Category Archives: Technology

Gartner Inadvertently Makes the Case for NO AI in Supply Chains (which includes Source to Pay)

Gartner, which promotes the use of Generative AI in customer service, even though it did place Generative AI on the Peak of Inflated Expectations on the Hype Cycle for Emerging Technologies, just inadvertently made the best case for never, ever, ever using AI anywhere in the supply chain, including Source-to-Pay, and we love it!

In a press release on their newsroom in late September, where Gartner Says 80% of Supply Chain Not Accounted for in Current Digital Decision Models, the subheading clearly stated that Digital-to-Reality Gap Shows Current Technology Use Fails to Improve Outcomes for Supply Chain Decision Makers.

As a result of this “digital-to-reality” gap, Gartner’s research, based on an analysis of 600 survey responses of supply chain decision makers, not only found that current use of digital models to analyze trade-offs made no meaningful impact on the rate of good decision outcomes but actually found that slightly more bad decisions were made with the use of digital tradeoff analysis than without and marginally increased the percentage of bad decision outcomes. Moreover, More than half of supply chain leaders reliant on digital technology to make a recent strategic decision told us that they felt they would have landed on better decision outcomes without the use of their models, and our analysis suggests that they are correct.

In other words, if source-to-pay and supply-chain decision makers cannot even make decisions when relying on traditional, focussed, machine learning and modelling technology, there’s no chance an unpredictable probabilistic incarnation of Artificial Idiocy that randomly changes its output by the millisecond is going to make good decisions. And the reason is the same — just like traditional (guided) (machine learning) models require good data and a digital representation that covers the majority (if not the entirety) of the process and relevant variables, so do Generative AI models and, in just about every organization on the planet, this necessary digital representation DOES NOT EXIST!

As a result, applying AI without the data it needs to have even a snowball’s chance in h3ll to make a decision is pretty much guaranteed to lead you to worse decisions than you, or any other intelligent human with a decent understanding of the situation, will make without the use of any technology whatsoever.

You don’t need AI, you need end to end process modelling, data collection, data enrichment, data validation, and the ability to use those end-to-end digital tools, interpret the data and recommendations, and make good decisions off of that. And since, with the current rate of digitization, it’s unlikely the majority of organizations will go from 20% supply chain digitization to 80% supply chain digitization (which is the minimum level of digitization you should have before even considering any AI, even for inconsequential decisions) by the end of the next decade, you should not even have AI for decision making on your future roadmap before the next decade rolls around.

the doctor doesn’t say this often, but thank you, Gartner. (Because it really is the case that stupid is as stupid does.)

ERP at the Center of Sustainability and Human Impact?

ERP Today recently ran a brief editorial insight entitled ERP at the Center of Sustainability and Human Impact which caught my eye because ERP is generally not at the center of anything that is not manufacturing but yet should be at the center of sustainability data because it’s the ONE system that should be accessed, or at least be accessible, organization wide. However, in most organizations, all it stores is the manufacturing / order data, purchase orders, and invoices.

The article states that, within some organizations, they are providing the financial clarity to drive meaningful environmental and human impacts, however it only lists TWO (2) (Blue Marine Foundation and Oracle), and the doctor‘s experience, which is similar to other analysts he’s worked with, is that, for the vast majority of companies, this is JUST not happening.

Why? A few reasons, but the main ones are:

  • most ERPs don’t store complete financials; they’ll store POs and Inventory, but the complete financials will be in the organization’s AP/I2P/P2P systems
  • most ERP’s don’t store/calculate ANY sustainability data and
  • most ERP’s weren’t/aren’t configured to store ANY sustainability data

This means that, for an ERP system to provide financial clarity around meaningful environmental and human impacts, an organization needs to

  • integrate it’s accounting systems with the ERP and push all invoices and payments into the ERP
  • get subscriptions to third parties with the sustainability data and push that into the ERP after
  • updating the ERP configuration to store all of the relevant data around sustainability and responsibility that the organization wants to track

And while this will be doable with most modern ERPs, it could be expensive and force an organization to use another platform, such as a modern SRM (Supplier Relationship Management) platform as its core sustainability and responsibility platform instead. But it would be nice if the ERP could be the one platform that at least stores all of the organization’s golden records, because data warehouse, lakes, and lakehouses aren’t the answer (as all they do is duplicate data and make it harder to find the single source of truth) — the answer is a central source of sustainability and responsibility data that is, or could be, accessible organization wide so everyone can know the impacts of their (financial/supply) decisions. And while it could be the ERP, given the sheer cost of any customization work on any of the big ERPs, the doctor doesn’t think it’s very likely.

