Monthly Archives: October 2016

Just What Does Modern Sourcing Need?

Every year vendors, analysts, and even bloggers come up with their view of what next generation sourcing is, and what is going to get us there. This year, there is a big push towards AI (Artificial Intelligence) and not just predictive, but prescriptive analytics. Apparently, the Sourcing (and Procurement) of the future will be managed by computers, and not by experts. This is not only unnecessary, but a bit scary.

Why is it scary? Because computers run on algorithms and algorithms are not intelligence. Even though it’s theoretically possible for a sufficiently powerful computing platform to pass the Turing test*, all it does is point out the insufficiency of the Turing test for assessing the intelligence of a computer program. While computers can process significantly more data than we can, and modern predictive (trend) models, given enough data, can be more accurate than our intuition, they cannot detect when they are likely to fail or there is information or factors that need to be considered not baked into the model.

For example, if the prescriptive analytics relies on predictive analytics that relies on price trend modelling that simply takes into account price history, currency fluctuations, demand for related products or commodities, and demand for commodities that are usually used to hedge against price fluctuations in the product or commodity category, it will not detect when a natural disaster will result in a supply chain disruption that will result in less product or commodity availability in two months, which, of course, will have a drastic impact on price. As a result, the recommendation to spot-buy while the price is dropping is the wrong one, because as soon as supply drops, prices will skyrocket and it will be too late to lock in the current price.

But this isn’t the worst that can happen. If the AI that monitors multiple pricing trends, expiring contracts, and supplier performance not only ignores this blip but, instead, not only directs a resourcing for an expiring contract on an unrelated,but highly strategic, category, but encourages the inclusion of a supplier (that normally does not supply that category) that is currently in financial distress, the organization could end up blindly awarding a critical category (that is currently being served by a stable, reliable, reasonably low cost supplier) to a different supplier that is about to go bankrupt and then, seemingly without warning, stock-out on a critical product for months.

As you can probably guess, the doctor still believes that Sourcing and Procurement do not need AI and prescriptive analytics. What they really need are powerful and modifiable rules-based workflows, exception monitoring, suspicious transaction identification, and event monitoring.

The true power of a platform is to automate the tactical and streamline the strategic. Every minute of a professional’s time should be spent on strategic activities or issue resolution, not electronic paper pushing. Document matching, data collection and verification, contract monitoring, automated trend computation, etc. are all tasks that should be done automatically, but no actions with any strategic impact should be taken without intelligent human intervention.

A good rules-based workflow can allow tactical tasks, such as invoice matching and marketing monitoring, to be automated according to accepted rules and ensures that professionals only need to be involved when the parameters exceed the specified norms. Exception monitoring can insure that when something is out of expected norms, or exceeds ranges, or happens too often, it is immediately brought to the attention of the right individual. A suspicious transaction monitoring system, even if statistically based, minimizes the chances of duplicate payments, fraud, audit trail tampering, and so on. And event monitoring, even though it will produce a number of false positives, will enable a human to identify events that might impact the projected supply and cost trends for commodities and products purchased by the organization, and mark those for manual review and, possibly, re-sourcing if need be.

Modern sourcing does need better technology, but it doesn’t need artificial intelligence. It needs platforms that can help the sourcing professional focus appropriately, not guide the professional down a programmed path that will only give Sourcing and Procurement a false sense of security. The solution can track best practices for different situations, but the human still needs to determine if the system’s assessment is proper. Sourcing needs a system that empowers it with the intelligence it needs to make the right decisions, not a system that makes decisions and acts on those decisions (with automated contracts, orders, etc.) without human review and approval.

* A computer that is capable of sampling all conversations archived and currently taking place in real time, finding the one that best matches the conversation you are having, and providing that answer will provide a conversation indistinguishable from that provided from a real human, but it’s not intelligent. It merely proves the infinite monkey theorem.

The Coupa Factory … Reboot!

