What Do Rubik's Cube and Expressive Bidding Have in Common?
From time-to-time, I enjoy checking-in over at CombineNotes, a CombineNet sponsored blog. CombineNotes digs into everything from advanced negotiation and combinatorial -- they now call it "expressive" – bidding approaches to back yard BBQ recipes. It's a fun mix of sourcing and culture. Back in September, Paul Martyn, CombineNet's CMO, came up with a sourcing analogy that I thought would be worth sharing with Spend Matters readers. In his post, Paul compares solving a Rubick's cube to making the right sourcing decision. He writes that a "3x3 Rubik's Cube has over 43 quintillion positions, a vast unfathomable amount also expressed as: ~4.3 × 10^19. But, if you know the right combinatorial magic, ANY cube can be solved in 29 or fewer moves. A 12-year-old boy in 1981, according to the article, wrote a book on some techniques to solve a Rubik's Cube, selling 1.5 million copies ... Sourcing is another puzzle with seemingly endless possibilities, and we say Expressive Commerce is the same kind of elegant solution." Check out Paul's post and decide for yourself whether you think he's onto something or if he's simply drinking too much Pittsburgh tap water.
- Jason Busch
Tweet
TweetBacks
Paul is onto something AND he's drinking too much Pittsburgh water. And this is not a contradiction.
Paul has made some good posts as of late (this post in particular is hopefully the first in a series that will close in on CombineNet's capabilities: http://www.combinenotes.com/PermaLink,guid,26b3261...), and I was waiting for the penultimate post I'm sure Paul is building up to before commenting, but now that it's spreading, I need to point out that the Rubick's Cube is simultaneously the best AND worst example you can use to describe optimization.
In a Rubick's Cube, as you pointed out, there are over 43 quintillion combinations but ANY cube can be solved in 29 or fewer moves. In other words, there appears to be infinite complexity but the actual complexity is infinitesimal.
In optimization, there are often an infinite number of possible solutions, but a very small finite number of optimal ones. IF you can express the model as a LP, then there are a exponentially large, but finite, number of possible solutions that may be optimal. And IF you can express the problem properly, add some good data structures, and apply some good algorithms, then, chances are you can get away with only checking a small polynomial number of them. This is still a lot, and more than you can ever solve by hand for any reasonably-sized problem, but from a mathematical perspective, on today's high end computing platforms, you've reduced an unsolvable impossible complexity to a finite one that can often be solved in a few minutes, or hours at most. In other words, the apparent complexity is that of a Rubick's cube (the good part of the analogy) but the actual complexity is much more than a Rubick's cube ... much, much, much more ... but not unsolvable with the right platform supporting the right model. In other words, optimization is not necessarily as hard as you think, but a lot harder than the example that is used!
And as for Expressive Commerce, it's not an elegant solution - it's an elegant model. The difference is subtle, but important. The solution algorithm is not pinned down, could be different for every problem, and the actual implementation may or not be elegant, but the model, which allows you to identify potential solution algorithms, and which is the key to solving an optimization problem, is elegant. (As for uniqueness, that's a different debate that I'll save for a later time.)
It's a good model, and their platform has some good solution algorithms, but what's important is not the solution capability, since there are lots of optimization engines on the market that can solve appropriately represented optimization problems, but it's ability to represent a problem, as there are fewer significant products that can accurately represent the model you need to generate a reasonable solution to a real-world problem.
I know it sounds like I'm splitting hairs, but there is a sharp distinction between problem, model, and solution algorithm in optimization and confusing them can be dangerous. It's hard to explain as there are not many real world processes that exemplify the distinctions. The best I can think of off the top of my head is computer networking. Problem: I need a network that spans my offices. Model: I need a network that spans these floors of these buildings. Solution 1: Wireless networking base stations at points A, B, C, D, etc. Solution 2: hard-wire each office, install routers at W, X, Y, etc. and map each connection point to the nearest router. In other words, it's the model that's key - not the solution algorithm.
Check out CombineNet's blog, and Paul's recent posts in particular, and keep watching. I'm sure you'll hear more from me as time goes on. There are some advantages of CombineNet's solution that I believe are relatively distinct (if not entirely unique) in the marketplace, but I do not think Paul has really dived into them yet ... so you can probably expect some more goading on my part in the future!
