Optimization: The Only Solution to Complex Spend Management
Today's guest post is from Paul Martyn, Vice President of Marketing of Bravo Solution.
Paul can be reached at p <dot> martyn <at> bravosolution.com or 312 279 6793.
|
Most organizations have a diverse spend portfolio that includes many simple, several moderately simple, and a few complex spends. To address each spend appropriately, we need to understand the dynamics that make each event complex. For starters, let's look to another 'multi-faceted' puzzle; the Rubik's Cube. Invented in 1974 by Ernö Rubik, Rubik's Cube has puzzled generations around the world with its utter devilishness. The multi-faceted nature of the Rubik's Cube makes for a good analogy to spend management and sourcing. |
 |
Read the very interesting Wikipedia article linked above and you'll find out that a 3x3 Rubik's Cube has over 43 quintillion starting positions. But, if you know the right combinatorial magic, ANY cube can be solved in 29 or fewer moves. Like spend management, the Rubik's cube has an extraordinary number of possible starting positions but a logical process (algorithm) can elegantly solve the problem with minimal effort.
|
Complex Spend Management is another multi-faceted puzzle with even more complexity and 'faces' (internal and external to the buying organization) than Rubik's famous cube. There are many factors which influence decision-making and, like a Rubik's Cube, each factor of Complex Spend Management is related to the other factors. For example, let's look at the supplier 'facing' decision factors inherent in sourcing decisions; price, incumbency, risk and timing: |
 |
 |
If incumbency, supplier risk factors and timing are not important, the spend management puzzle is relatively straightforward to solve. We could use a reverse auction or simple RFI/RFP template and get the cheapest possible price, all other things being equal, pretty easily. This is akin to solving one face of a Rubik's cube, something most of us have the skills to do. |
But, if we are to focus on the multi-faceted nature of our negotiations and explore new, and potentially more efficient, ways of dealing with suppliers while balancing the satisfaction of our internal stakeholders (in operations, finance, marketing, etc) we must recognize how our efforts to solve one 'face' or dimension of the puzzle impact the other faces and work to find a solution that satisfies each dimension. We've all observed that the price visibility we see in a reverse auction inspires pricing creativity by suppliers. In the same way, if we can offer more visibility into other, non-price stakeholder requirements, we will stimulate suppliers to respond creatively in those areas as well.
Successful sourcing managers find creative ways to drive financial results for their company. Effectively reducing costs means challenging internal stakeholders' assumptions, preferences, and processes with scenario analysis that quantifies trade-off costs. Buyers need to expand simple price analysis to quantify the total costs (of ownership) absorbed by the operational stakeholders. Complex Spend Management requires that buyers include the inventory and logistics impact in their financial analysis. Buyers are often challenged to explore a wider variety of options to redesign their supply plan while evaluating strategic considerations like 'make versus buy'.
To address this explosion of complexity, many buying organizations have developed and maintain 'big ass spreadsheets' (BASS). BASS were often designed for a single project and then reused on subsequent complex spend events. This approach does not take into account the dynamic nature of Complex Spend Management. The BASS approach is akin to knowing the moves to solve one specific starting position of a Rubik's cube and applying it to other starting positions - it simply does not solve the 'new' puzzle. In short, buyers need a more dynamic and flexible solution.
Fortunately, today's optimization algorithms provide buyers with a technology that identifies the optimal sourcing solution for each combination of supplier pricing, buyer preferences, business rules and risk factors. This allows buyers to define a problem (starting position) and work with the suppliers in a collaborative manner to propose solutions. The buyers then use optimization to determine which combination of supplier proposals is best.
|
In essence, optimization technology provides the buyer with the ability to increase collaboration with suppliers and solve any starting position of a Rubik's Cube by simply pressing the "Solve Now" button. |
 |
Categories: Supplier Management, Decision Optimization, Cost Reduction
Tags: Strategic Sourcing Decision Optimization Spend Management
Optimization can also deliver extremely poor answers, sometimes answers that are deeply wrong. That's because the "solution" that pops out when you hit the "solve" button is almost always deeply sensitive to one or more of the assumptions ("constraints") built into your model. Rubik's Cube has exactly one correct solution, so it's really not a good analogy for understanding optimization. Your procurement problem has many, many solutions -- many shades of gray. The "best" solution isn't necessarily the "right" solution.
If you don't do literally dozens or even a hundred or more runs wherein you vary your constraint parameters by small amounts to see where the sensitivity lies (and therefore perhaps identify alternative solutions that might actually be better for your business), then the power of the optimization tool can easily walk you down entirely the wrong road.
That's why optimization is a product that, in my opinion, is best delivered with a combination of software and services. If you aren't a pro at optimization, you need someone walking alongside you and helping to explore your problem thoroughly. And in case you've been reading Spend Matters recently, the fact that an optimization vendor provides both software and services doesn't mean that the vendor is a "supply risk." It means, rather, that the vendor has approached this particular issue intelligently and rationally.