The Seven Patterns of Artificial Idiocy … in Procurement

AI proponents, who keep telling us it stands for Artificial Intelligence which does not exist, keep pushing the benefits of AI while sweeping all of the detriments under the rug so you’ll sign that multi-year deal now (and they’ll have the money to keep researching the technology in the hopes that their continued efforts will prevent the bad from happening again). (And while the tech will get better and the success rates will improve, the very nature of the technology they are deploying is such that it is impossible to prevent the bad because the technology is not intelligent and not deterministic, as per our many previous posts on the subject. The best will eventually get success rates up to 99%, but that’s still a 1% failure rate and, in Procurement, it only takes one catastrophic failure to wipe out all of the successes made in the rest of the year. It only takes one bad decision that shuts down a multi-million production line, results in class action lawsuits for the release of unsafe products in to the market, or results in seizures and destruction of millions of dollars of inventory when the products violate import restrictions, etc. to deliver masses losses.)

And the benefits they push typically fall into one of the seven patterns of AI that the AI proponents keep telling us AI will deliver. To help you better classify the false promises, we’ve decided to cover the seven patterns, example promises, and realities you will encounter if you implement current iterations of technology that employ Artificial Idiocy.

HyperPersonalization
Promise: The system will adapt and evolve over time so that when you log in, you see exactly what you need to see, in priority order for you.
Reality: The system comes with a set of widgets, and all the AI does is reorder the alerts/notifications/tasks/reporting views in each widget based on a priority weighting where the weights are recalculated based on recency in access so, at the end of the day, it just keeps showing you what you just looked at and truly important alerts, because you haven’t regularly looked at them, are at the bottom of the widget and off the screen since you never scroll down inside the widget. Classic rule based systems work better.

Recognition
Promise: 100% effective automated invoice processing, routing, and approval
Reality: it only recognizes invoices from suppliers who invoice regularly in a format that never changes; and only if the line item descriptions are never abbreviated; and only matches properly if the PO number is included; 10%+ of invoices have to be manually processed, and abbreviation errors cause misclassifications of units that sometimes don’t get caught until after payment is issued

Conversation and Human Interaction
Promise: natural language interaction, including voice to text
Reality: due to the ambiguous nature of the English language, the number of follow on questions the AI has to ask just to produce a simple report requires the user to spend five minutes giving clarifications and specifying parameters that could be point-and-click selected in 30 seconds; efficiency is flushed down the virtual toilet

Predictive Analytics & Decisions
Promise: what the price will be when, and why
Reality: works really well 95% of the time, with price accuracy often within 2% to 3% and demand predictions (outside F&B and other highly unpredictable industries) within 3% to 5% at the macro/rollup level, but when a trader illegally tries to run up the market with excessive trades, and prices start to skyrocket and the algorithm doesn’t know this is unusual/short-term, it may predict extreme price increases at contract expiry in 6 months, and automatically early renew the contract for you at rates 30% to 130% higher than it should (before the costs become unaffordable)

Goal-Driven Systems
Promise: Sustainable Buys with Cost Savings
Reality: Unsustainable buys as cost is overweighted and over-prioritized in all situations; ’nuff said

Autonomous Systems
Promise: They will procure automatically and do better than humans
Reality: They procure automatically and occasionally do better than humans, usually do on par, and occasionally make such disastrous decisions that the company does well to avoid bankruptcy …

Patterns & Anomalies
Promise: They will detect unusual spending patterns and detect the best opportunities for savings and the most likely instances of fraud
Reality: Unusual spending patterns don’t mean savings, and usual spending patterns don’t mean absence of fraud, and you get all kinds of “priority alerts” with no savings opportunities while the largest opportunities go unidentified and collusion frauds are never detected

At the end of the day, as we’ve said again and again, Procurement Automation: Good, (AI) Automated Procurement: Bad. Only you, dear reader, are intelligent and, thus, only you should do the thinking and only use technology for what it’s good at, the thunking.

Procurement Automation: Good. Automated Procurement: Bad.

We shouldn’t have to say this. It should be very clear by now. But given that a number of vendors are using the terminology interchangeably, possibly to convince you they have the right solution, maybe it’s not clear. But it needs to be. Because procurement automation is NOT the same as automated procurement and while procurement automation, properly done, is the best investment an average over-burdened and under-resourced Procurement department can make, on the flip side, AI-driven automated procurement is the absolute worst. To put things in perspective, downgrading Excel to Lotus 1-2-3 would be a better move. But let’s back up, and start with some definitions.

Procurement Automation is the process of automating certain procurement tasks that can be best accomplished by machines and procurement automation technology is the technology that automates the tasks that can be best done by machines. In simpler terms, it automates the “thunking” by doing all of the tactical, almost mindless, work that is a waste of a senior Procurement professional’s time.

The Source-to-Pay cycle is full of tasks that are best done by machines when appropriate rules and boundaries are defined. For each major area, we’ll outline some of these tasks as an example.

Intake/Orchestration

Procurement Automation will analyze the request, identify similar requests made in the past, identify the actions used to resolve those requests, identify the suppliers considered and selected, the products and services used, and other information. It will present that information to the buyer, including the suggested actions, and allow the buyer to one-click initiate any of the suggested actions, which might include a sourcing event, contract renegotiation, catalog purchase, etc.

