Back in late 2018 and early 2019, before the GENizah Artificial Idiocy craze began, the doctor did a sequence of AI Series (totalling 22 articles) on Spend Matters on AI in X Today, Tomorrow, and The Day After Tomorrow for Procurement, Sourcing, Sourcing Optimization, Supplier Discovery, and Supplier Management. All of which was implemented, about to be implemented, capable of being implemented, and most definitely not doable with, Gen-AI.
To make it abundantly clear that you don’t need Gen-AI for any advanced enterprise back-office (fin)tech, and that, in fact, you should never even consider it for advanced tech in these categories (because it cannot reason, cannot guarantee consistency, and confidence on the quality of its outputs can’t even be measured), we’re going to talk about all the advanced features enabled by Assisted and Augmented Intelligence that were (about to be) in development five years ago and are now (or should be) available in leading best of-breed systems. And we’re continuing with Supplier Management.
Unlike prior series, we’re identifying the sound, ML/AI technologies that are, or can, be used to implement the advanced capabilities that are currently found, or will soon be found, in Source to Pay technologies that are truly AI-enhanced. (Which, FYI, may not match one-to-one with what the doctor chronicled five years ago because, like time, tech marches on.)
Today we continue with AI-Enhanced Supplier Management that was in development “yesterday” when we wrote our first series five years ago but is now available in mature best of breed platforms for your Procurement success. (This article sort of corresponds with AI in Supplier Management Tomorrow Part I and Part II that were published in May, 2019 on Spend Matters.)
TODAY
Auto Profile Updates with Smart Information Selection
In our last article, we noted that in first, and many second, generation Supplier Management solutions, a supplier was always forced to create a profile by scratch, filling out a bevy of pre-defined form fields — even if they had all of that data in a well formed (metadata rich) xml or csv file. That’s why yesterday’s Supplier Management solutions contained functionality to auto-complete profiles wherever this data was easily available in standard formats.
But the biggest problem remained — supplier profile maintenance. A supplier profile is only accurate the second a supplier hits confirm/complete. Then, their main contact changed. They changed their mailing address. They moved HQ. They offered a new product. They dropped an old one. And so on. And, of course, they never maintained their profile, and you never verified it until you went to call, mail, or order and that person wasn’t there, the mail got returned, or the order was rejected (because the supplier no longer made the product). Then, you went to the website, found the new main line, called, navigated to the right person, got the right info, and maybe remembered to update the system.
So, as errors were discovered, some critical ones would be corrected, but most would remain unchanged or unnoticed and over the years errors — including information on critical insurance, regulatory approvals, and other key business requirements that put the organization at high risk if not verified — continued to pile up. After a few years, the record becomes more wrong than right. Not good.
So today’s solutions make use of the fact that information typically gets updated somewhere, even if not in the application. They monitor the supplier’s website for changes in contact information, invoices for address and product information, state and country registries for business information, and so on and when changes are detected, automatically update the supplier profile if the changes can be independently verified (through a third party authority, to prevent hacks or fraud from changing the system) or present the new data for approval to the relationship manager. All this takes is simple website and data source monitoring, scraping, reg-ex based pattern matching, and automated workflows. For complex information, a bit of semantic processing. Nothing beyond classical, proven, tried-and-true AI is needed.
Market Based Supplier Intelligence
Today’s supplier management platforms can integrate with multiple marketplaces, communities, partners, GPOs, and specialized compliance, sustainability, and risk data platforms, use rule-based transformations to harmonize all the data, and use built-in algorithms to extract intelligence at a market level.
Your company data gives you one view into a supplier; your vendor-based community, which is usually limited to similar companies in your industry that the vendor was able to sell, gives you another view; but the market gives you yet another view yet. Mathematically, one data point doesn’t tell you anything. If only nine other customers use the vendor and share their data through community intelligence, that gives you 10 data points, which gives you some data on the supplier’s performance and their performance for you relative to others, but 10 data points is not statistically significant. But if 30, 50, 100 data points can be collected from the market, that gives you deep insight with deep statistical significance.
On top of the data, and a few powerful cores (few, not a few thousand), all these platforms need is basic statistical calculations, trend analysis, classical machine learning, semantic processing, and sentiment analysis … all of which have been market ready for over a decade.
Real Time Relationship Monitoring
Relationships are more than just performing to a contract. They are about building a working arrangement that is beneficial to both parties. One where both are willing to admit problems, collaboratively explore potential solutions, and work together to achieve them. One where, when there are no problems, both are willing to find ways to improve.
As a result, relationship monitoring is more than just supplier performance monitoring. Especially since the relationship can be bad even when the performance is (still) (surprisingly) good, and the relationship can be (reported as) good when the performance is bad.