In Honour of Coupa’s Recent IPO …

Oompa Loompa Doom-pa-dee-do
We’re still building great product for you!
Oompa Loompa Doom-pa-dah-dee
Humble is hard when your work is marquis

What do you get when you go enterprise?
Encroaching upon the ERP course?
Where they are going terribly flat!
What do you think will come of that?

I don’t like the look of it

Oompa Loompa Doom-pa-dee-dar
If you are willing, you will go far
You will live in happiness too
Like the Oompa Loompa Doom-pa-dee-do

Oompa Loompa Doom-pa-dee-do
We have a great product for you
Oompa Loompa Doom-pa-dee-dee
Humble is hard when your work is marquis

ERP’s fine when your systems are old
They store all your data on drives that are cold
When you need real-time you’re hung out to dry
To watch the vultures circ’ling high

Up in the dark’ning skies

Oompa Loompa doom-pa-dee-dar
But now there’s Coupa, you can go far
You will buy in happiness too
Like the Oompa-Loompa doom-pa-dee-do

Oompa Loompa Doom-pa-dee-do
I have a great product for you
Oompa Loompa Doom-pa-dee-derd
If you are wise, you will heed our word

Ask who they’ll blame when your spend is off track?
Deep in the red and there’s no turning back
Contracts alone are never enough
If your software’s not up to snuff

And doesn’t track your stuff

Oompa Loompa doom-pa-dee-dar
But now there’s Coupa, you can go far
You will buy in happiness too
Like the Oompa-Loompa doom-pa-dee-do

Oompa Loompa Doom-pa-dee-do
I have a great product for you
Oompa Loompa Doom-pa-dee-dise
If you are wise, you’ll buy enterprise

Requisitions with a single mouse click
Budget checks integrated and quick
One click receipts when the order arrives
Invoice matching that always jives

You’ll have no … you’ll have no … you’ll have no regrets

Oompa Loompa doom-pa-dee-dar
But now there’s Coupa, you can go far
You will buy in happiness too
Like the Oompa-Loompa doom-pa-dee-do

Sixty Years Ago Today

Sixty years ago today, Fortran, possibly the first modern computer language, is shared with the coding community for the first time. Originally developed by IBM, Fortran is a general-purpose, imperative programming language that was designed for numeric computation and scientific computing that dominated science and engineering program for decades.

Updated significantly in FORTRAN II (procedural programming), FORTRAN III (inline assembly), FORTRAN IV (logical data types and statements), FORTRAN 66 (ANSI standard), FORTRAN 77 (structured programming and character-based data), Fortran 90 (array and modular programming), Fortran 95 (high performance), Fortran 2003 (object oriented programming), and Fortran 2008 (concurrent programming), this ancient language is still in use today. In fact, due to the continued widespread use in the scientific community, the next version of Fortran (currently dubbed Fortran 2015) is intended to be completed mid-2018.

What do you think, LOLCat?

Really? Why?

Spend360 – Applying Deep Machine Learning to Spend Analysis

Regular readers will know that, generally speaking, the doctor has not been impressed with the auto-classification and mapping offerings by any spend analysis vendor he’s ever blogged about as all have failed pitifully on tail spend, performed poor on any supplier or category the provider hasn’t processed extensively, and worked poor in new geographies and even poorer in foreign languages.

However, this year, he’s been impressed by two vendors with auto-classification. TAMR, which are trying to tame the data deluge, and now Spend360. While a new name on this side of the pond, it is not a new name across the pond, having opened its doors for business in 2011, after two plus years of intense development. Plus, it is gaining reputation pretty quickly since it’s foray to this side of the pond a couple of years ago and now has over 100 North American clients, which brings its total client base to over 400 global customers, which is impressive for any company in this space. (Even more impressive is the fact that, to date, it has processed over 1 Trillion of spend.)

While it’s still not perfect, and still can’t outmatch the best human expert with a multi-level priority mapping engine, it is decades ahead of its competition and has the ability to learn and evolve and, over time, approach 98%+ mapping accuracy, leaving little that has to be mapped, or corrected, by a human user (which is quite valuable when the user is not an expert in spend analysis but still wants to reap the benefits).