Sourcing

Procurement Automation will, when a user kicks off a sourcing event for one or more products, automatically bring up the suggested suppliers, automatically suggest the appropriate questionaries and forms, automatically suggest the appropriate Ts and Cs to insist on up front, automatically send the RFP to suppliers, automatically analyze the responses to make sure they are complete, in the correct format, and in an expected range; automatically compare the responses to find deviations from the norm; automatically highly the lowest and highest costs, CO2 factors, etc. and present all that information to the buyer.

Supplier Management

Procurement Automation will, when a supplier is selected, automatically handle the onboarding; monitor the data for changes; monitor the performance metrics; monitor the OTD; monitor third party financial and risk metrics; and alert the buyer to any issues and performance changes that are detrimental or may indicate forthcoming problems.

Contract Management

Procurement Automation will, when an award is selected, push the award into the Contract Management system, automatically generate the draft contract, send it to the supplier, highlight any redlines the supplier makes when it comes back and automatically inform the supplier if any non-negotiable terms and conditions (including those they agreed to when they responded to the RFP), and automate the generation of the response email when the buyer does their redlines.

e-Procurement

For catalog buys, it will automatically generate the POs, route them for necessary approvals, distribute them to the suppliers when approved, automatically match the ASNs when they come back, alert the buyers if ASNs are not received in a timely basis, and match the invoices when they come in.

Invoice-to-Pay

When the invoice comes in, it’s automatically matched to the purchase order, it’s checked for price accuracy, identified as partial or full, verified to be non-duplicate, and if any checks fail, it’s bounced back to the supplier with a description of the issues and a request for correction and resubmission. If the resubmission deals with the problems, it’s queued waiting for goods receipt/confirmation if not present, or matched if present. If the match is made, then it’s automatically sent down the approval chain, and if it’s not made within a certain time period, an alert is raised.

In all cases, it’s automating the tactical tasks that don’t require any decision making and only involving the human when necessary.

In contrast, Automated Procurement is the process by where entire procurement processes are handed over to the machine to fulfill instead of the human. In other words, when an intake request comes in and the buyer marks it for sourcing, an Automated Procurement solution will handle the entire event up to and including the award and auto-generate and distribute the Purchase Order(s). The buyer is completely bypassed and the right inventory showing up at the right time at the right price is left entirely up to the machine. Sounds good in theory. Looks good in practice when it actually works, which it will some of the time. But grinds the company to a halt when it fails.

A machine that pursues lowest cost will select an unproven non-incumbent supplier for a critical part when the suppler, who has not supplied that particular part to the company before, outbids the incumbent. It will not detect that the bid was made in an desperate attempt to help the financially struggling supplier stay in business, that the bid is not sustainable, and that the supplier is not capable of producing the part at the indicated level of quality. Then, when the first shipment is mostly defective, and the promised rush replacement order never arrives because the supplier goes out of business, the production line for the 75K luxury car folds all for lack of a single control chip. (A similar situation has occurred in the past. Recently, chip shortages stopped Cherokee production in 2021, and that wasn’t the first occurrence. Or even the second, or third.)

Machines are not intelligent. Not even close. And expecting them to make a good decision every time with no logic whatsoever (as modern Artificial Idiocy algorithms just stack probabilistic equations on top of probabilistic equations almost ad infinitum) is lunacy. So while you should invest in the best Procurement Automation tech you can get your hands on, you should steer clear of any and all Automated Procurement Solutions those fancy new startups try to sell you. While those solutions may work 90% of the time, that last 10% of the time, they won’t work that great. And, in particular, that last 1% of the time they will fail so miserable that the disruptions and losses that result will more than cancel out any and all savings and efficiencies you might get from the 90% of the time the tech worked in the beginning.

The 1-Step Guide to Responsible AI in Procurement

Forbes recently published an article on Responsible AI Procurement: A Practical Guide For Selecting Trustworthy AI Vendors. It wasn’t bad, but it missed the point.

Today, there’s only one way to responsibly address AI in Procurement.

JUST SAY NO!

1) We don’t really understand proper AI Governance (especially when most vendors are using third parties which are illegally scarping content, not checking for bias, and tweaking models on the fly without consideration for the new problems the on-the-fly tweaks will cause).

Plus, it’s not just ethical codes of conduct, it’s agreeing on what the ethics are, and, most importantly, making sure the models are transparent and unbiased — but we don’t know how to do that today, especially since all these models are huge black box models.

2) You can demand all the evidence you want from the vendor as backup for the vendor claims, but if you can’t verify it, how can you trust it?

3) These models require huge datasets to train. Even if you know the data set used and the processing method used, how can you be sure every element was properly vetted? Just like one bad apple can spoil the bunch, just one bad element in a clustering or optimization model can spoil the entire model. Just one!  It only takes a small amount of bad data to spoil a model, regardless of the model used.

4) These models can fail, and sometimes fail spectacularly. If you don’t understand the model, you don’t understand where it can fail, and thus what to look for. Also, many minor incidents (which can foretell future catastrophic failures) will go unnoticed if a human isn’t checking everything.

5) These models are not secure … the AI can leak any training data at any time without warning. Your vendor can have every security certification under the sun, and all will be for naught if they use LLMs.

So, JUST SAY NO!