My point of view is that I'd like to see end-user-operable optimization solutions that automatically vary parameters and automatically run hundreds of scenarios, giving the user a very good idea of where he may have painted himself into a solution corner. Perhaps someone is already doing that; if so, great. Because I think that's key to making optimization practical for end users who don't necessarily understand what they're doing when they add or modify a constraint.
Eric,
Good points regarding the scenario analysis process.
However, optimization, by definition, is the one ("opti") right answer. I know I am splitting hairs but think it's an important point to raise. Optimization is a computer algorithm that does all the calculations to determine the single best answer to the question asked – hence the Rubik’s cube analogy. That is, for each scenario, optimization will calculate the minimum cost to meet your constraints – solve that problem for the ONE best answer.
Your point regarding ‘preference’ sensitivity is important as you talk about scenario analysis, constraints are hard and fast rules that are not subject to negotiation. Again, I may be splitting hairs but the misuse of technical terms (optimization/constraints) confuses business users that are not as comfortable with the technical terminology. The iterative scenario analysis you refer to is a great method for exploring the trade-off costs of preferences but not a shortcoming of optimization.
Scenario analysis helps quantify the trade off costs of preferences and helps the business users measure the difference between two scenarios. For example, I may have a corporate mandate (constraint) that at least 10% of my business is awarded to small, minority and/or women owned businesses. I can optimize my sourcing decision to ensure I comply with this constraint while minimizing my costs. Preferences, such as how many suppliers I award business to are great decisions to explore using iterative scenario analysis. Iterative scenario analysis allows me to compare optimized scenarios for many different preferences. For example, I can create individual scenarios for no more than 1 supplier (sole source), no more than 5 suppliers, 10 suppliers and an unlimited number of suppliers and compare the optimized total cost for each scenario. This comparison allows business users to measure the trade-off costs between the various number of suppliers in each award scenario while guaranteeing that the comparative basis is lowest cost for each scenario.
Again, optimization guarantees that you are comparing the lowest cost scenario for each of your preferences and makes sure your scenario comparisons are apples to apples, i.e. minimal costs given your constraints and preferences.
Paul:
Good points but:
(1) You're wrong when you say a scenario has ONE best answer. Sometimes it does. Sometimes it doesn't. It can have 10. It can have 100. It can have 10,000.
The reality is that more than one supplier can make the same bid, and from the total cost equation viewpoint, if one supplier bids 10% less but freight is 10% more, while another supplier bids 10% more but freight is 10% less, two suppliers can have the same cost. If they are both the lowest cost, it is possible that both could be part of an "optimal" solution, depending on the constraint and cost mix. While it may not happen often in certain categories, in some commodity categories, it can. And I've seen it even in complex real world scenarios.
Plus, we have to remember that computers are fixed precision machines and since optimization often involves millions of calculations, we have to account for roundoff. This means that we can never get optimal, only within some small fraction of a percent of optimal (depending on precision and scenario size). Sometimes a 1/10,000 of a percent is enough to permit a set of "optimal" solutions.
(2) I think you missed Eric's point. Sometimes you want a 3-way split, with one supplier within 10 and 20, another within 30 and 50, and another within 50 and 60 percent. Some platforms only permit rigid splits, so you have to run sensitivity analysis to truly find the best fit.
Also, and this is the point, sometimes you want a 3-way split for a demand forecast that is only 95% accurate. Depending on how you define "optimal", the best solution might be 15%, 36%, and 51% of base demand because this gives the lowest average cost per unit, which is not accounted for by optimizing the lowest cost, as that would likely be awarding only 100% of base demand, instead of 102%. Most optimizers aren't powerful enough to truly capture the desires of the modeller, which means that the models are only approximations, which means that not only can there be more than ONE lowest cost answer, there can be more than ONE right answer. In other words, when you're trying to solve an optimization problem, what you're really doing is trying to solve a whole set of hyper-dimensional Rubik's cubes simultaneously. No easy feat!
As Michael alluded to in (2) above, my main point is, when is a constraint really a constraint? Perhaps 10% MWBE is just an aspirational goal. What if I suddenly saved $400,000 by relaxing the 10% to 9%? How would I know that a huge peak of value was "hiding" just to the other side of one of my constraints?
That's one reason why I think walk-along services provided by the vendor are extremely important. Models are never perfect, so one model, and one "optimal" solution, is never enough. One cannot expect naive users to navigate complex models effectively. In general, it requires significant time and training before an average user can take the optimization helm without a consigliere.
The three of us are in violent agreement about the 'dynamic'.
My point is one of nomenclature.
Calling scenario analysis 'optimization' is like calling cooking a knife. That is, you're describing a process (cooking) by the tool (knife) that is used.
I think complex spend management is the high order process, scenario analysis is a sub process and optimization is a tool to make certain you calculate the minimum cost for each scenario.
Further, a constraint is a constraint and when it's not it's not. That is, call your goals/assumptions/preferences something other than a constraint or it's confusing to a non-technical audience that understands the word 'constraint' as a mandate and not as optimization programmers refer to the word.