Not only can its deep machine learning identify tail spend suppliers, company specific categories, and even individual items coded in obscure ways, but it can learn over time and adapt to different data models, especially since it can use evolving knowledge bases. Whereas the majority of first generation classifiers used naive statistical classification that could not learn and had to map to a fixed (UNSPSC) model, Spend360’s uses deep machine learning (based on LSTM and encoder/decoder technology) that maps to custom data models using extensible knowledge bases (which can be created and maintained by the organization) that can encode organization and industry specific knowledge (and negate the need for custom mappings or override rules).

The fact that the knowledge base can be extended anytime a mis-classification occurs negates the need for manual mappings or override rules common in so many first generation spend analysis systems is a very powerful concept. It means that every erroneous mapping need only happen once and will never need to be manually corrected again. Plus, the fact that the data model can be extended as analytic needs evolve means that the platform can continue to deliver value year over year over year, unlike most first generation platforms that only delivered top N reports and failed to deliver value after the first twelve to eighteen months.

But this isn’t all Spend360 has to offer. In addition to a powerful classification ability, which can be trained to actually work, it also has a very powerful front end that allows the user to drill through the cube using custom filters in real time, compared to first generation systems that had fixed OLAP with limited filter capability. Reports can be cross-linked and all linked reports auto-update as one is drilled into. And data can be uploaded and incorporated into the cube in real-time if additional data is required.

And, to top it off, based on the 1 Trillion in spend they have classified over the years, Spend360 also has deep spend benchmarks across all of the major verticals and categories, which is often mapped down to UNSPSC level 4. This allows an organization to quickly understand how its spend on a category compares to the average in its vertical. Simply augmenting this data with pricing trend data can give an organization quick insight into where some significant cost normalization opportunities may lie.

In short, Spend360 is a provider the doctor expects you to be seeing a lot more of in the years to come, and recommends that you check out the upcoming deep dive, co-written with the prophet, over on Spend Matters Pro [membership required] if you are able. This is one best-of-breed provider you want to know.

3 More Terrible Reasons NOT to Use e-Procurement

Over on Procurement.World, the procurement dynamo gives us 3 Terrible Reasons NOT to Use e-Procurement, which, sadly, are still used by many organizations in the bottom 40% to 60%, to fight the implementation and adoption of e-Procurement systems.

If the reasons given in the procurement dynamo‘s post were the only reasons, that would be bad enough. But these are just a few of the reasons that Procurement organizations don’t use e-Procurement. In this post we are going to discuss other reasons, and, in particular, reasons that are a bit more believable — which are the worst kind of reasons.

1. Our Processes are Not Supported in the New System

While it’s true that the processes used by organizations that are still operating like it’s the last century are not supported out of the box, modern procurement platforms come with adjustable workflows that can be tailored to support just about any process the organization needs, good or bad. This may have been an excuse with first generation systems with fixed rules and workflows, but it’s not an excuse anymore.

2. The system won’t work with our current ERP or AP system

Most organizations require that all POs get in the ERP, all invoices in the AP, and all goods receipts in the inventory system. Because no recommended e-Procurement system will integrate with these systems out of the box, anyone against the implementation of such a system will insist it won’t work. And, again, wile this may have been an excuse with first generation systems that were almost impossible to integrate with anything, it’s not an excuse anymore when most e-Procurement vendors realize that their systems have to integrate with other systems and have published data models, open APIs, and middleware that enables the easy integration with such systems.

3. We don’t need Supply Management System X, we need Supply Management System Y.

Sometimes, knowing that a system they don’t want is inevitable, an opposing employee will suggest that a system is needed, but the system under consideration is not the right one and a totally different system is needed. For example, you are looking at a P2P and they will insist that a S2C is needed, or vice versa. Or they will insist that the ERP needs an upgrade. Or so on. But it will all be a distraction.

Systems will always be opposed, but when they are needed, they need to get implemented. The key is to select the right one. But with proper homework (and many posts on this blog will tell you how to do it), the right one can be selected.