Introduction
In our first instalment, we noted that the ambitious started pumping out 2025 prediction and trend articles in late November / early December, wanting to be ahead of the pack, even though there is rarely much value in these articles. First of all, and we say this with 25 years of experience in this space, the more they proclaim things will change … Secondly, the predictions all revolve around the same topics we’ve been talking about for almost two decades. In fact, if you dug up a Procurement predictions article for 2015, there’s a good chance 9 of the top 10 topic areas would be the same. (And see the links in our first article for two “future” series with about 3 dozen trends that are more or less as relevant now as they were then.)
In our last instalment, we continued our review of the 10 core predictions (and variants) that came out of our initial review of 71 “predictions” and “trends” across the first eight articles we found, in an effort to demonstrate that most of these aren’t ground-shattering, new, or, if they actually are, not going to happen because the more they proclaim things will change …
More specifically, we began our discussion of the 10th prediction … AI.
AI continued
We began our discussion by noting that this was the only prediction where the visionaries were not in synch, and that the predictions ranged from continued adoption to adaption to analytics enhancement to seamless integration to true advancement in underlying technology, and with the exception of continued, mostly unbridled, and definitely unresearched, adoption, they are more-or-less all off the mark.
As for adaption, most vendors don’t understand the technology they’ve embraced well enough to properly adapt it for Procurement needs, especially where Gen-AI is concerned. So “adaption” will be limited.
(Gen-AI is fundamentally good at only two tasks:
- summarizing large documents
- creating natural language responses to queries based upon a large data archive
If your task can’t be fundamentally reduced to one of those two tasks, then Gen-AI is NOT good for the task!)
As for analytics enhancement, a few of the smarter vendors who understand the true power of traditional AI solutions (based on ML and automated reasoning) will look for ways to use AI to enhance analytics for better results, which are easily obtainable given the increases in computing power and explosion in readily available (verified) (third party) data sources, and those that do will get better results across the spectrum of applications for predictive analytics in Procurement.
Seamless integration is a ways off. The current level of integration, especially around Gen-AI, is quite choppy and most of the results are (much) worse than not using it. We’ve spoken to a number of vendors who integrated Gen-AI since (potential) customers wouldn’t even speak to them unless they had it, only to hear that those same customers wouldn’t buy the solution unless they could “turn it off” (where it is the “Gen-AI” they insisted they needed). All of the “orchestration” vendors think Gen-AI chat-bot integration for Procurement is cool. But it’s not. For example, it currently takes up to ten times as long to use a Gen-AI chat-bot to complete a requisition in a well designed e-Procurement system than to use an expertly designed catalog.
Take a simple example where you want medical gloves. In Gen-AI, you’ll have a process something like the following when interacting with “Gormless”:
- Hey Gormless, I need some medical gloves.
- … 5 to 15 second wait while it processes …
- “OK Gary. I can help you with that. Do you want latex, polyvinyl, polyethylene, neoprene, cryogenic or surgical.”
- The same ones I always order you Gormless idiot. Nitrile!
- … 5 to 15 second wait while it processes …
- “Sorry Garry. I had those classified under dentistry. Do you want small, medium, or large.”
- The same ones I always order. Large!
- … 5 to 10 seconds while it processes …
- “OK, Got It. Now, do you want 50 packs, 100 packs, or 500 packs.”
- I want 1000, whatever packaging is cheapest.
- … 5 to 10 seconds while it processes …
- “The 50 packs are cheapest. Do you want 20 of those.”
- Cheapest per box? Or per unit?
- … 5 to 15 seconds while it processes …
- “I don’t understand Garry. The 50 packs are $10; The 100 packs are $18; The 500 packs are $85”.
- The 500 packs, Gormless.
- … 5 to 10 seconds while it processes …
- “Got it. Two 500 packs. Do you want same day shipping for $29.99 or next day for free?”
- Next Day is Fine.
- … 5 to 10 seconds while it processes …
- “Ok. You want two of the 500 packs of nitrile gloves, next day shipping. Shall I place the order?”
- Yes, Gormless. Place the f6cking order please!
- … 5 seconds while it processes …
- “The f6cking order has been placed. Your F6cking Purchase Order ID is 984567.”
In an integrated and properly indexed catalog with a traditional search bar and priority sorting based on order history and preferred suppliers, you will:
- open the catalog with a single click on the icon
- type “large nitrile gloves” in the search bar and press enter
- see, with pictures, images of all options in priority order, with the ones you always order first
- click on it, see you have 3 options, with an already calculated cost per unit
- select the “500 pack” option by entering “2” units next to it and pressing “buy it now”
You’re typing 3 words, 1 number, and clicking submit four times and you’re done in 15 seconds. Not 3 to 5 minutes of explaining your simple request to Gormless, the Artificial Idiot.
And this is just one example where trying to integrate state-of-the-art AI technology just to keep up with the hype train is making ProcureTech worse instead of better. So seamless integration is still quite a ways off!
What Should Happen? (But Won’t!)
1. Fuck Gen-AI.
As per our previous series, there are almost no valid uses for Gen-AI and very few valid uses for Gen-AI in Procurement. As we have indicated in previous posts, what Gen-AI is good for, and the only thing Gen-AI is good for, is massive text processing, summarization, and natural language query response generation to natural language queries. It’s only accurate with high probability, for non-critical decision support only, only when there is enough verified data for training. And then only of it is properly used by an expert who can identify when it makes a mistake (which it will do regularly). But any use that does not reduce to document processing and natural language response generation from natural language text blocks, in a manner where the response will be reviewed by a human before a decision is made (because accurate with high probability means it makes mistakes ALL THE TIME), is inappropriate.
2. Embrace Point/Function ML-based Predictive Analytics
With enough good, verified, numeric data, these algorithms, which have been researched, refined, and verified for decades, produce great results with high, known, confidence (compared to Gen-AI where the confidence is never known). (With enough data, the confidence can be 99%.+ Guaranteed. For many simple classification tasks, Gen-AI struggles to produce 70% accuracy. And that’s a good scenario!)
3. Embrace (Strategic Sourcing) Decision Optimization
As we’ve noted in previous entries, this technology has produced great results for almost 25 years, but yet the majority of organizations have not yet adopted it when it should be used, at least to generate a baseline, in every sourcing (and logistics) scenario! Moreover, it’s not just limited to cost optimization, it can also optimize carbon/GHG emmissions, delivery times, risk, or any combination of with the right data. It’s a value-generating life-saver for any organization.
Just remember. If you want true advancement, let us remind you that the majority of advancements in “AI” technology over the last seventy years (and you read that right, 70 years because “AI” is not new and has been under active research for over seven decades, with the first program generally considered to be “AI” written in 1956) has taken close to two decades to be ready for industrial use. Gen-AI still needs at least another decade (if not more) to reach the reliability we need to depend on it for critical use. Right now, as the disclaimers say, it can, and will, be wrong way more often than you think.
So while the focus on “AI” will continue, the focus should back off from experimental technologies unproven in Procurement when we have existing analytics, optimization, and ML-based predictive analytics that, in the right hands, with good data, can achieve results that many would organizations would consider a miracle. Leave the experimental stuff to the research labs and the creative teams, who aren’t impacted if what is generated is totally useless, as the creatives may still be able to use the useless garbage as inspiration.