In Part I, after noting that I’ve been hearing and seeing the following too often lately:

• smaller vendors struggling to close/losing deals because the bigger/splashier vendors have more “Bells and Whistles”
• vendors getting lots of (and possibly too much) funding focusing heavy on bells and whistles

I said that we were going to dive into some of the most common bells and whistles and why, in the best case, they’re a complete waste of money and, in the worst case, the thorn prick will end up being so painful that your team will simply stop using the software. (And often do so before the subscription is half, or even a third, up, which leaves you in an expensive subscription jail you cannot stop paying for even though no one is using it.)

We started by diving into Intake before addressing the general situations of Flashy UX and Adaptive AI-Driven Automation which are permeating the space for little to no value whatsoever. Today we continue discussing the bells and whistles in Source to Pay that offer little to no value.

RFX: Generation

This sounds super cool and super helpful, except, in all but one case I’ve seen so far, it’s not. The way most of these products are implemented is they ask you a few questions, create a very detailed prompt, feed it to a generic LLM, and pull back what looks like a good RFP because it’s reasonably long and really well written (compared to the average RFP written by a junior buyer who has written anything without the aid of an LLM in 5 years).

Except what it brings back is an RFP that is written to the LOWEST COMMON DENOMINATOR, not average, and definitely not the wisdom of crowds because these LLMs are trained on whatever data can be scraped from the internet, and the data there is the most of becomes the data that influences the LLM the most, and that data is the lowest common denominator data. So, you get a rather generic RFP with little to no useful specifics, which means that the RFP is okay for indirect and standard services at best, and poor for direct and complex services, if it’s even usable.

RFX: Response

You have some vendors that are building LLM solutions to help vendors build RFX responses based on their data (pretty good) and standard responses, which is not so good because, again, it’s lowest common denominator responses with absolutely no unique capabilities conveyed or any unique thought. You want a vendor who took the time to review your RFP, understand it, and come up with a thoughtful, customized, valuable solution to your problem — not a vendor who trusted an LLM to create a winning response.

Supplier Discovery

Quite a few vendors are now integrating AI, and Gen-AI in particular, into their supplier directories, networks, and internet search and promising quick, easy, powerful supplier search for any need you can think of. A few of them, primarily the industry specific solutions, work pretty good, but the majority of them give you results that are no better than a random Google search. This is because, without deep supplier profiles and deeper product and service files, you can’t do good supplier discovery.

Automated Supplier Risk Profiles

Now, this sounds really useful, and if they were reliable, they would be. But most of these platforms are pulling in third party rating data, computing risk statistics off of incomplete web data (supplier site, articles, social media sites) based on LLM/statistical processing that’s not fully accurate. This results in both false positives and false negatives. False positives because the incomplete data results in a poor reliability or performance or brand score, which isn’t too big a deal if you think you might need the supplier, dive in, identify the issues, provide the missing data, and fix the scores. (But if the suppliers who are the false positives are the best options and you overlook them because you choose to focus on the suppliers who were all green, and only middle of the road, then you end up with mediocre suppliers when you could have had good ones.)

But if the suppliers turn up green, and you trust the scores because the handful of green scores you verified when the system went live were reasonable, and you select a supplier that was on the verge of bankruptcy (because the financial ratings agencies didn’t have the data from recent events because the supplier was private) who has been turning out poor quality products over the last few months, you could end up with a supply shortage on short notice followed by a lot of returns and demands for replacements you can’t provide — and that’s not a good thing.

The best solutions are semi-automated, force the buyers to build the right profiles and metrics, compute data confidence, and force buyers to verify data of low confidence. They don’t promise miracle ratings, they promise reliable solutions.

In other words, as we said in the last part, bells and whistles look and sound cool, until you realize that the ringing can be deafening to the point it gives you such a headache that you can’t get your work done.

Finally, because we want to make it extra clear how useless bells and whistles are, we are going to continue this series and address a few major areas of Procure to Pay as well.