Revolution's calling!
Michael
2 design engineers at competing aircraft firms are arguing who's got the best real-time guidance systems (poke at CombineNet here - like the SAP SEM cockpit) for their new jets. Meanwhile, a passenger wants to get from point A to point B. Now, from the rumsfield school of Q&A:
Does the passenger care about their argument? no
Do she care about getting to B safely and inexpensively? yes
Should she fly the jets herself? no
Should she fly it alongside the expert? maybe
Should she pick the best airplane to take her there? yes
Do the airplanes have different capabilities? yes
Is it a function of any one sub-system? no *everyone agrees it's as much the model and expertise of the modeler than the power of the solver
Should she get help to pick the right airplane? yes
Is a blog the best place to do this? no. she should talk to an unbiased expert.
Does CombineNet's capabilities blow away Iasta's for complex logistics environments, such as solving a complete FTL network with thousands of OD pairs and constraints? you betcha. sorry, the fool digresses here.
But, most importantly, does she really need a high-powered jet? occasionally
Does she need to own it? absolutely not
Might she just want to lease a honda (btw, who are getting into the jet business, but I again digress) for the other 99% of her travel? you betcha (advantage Iasta/Procuri)
Don't laugh at the Honda/Toyota. We did that in 1973. oops.
Technology trickles down and complexity inevitably gets solved, automated, and deployed to the masses. Until then, take a cab.
Power to the people. Stay foolish all.
"Make everything as simple as possible, but not simpler". Albert Einstein
I have, since joining CombineNet, enjoyed the banter around 'Good Enough'. That is, why buy a Lear Jet when all you need is a bicycle to get to work. Fortunately for the Jet manufacturers, we live in a Global economy and the bike only gets you to the end of your block.
Where in the 'Good Enough' argument is there room for continuous improvement? Today's leading companies are not looking to 'copy' the success of their competitors (trickle down theory), they are looking to drive innovative improvements - continuously.
More specifically to the sourcing arena. The argument for 'Good Enough' is expensive, costing organizations at least (according to several analysts) 5-10% savings that's left on the table. Compound this with the failed implementations from 'Good Enough' events that didn't capture the execution stakeholders constraints/cost preferences or all of the elements of Total Landed Cost and the 'Good Enough' camp replies - 'at least we have contract management'.
At the end of the day, if the Lear Jet is as inexpensive and as easy to drive as the bicycle....
Gotta run to the toilet, too much of a good thing (Pittsburgh water)...
You heard it here folks, CombineNet software delivered as a service for $5K an event or $50K a year unlimited!
but seriously, talk to the users, not just the analysts.
the spend warriors are out there with friggin pitchforks and since they don't have a lot of money to spend on this muffler, it's either the BASS system (Big Ass SpreadSheet - no offense to people with big butts intended) or nothing. this is why the simpler on-demand guys are kicken BASS right now - the "Bondo" vendors that fill a hole in this market until the Borg's (SAP, Oracle) assimilate us all (sigh). Meanwhile, MindFlow? XPORTA? eventually, Emptoris? Bueller? Bueller?
From BASS to SaaS: yeah, ok, the combinatorial optimization squeezes out the extra value (and recovers potentially destroyed value from reverse auction - even when it's a 'price discovery' step in a texas 2-step process a la Jason), but, I'll take 'simple' tools deployed to the masses (a la Toyota) to touch way more spend, even if the savings are slightly less. Simplicity is key - especially for compliance. How well are companies doing with basic compliance? not so good. Now what happens with superfantasticbubbleplastic optimization events? Take CombineNet's forte - transportation. MIT estimates truckload contract compliance at less than 50%. Cool tools are slick, but the fool's tools are quick. remember, the smallest spiders can be the most deadly.
Fool out.
Reverse auctions can include optimization. Feedback (in various forms from lowest price to winning based on side constraints, such as, no more than 10 winning suppliers) can coexist with optimization.
We offer our Advanced Sourcing Application Platform (ASAP)with the option to run sourcing events with many different flavors of feedback and constraint options. When you can solve the optimization problem in less than a second, you can provide a richer marketplace for auctions - including feedback.
