Friday Rant: Most E-Sourcing Optimization Stinks

One of the biggest shams in the e-sourcing world in the past few years has been software-driven optimization. Don't get me wrong -- the general concept of optimization is extremely powerful. And as executed as a service from such companies as CombineNet, Trade Extensions, Emptoris and BravoSolution, it can deliver an exceptionally unique value across complicated categories. Even in software, from providers like CombineNet and Trade Extensions (the later of which has more experience delivering stand-alone software capability), it can live up to its promises. But most off-the-shelf sourcing optimization from the big name players (and some of the smaller names) is marred by similar -- pardon the pun -- constraints.

Perhaps chief among these is the limited ability to define constraints based on what is relevant to a specific market versus what the developers engineered around the CPLEX solver. For example, most off the shelf sourcing-suite based optimization can support the following scenario in a single event (versus having to run multiple markets): on the front end, have global suppliers submit different price quantities across thousands of line items and and price points for different ship-to locations both FOB and CIF. They might even enable users to also break out raw material costs from value-added components (bid both ways). But it is unlikely that in addition to the above examples, that they would also alllow suppliers to submit alternatives that might be more competitive (e.g., producing at night to take advantage of cheaper utility rates) while also showing an impact on a production schedule. To further complicate this example, it's also unlikely that they could also show prices locked for 6 and 12 months in addition to hitting all of the above requirements.

On the back-end of the bid, run scenarios that encompass: best total landed cost (with split of business requirements as dictated by individual BUs or facilities), performance/risk adjusted scores (based on internally derived metrics/system data) and offset or diverse supplier requirements. And solve these scenarios in less time than it takes to run to Starbucks. Again, unlikely most off the shelf software can accomplish this.

Now, this combined example is no doubt extreme. But hopefully those in the audience who've tried out an e-sourcing optimization package will catch my drift and it will be educational for others considering using optimization. My personal takeaway from trying out this stuff myself and writing MRDs for a failed optimization product years ago -- not to mention interviewing a number of users today --- is that nearly all off-the-shelf software-based optimization not only falls apart when you add just a few elements of complexity versus what it was designed for -- it also requires that you kludge together an event-based solution such as running multiple events or copying events to do the analysis in a way that could be described as anything but elegant. Solve times can also increase beyond acceptable limits as well.

So who wins in all of this? Two groups: service providers (who can do this for you). And ILOG, of course, who gets license revenue on their solver. So if you're serious about making optimization work, find a service provider you like to work with or call a true optimization specialist (you know who they are) if you want software. Don't be fooled by awards or other claims. And don't waste your time testing the limits of tools designed for very specific types of bids (e.g., transportation) in other categories with the need for more flexible constraints.

Vacation footnote: after this post was initially published, within an hour I got a handful of emails from vendors claiming that their software can account for examples like this. In January, I will revisit this post after seeing if they can prove me wrong. - Jason Busch

Share on Procurious

Discuss this:

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.