The Role of Optimization in Strategic Sourcing – Optimization in the Strategic Sourcing Process

This series discusses the recent report from CAPS Research on “the role of optimization in strategic sourcing”. The primary goal is to highlight, clarify, and, in some cases, correct parts of the report that are important, confusing, or incorrect to insure that you have the best introduction to strategic sourcing decision optimization that one can have.

The first chapter in the report provides a brief history of optimization, which dates back to 1947 when G. Dantzig developed the simplex algorithm, defines sourcing optimization, describes some inherent complexity of sourcing models that makes them well suited for decision optimization, and places optimization in the sourcing process. It also defines “expressiveness“, which is just a fancy term used by a couple of providers to say that the models can handle sophisticated bidding with tiered and volume-based discounts and bundles.

Since I have already provided you with a good introduction to decision optimization in the wiki-paper, which includes the basic requirements for a true strategic sourcing decision optimization solution, I am instead going to focus on those aspects of the chapter that I found to be misleading or, in a couple of cases, incorrect.

First up, the statement that optimization applications for strategic sourcing are rather new is just plain wrong. These solutions, which turn 10 next year, have been around since 2000! To put things in perspective, the first generally available e-auction platforms did not start appearing until 1996, the year after FreeMarkets was formed. I’ll admit that there were only a few solutions at first, and that the first instantiations were primitive by today’s standards, but by 2003, a few of these solutions could handle models that were very sophisticated. It’s true that solution times were in the hours, and sometimes days, for some of these models, but that was still better than sending a scenario over to an operations research group and waiting 2-3 weeks for them to do a custom analysis. And today, with increased computing power and better solvers, those models generally solve in a matter of minutes, and sometimes seconds.

Next up, Figure 1 on the Strategic Sourcing Process. While this figure does capture all of the options and steps, it’s a little misleading because optimization can precede, follow, or be used simultaneously with sealed bids, negotiations, and reverse auctions, and the sub-cycle can be repeated as many times as you like. You can bid to get potential suppliers, optimize to find those suppliers who qualify and fall close to the required bid range, do an e-auction to get initial bids, optimize, negotiate with the top 5 suppliers, optimize again, and then make awards to the top 3 in a 50/30/20 split, for example. (Provided that you explain to the suppliers up front it will be a multi-round sourcing event so that you can’t be accused of unethical conduct.)

Continuing on, while the statement that any category with a medium to high level of spend and complexity is a candidate for optimization is true, it is a bit misleading because it doesn’t define “medium to high level” and conjures up images of global sourcing models with dozens of suppliers supplying hundreds of items across thousands of lanes in the minds of some potential users. The reality is that if you have a model where only three suppliers are bidding on only three items for only three locations and where shipping costs are dependent on the total volume on a lane, you already have a complex model that most professionals will solve sub-optimally even though it can be solved, with effort, in a spreadsheet. Consider the simple example discussed in the NLP podcast (transcript). Three bidders, one item, three locations, volume discounts on the bids, and a supply constraint. An “obvious” solution could easily cost you 2.5% more than you need to pay. An even worse solution could cost you 4.5% more than you need to pay … on a model that you would probably label “child’s play”. So just imagine how much you could be overspending on even your average sourcing event!

This says that any company who followed the lead of the company that put a lower limit of 5M on a sourcing event before optimization could be used would probably be foolish. Considering that repeated studies have found that strategic sourcing decision optimization returns an average of 12% beyond what reverse auctions and other standard negotiation methodologies can deliver, this company is likely leaving hundreds of thousands on the table in an average sourcing event over 1M, if not more! A number of sourcing providers have delivered returns of 20%, 30%, and 40% on categories under 5M using strategic sourcing decision optimization. If you acquire a blanket license, and appropriately train your team, the cost per event becomes ludicrously cheap in comparison to the potential savings and it becomes stupid not to at least run an unconstrained scenario to understand your base cost.

Next we have the statement that if it is expected that a supplier’s pricing will depend on the total amount of business they are awarded, then they must be asked to bid on larger bundles as well as on discrete parts, which is just plain wrong. Maybe a few of the older products have this limitation, but the new products don’t. This is fairly easy to encode on a true strategic sourcing decision optimization platform that uses a powerful Mixed-Integer Linear Programming (MILP) solver at its core. Good products support tiered and volume-based quotes on individual products and arbitrary product groups, negating the need for a supplier to provide multiple quotes at pre-defined price points and for pre-defined lots. They only have to define the discounts they offer. No more. No less.

Finally, while the statement that there are limits to the ‘expressiveness’ that optimization can accommodate is true, the example provided is not. The report states that “one bidder submitted a bid that had the potential to save the buying company 10M, but would require the buying company to hire 18 additional people and place them on site at the supplier’s location … because it was different, this bid could not be included in the optimization analysis. Again, while this may have been a limitation of the particular tool being used, it is not a limitation of strategic sourcing decision optimization models in general. There are a number of ways this could have been handled. The easiest way is to treat it as a fixed cost. Since the optimization model is for an expected demand over a fixed time frame, we have 18 bodies at an average overhead (salaries and benefits) of $Y/month for X months. This says that utilizing this supplier’s bid comes with an overhead of 18 * Y * X. Fixed costs are very simple binary constraints in MILP models and a number of optimization products on the market today can handle fixed costs.

Other than that, this chapter served as a good introduction to the report and strategic sourcing decision optimization in general.

Next Part II: The Benefits of Optimization

Share This on Linked In