The tool is simple and includes the following:
Self service interfaces
Best Practice - category based - templates
We also offer Program Management to help our new clients get over their fears and process hurdles that prevent them from realizing 'easy savings' and move away from the less strategic time spent on BASS work. Which Buyer do you know wouldn't want to spend more time with suppliers and less time in front of a spreadsheet?
The solutions we offer address spends from $1million to $1billion and cover the full range of categories (not just transportation).
The solutions integrate with ERP, e-Sourcing, Supply Chain Management and Transportation Management Systems helping our clients bridge the gap between finding savings and realizing savings.
Additionally, considering organizational side constraints and preferences as PART OF YOUR SOURCING PROCESS is a surefire way to improve compliance - that is, finding the optimal IMPLEMENTABLE allocation of business to suppliers significantly increases stakeholders compliance.
Let's face it, it's not about how much better/faster the optimization engine is (nobody doubts we have the best/fastest), at the end of the day, it's what you do with it.
Simply put, we make flying a jet as easy as driving a car.
Thanks for jumping in! These comment threads are much more fun when I'm not the only one elbowing the vendor! As to your points:
Yes, CombineNet's capabilities blow away Iasta's, Procuri's, Emptoris', etc. capabilities for complex logistics environments - I've never denied that - but, as you point out, the important point is when you need the lear jet and when a honda (or used cesna) will do nicely.
Paul:
I love your last point - at the end of the day it is what you do with the optimization engine - and this means that the key point that you need to make is what the optimization engine can do, versus how it does it. After all, the user needs to understand why - and when - she should spend as much on an optimization tool as on an entire on-demand sourcing suite from Iasta or Procuri. As I've said before, and will say again, POE (Platform Optimization Engine) solutions based on COTS (Commercial Off-The-Shelf) optimizers can solve many problems well, but not only is their problem set limited when compared to BoB (Best of Breed), there are problems where the quality of the solution will not be as good as what a finer grained BoB model can produce.
However, often the better model only saves 2-5%, not 5-10% - which means the money left on the table is very significant on a spend > 100 M, and BoB pays for itself many times over, but the money left on the table on a spend < 10 M wouldn't even cover the cost of the more expensive tool and the extra effort because - and let's face it - the quality of solution is limited to the quality of the data! Better models, require better data - and collecting, cleansing, and aggregating that data - which always requires more manual intervention than you would like - always costs (significantly) the first time.
So, in summary: the SpendFool is right. Paul is right. And I'm right. It all comes down to perspective - and the problem you're facing. I wonder what our moderator has to say after having a few hours on the course to dwell on the issue?
Not sure on your 'mythbuster' comment though. If you have a massive market basket, you can do winner-take-all (e.g., as an auction) or bust it up into smaller lots and do the same. Both are sub-optimal though for many reasons as you know. So, open it up and let suppliers bid 'expressively' and solve and THEN you can possibly do some lotting based on data you gather. So, they can work together, but they are not synonomous nor is optimization a part of a reverse auction - it's a complement to it. but we digress.
Enough folly for today. Hope you hit 'em straight jason!
CombineNet provides an advanced sourcing platform (Advanced Sourcing Applicaton Platform - ASAP) powered by optimization. ASAP allows CombineNet clients to create an event, invite suppliers, communicate requirements, collect expressive bids, model constraints and preferences, look at the event results in the context of a broader supply plan and automatically port the resulting allocation to the transactions systems - how hard is this?
I think folks get intimidated by what they don't understand - the underlying math that powers our engine.
How many of us understand the inner workings of our Honda engine?
We trust that we simply hit the right foot pedal and the engine roars - our optimization is no different - hit the Optimize Now button and see your savings accelerate.
And please don't tell me you don't wear the silicon valley software salesman uniform, you know: Tweed tan jacket. black turtleneck. slightly too chunky gold jewelry. little overweight. Loafers. occasional hair plugs. bluetooth headset. weathered face.
god bless them though. With the trends towards SaaS, let's pray for their children.
enough tomfoolery. fool out.
If so, let's get together to discuss this topic over a meal or drink.
Of course, you could attend our Optimization round table discussion, as well. I will refrain from wearing the Silicon salesman uniform (not sure i even own a turtle neck).
As for 'intellectual' debate - I'm still waiting.... ;)
Jokes aside, I think we've circled the topic well, with many excellent points - would be nice to synthesize the points (minus the Pompeil) into a comprehensive article.