I think the use of technical terminology and tools stands in the way of a broader understanding of complex spend management.
I understand and agree with Eric's assertion that you need to run many scenarios to measure and evaluate the trade-off cost of your goals/preferences.
I do not, however, agree that every event requires services. Eric, perhaps you're familiar with less intuitive tools than we have at Bravo.
As you know, Paul (you've been in this space a long time), for many years vendors have been claiming that their optimization platforms are easy to use. So Yet Another Claim along those lines doesn't quite fall on deaf ears, but almost.
I do think you'd agree that walk-along services are in many cases a useful insurance policy for good optimization results.
Eric,
I agree that walk along services are in many cases a good idea for complex spends, yes. I think there are more than just 'naive users' out there and wanted to throw the 'advanced users' a bone, as many of them are Bravo clients.
Our team's (formerly Tigris consulting) use of optimization tools to support successful spend management predates any provider in the marketplace. This experience and collaboration with category experts (clients) is the foundation for the design of our complex spend management solutions.
Paul,
Tigris Consulting was formed in 1996, before being acquired by VerticalNet in 2004, which was then acquired by BravoSolution in 2007, and the formation of Tigris Consulting does predate the formation of the oldest surviving SSDO providers in the space (CombineNet, Emptoris, and Trade Extensions in 1999/2000), but precisely when did Tigris Consulting start consulting on (strategic) sourcing optimization *with* a tool to back them up?
Most of the (public) Tigris Consulting success stories were in 2002 & 2003, and I don't personally know of any situations where they applied sourcing optimization (specific) technology before 2000/2001, which is when CombineNet and TE started offering OaaS (Optimization-as-a-Service). My understanding is that, prior to 2000, Tigris Consulting was using generic off-the-shelf optimization solutions, like CPlex, so could you fill us in as to the extent of capabilities they had prior to 2000 to support your claim that, through its acquisition, BravoSolution is the earliest provider of SSDO?
Thank you.
Tigris' first SSDO project was for a large CPG company in the fall of 1999. The project sourced a complex, blended raw material.
Later in 2001, Tigris' product strategy changed course and they established a relationship with an 'off the shelf' sourcing optimization provider.
But did they use a real home-grown SSDO tool that met the requirements for true SSDO of solid mathematical foundations (LP, MILP, etc.), true cost modelling, constraint analysis, and what-if capability ... or did they use spreadsheets which fail on two of the four requirements? If they used spreadsheets, then their use of true optimization tools does not precede the formation of CombineNet, which was the first "off-the-shelf" platform they used. And since CombineNet and Trade Extensions both formed in June of 2000, this would mean that, unfortunately, that while Tigris (then VerticalNet, then BravoSolution) was one of the first organizations to successfully apply optimization in the marketplace, they were not the first.
Also, MindFlow Technologies was incorporated in 1999 and the beta version of the product was successfully utilized at the first customer in the fall of 1999 as well. And MindFlow was acquired by Emptoris in 2006 ...
Yes, we used CPLEX as the solver, modeled true costs with constraints and had web-based RFP and Scenario Generation. It meets the minimum requirements as you've laid out.
Then it looks like you tied MindFlow for earliest application of true SSDO, and may be able to make the modified claim that your team is the oldest surviving SSDO team, assuming that the Tigris personnel who worked on that project are still with BravoSolution.
(The 1999 MindFlow beta was built by the founders and contractors. To the best of my knowledge, the founders are no longer in the space and none of the original contractors ended up at Emptoris.)
Yes, many of the original Tigris team lead our Bravo Collaborative Sourcing (BCS) team.
So, now that we've got the history sorted out, let's return to Eric's points regarding:
(1) services,
(2) constraints, and
(3) optimization goals.
While I believe that a user should be able to do it herself now that a number of solutions have turned 10, I have to agree with Eric and point out that, at least initially, a user should probably not do it herself as most people don't truly understand what optimization is, that WYAFIEWYG (What You Ask For Is Exactly What You Get), that one fat finger cost entry in a million will totally invalidate the model, or that there are more hidden gotchas when attempting to solve a complex problem than there are stars in the sky. Until you know what you're doing, it's very easy to make the wrong assumptions, build the wrong model, and make the worst possible award ... because you optimized the wrong scenario!
How do we find the right balance?
Furthermore, most platforms don't differentiate between hard and soft constraints and hard and soft goals. How do we approach this problem today, and how will we approach it in the future?
And what should the goals be. Optimization solves against an objective function which, traditionally, has been cost centric, but TCO is not always the best award - especially when social responsibility, brand power, patriotism, free market, carbon production, innovation, and other concerns are taken into account. The best award is not one that minimizes cost, but maximizes value, as companies are not measured on savings, or revenue, but profit. If paying a little more helps you charge a lot more ... then profit is maximized, and that's the ultimate goal of the company.
How do we take these concerns, and others, into account and balance cost vs. value?
And where else does optimization need to go?
Sounds like good topics for a series of 'proper' blog posts...
Should we continue our discussion in the comments or promote this to the front page?