However, if you turn that semantic and sentiment analysis that was typically done on market data and public comments on internal communications, you can start to build up a picture of the overall viewpoint and sentiment on the relationship from both sides, what successes or issues are contributing to that, and if the situation is improving or deteriorating over time (by trending the number of spikes in communication with sentiment that is overly positive or negative). It’s not foolproof, as both sides could adopt strict, formal, communication no matter what, but since people are human, they tend to get hotheaded and lose tempers (and let the words fly) if they are really upset or jubilant when they are really happy (and let the praise fly), and while minor changes in relationship sentiment might not be caught (within tolerance), major changes will. Moreover, you’re not going to get rigid, controlled, strict, formal communication until threats of a lawsuit fly, but then it’s too late!
Automated Resolution Plan Creation, Monitoring, and Adjustment
Not only can supplier management platforms automatically detect issues (by rapid increases or decreases in trends or metrics), they can also correlate them to included resolution plan templates, automatically instantiate them and customize them to the issue in question, walk the supplier relationship manager through the resolution process, monitor progress, and automatically adjust the plan, and timeline, as needed as new information, good or bad, comes in.
Each default template can be correlated to a particular metric, trend, or sentiment driven situation, so selecting it is just a lookup. Instantiation is just filling in the blank with the appropriate category, product, service, and metric information, through reg-ex matching and search and replace. Robotic Process Automation (RPA) walks both sides through the process. Monitoring alerts either side when something is updated or not completed on time through more RPA. And adjustments can be made to trend lines based on average timelines on similar projects and current trends at each milestone.
Automated Risk Mitigation Strategy Identification
It’s one thing to detect risk, which is pretty easy along many dimensions when you have a lot of data at your disposal, and relatively straightforward to predict the likelihood of some risk events, but it’s a lot harder to determine which mitigation strategies should be employed when it looks like a risk is going to materialize.
But that doesn’t mean it can’t be done, or isn’t doable by the best of platforms. Just like a platform can come equipped with issue resolution plan templates, it can also come equip with standard risk mitigation strategies, which are essentially action plans to be automatically customized with the specific category, product/service, logistics, and supply line details. This is just pattern matching and semantic contextual awareness.
When all of this is combined with (near) real time monitoring across data sources, that are continually looking for relevant news sources, changes in metrics / prices / trends, etc, it’s like magic (although it isn’t). The platform detects risks, finds the most appropriate mitigations, and present it to the relationship manager. An all it uses is math, traditional machine learning, and traditional semantic/sentiment analysis. And, of course, a lot of up-front human intelligence (HI!) in the creation of this solution.
Automatic Real-Time Resource Re-Alignment
Corrective action plans and risk mitigation plans have something very important in common — people. People who create them, approve them, execute them, and monitor them. This requires resources to be constantly assigned, monitored, replaced as soon as they are unavailable or needed on more pressing assignments, and reassigned as the issue is resolved or the mitigation complete.
And while it will often be difficult for a project manager, or even a resource manager, to determine when to remove an organization’s best problem solver from a critical corrective action project to address a less critical risk mitigation project, or vice versa, even when the manager can’t think of someone else who could address the less critical risk mitigation project effectively, even when there is another moderately experienced problem solver that could step into the critical project, the software will be able to compute when that should happen if the organization defines the rules as to when that will happen based on hard metrics.
For example, if you define assignments to correlate resources to the projects with the highest cost (should the issue persist or the risk mitigate), and you define the cost of an issue based on its expected impact if unsolved, and the cost of a risk as its expected impact if unaddressed (using a fixed cost or a formula if those 10,000 processors don’t arrive and you have 10,000 vehicles you can’t complete), and you associate a seniority with each resource, it’s simply rank ordered matching.
If there aren’t enough resources for all problems, you can apply simple optimization to maximize the impact of your most senior resources. And, again, there is no Gen-AI needed!
SUMMARY
Now, we realize some of these descriptions, like yesterday’s, are also quite brief, but again, that’s because this is not entirely new tech, as the beginnings have been around for years, have been in development for a few years and discussed as “the future of” Procurement tech before Gen-AI hit the scene, and all of these capabilities are pretty straight-forward to understand. Moreover, if you want to dive deeper, the baseline requirements for most of these capabilities were described in depth in the doctor’s May 2019 articles on Spend Matters. The primary purpose of this article, as with the last, is to explain how more sophisticated versions of traditional ML methodologies could be implemented in unison with human intelligence (HI!) to create smarter Supplier Management applications that buyers can rely on with confidence.