In my first column covering Oracle's new Spend Classification product, I touched on a number of background elements about the solution. In today's post, I'll continue this examination, digging into how the tool works in practice (and in the final part of this series later in the week, I'll explore how it compares to other classification solutions in the market). Let me say upfront that I've not spoken to any Oracle references as yet -- none are available -- nor did I ask to try out the product on a sample data set. In other words, please consider this analysis more of an initial -- and hopefully informed reaction -- to the solution rather than an in-depth study of the product that is substantiated, in part, by discussions with current users. This is an important distinction.
During an initial walkthrough, Oracle made some fairly impressive claims regarding how the tool worked in early field trials. For example, they claim it was able to grind through 2 billion spend transactions in a 7 hour period, delivering an overall classification rate of close to 90% accuracy. And in cases where users were involved in assisting the tool as it learned, initial pas results have achieved 95% accuracy. These numbers, while impressive, are obtainable by other solutions in the market as well. But the combination of speed and initial pass accuracy that Oracle is claiming -- again, I've not validated this from customer's testimony -- is certainly impressive. Moreover, Oracle wants the tool to stand on its own and the product leads for the solution suggest they have no interest in "bundling in labor" to make it work. In their words, "we don't want to be in this business of services -- our goal is to enable the customer and potentially partners [to handle the spend classification process on their own]."
The demonstration of the product was relatively straightforward and it appeared to be well integrated into the user interface with the rest of the oracle spend and procurement suite. For example, the tool is plugged into Oracle's overall spend dash dashboard itself, which visually depicts high-level information about an organization's data and trends. The interface enables users to see and dive into previous batches of spend that they've classified. When it comes to classifying raw spend data, the system lets users set a confidence threshold level (e.g., 80%) and if the data passes this level in the system's view, it is then classified. If it misses the threshold, it remains unclassified.
Oracle Spend Classification enables users to classify to multiple taxonomies at the same time. For example, a user might opt to classify to a standard or customized version of UNSPSC while also adding the ability to examine their data through a different hierarchical structure such as an Oracle / MRP material code. In total, users can classify their spend using up to 5 different taxonomies out of the box based on the way the system is designed (but if users desire the ability to classify spend to additional taxonomies, it's possible to customize the tool to enable this requirement).
From a usage standpoint and the ability to undo a classification run, Oracle's engine never overwrites previous data. Users can create each classified data set anew each time they put the tool to work. Perhaps most interesting from a broader Spend Management perspective, the tool is integrated with the rest of Oracle's procurement applications. For example, if a front-line business user is in the process of ordering a new type of toner cartridge that's never been ordered before in the Oracle procurement suite, the system can do a call out to the classification tool to let users classify spend at the point of requisition. In this case, the tool would provide the user with recommendations based on percentage probabilities based on past classifications (though the user can override the recommendation).
Stay tuned for Part 3 of my analysis on Oracle's classification tool when I compare it to other solutions in the market based on what I've seen so far.