyond what we can do individually. Whether that software code is implemented in our own neurons or within silicon bits and bytes, those analytics require mental/knowledge models and they require math/algorithms. Some of the models are fairly simple (e.g., text classification into a taxonomy) but some can get extremely complex — and valuable.
For example, sourcing optimization tools use some pretty heavy math, but these amplify our own cognitive horsepower to manage complexity that exists within the actual sourcing process that we could never do ourselves. It allows procurement organizations to create larger “unconstrained” market baskets that let suppliers bid flexibly to put their best foot forward — and allows procurement to facilitate a trade-off analysis discussion that elevates its own role beyond price reducers. The result: faster sourcing, more strategic sourcing, happier suppliers, happier stakeholders, more value unlocked and an elevation of procurement’s role to one akin to leading corporate strategy.
Some AI pundits might argue that a software tool is not “real AI” unless it uses machine learning and neural networks, but that’s silly. Some machine learning is downright simple. If you remember Bayesian statistics and multivariate regression from college, then you’ve got the gist of some of the basic spend classification approaches. But as the problem domains get more complicated, we’re seeing combinations of algorithms being employed to work together to solve different classes of problems.
In my next post, I’ll dive a little more into some AI areas within strategic sourcing (beyond spend auto-classification) that are emerging. I should probably delay it until next week, though, after I get back from a breakfast workshop in Atlanta that I’m doing next Tuesday with Dr. Alan Holland, the CEO of Keelvar, who actually has a PhD in AI but still has the ability to explain things in a way that mere mortals like me can understand. If you or anyone you know is free next Tuesday morning in Atlanta, I highly recommend attending — it’s free!
Stay tuned for my next post, where I’ll dive a little more into AI use cases in strategic sourcing and some of the work happening out there in these areas.
Get real time update about this post categories directly on your device, subscribe now.