All too often we don’t put processes in place which couple the human ability to evaluate nuance “intelligently” with the machine ability to evaluate vast amounts of data “dumbly”.  All too often we confuse efficiency with effectiveness, and pursue efficiency over effectiveness, exemplified by the use of the term “machine intelligence”.

Nothing brings this out more clearly than the recent stock market behaviour.  All the “quants” were quick to identify “human error” initially.  Not only did they say it was human error, but it was female error.  I’m surprised they didn’t suggest she was a blond too.  After all, we know how they confuse their B’s with their M’s.  How ridiculous!  Now that calmer analysis has taken place, it would seem that nothing of the sort happened, and not by a female either.  There is a very interesting article – I am sure there must be many more out there – in the Wall Street Journal (WSJ) by Aaron Lucchetti titled “Exchanges Point Fingers Over Human Hands” that analyzes what really went on last week Thursday. Lucchetti makes no bones about the fact that this is a man vs machine tussle:

“In the man-vs.-machine argument for financial markets, proponents of technology say machines do it faster and cheaper. Those in support of human involvement say people can use their experience and pull the emergency brake when the computers, or their programmers, make mistakes.

But when that happened Thursday, it appeared that some humans couldn’t react quickly enough, while those using computers just kept pushing the market lower.”

I would argue that human involvement should have been used to prevent the situation from occurring, not just as an “emergency brake”.

Let’s start by understanding the role of the “quants” in financial organizations.  A “quant” is short for a quantitative analyst.  These are math and physics whizzes that have been brought into financial institutions to create mathematical models to evaluate market behaviour, particularly algorithmic trading.  Algorithmic trading is a trading system that utilizes very advanced mathematical models for making transaction decisions in the financial markets. The strict rules built into the model attempt to determine the optimal time for an order to be placed that will cause the least amount of impact on a stock’s price. Large blocks of shares are usually purchased by dividing the large share block into smaller lots and allowing the complex algorithms to decide when the smaller blocks are to be purchased.  The use of algorithmic trading is most commonly used by large institutional investors due to the large amount of shares they purchase every day. Complex algorithms allow these investors to obtain the best possible price without significantly affecting the stock’s price and increasing purchasing costs.

Let me come clean;  I am an engineer, so I am a “quant” by nature and by training.  But I had the good fortune to study “decision under uncertainty” at the PhD level.  During this time I also came across “fuzzy logic”.  Forget the math and theory.  Fundamentally what it comes down to is that some people (quants) believe that any and all systems can be modelled exactly – given enough time and insight – and that the models can then be used to predict behaviour under any other circumstances.  I think this is a load of hogwash.  No mathematical model is ever complete and data is never 100% accurate.  However, when computers are used by humans to understand “directionally correct” decisions, they are of huge benefit.  In other words the model of the supply chain may indicate a 5.21% improvement in gross margin from 23.42% to 28.63% if supplier A is used rather than supplier B.  I would interpret the result to mean that it is highly likely that we could increase gross margin by more than 2.5% by using supplier A.  It would have probably taken a human months to gather, collate, and analyse the data by hand, and probably with a great deal of “human error”.  The same analysis could be achieved in a few hours using a computer, provided some of the primary data was already available.

There is an interesting little snippet in the Wikipedia description of quants which I think is of particular relevance.

“Because of their backgrounds, quants draw from three forms of mathematics: statistics and probability, calculus centered around partial differential equations (PDE’s), and econometrics. The majority of quants have received little formal education in mainstream economics, and often apply a mindset drawn from the physical sciences. Physicists tend to have significantly less experience of statistical techniques, and thus lean to approaches based upon PDEs, and solutions to these based upon numerical analysis.”

Statistical techniques are based upon uncertainty, or randomness.  Physicists, mathematicians, and engineers, on the other hand, hate uncertainty, and spend enormous amounts of time looking deeper and deeper into atoms trying to prove that everything is predictable, if only we had the knowledge and wisdom to understand the observations.  And they bring this perspective to the analysis of financial market behaviour, as pointed out in the Wikipedia quote.  Einstein once made the statement that “God doesn’t play dice with the universe”, which he came to regret, incidentally.  He was questioning the notion of randomness as opposed to determinism. Determinism is defined as understanding every event in nature as having a particular cause. Randomness defines an aspect in nature that has only a probability such as in quantum uncertainty.  My engineering training was replete with this deterministic attitude which informed Einstein statement, as was the training of my fellow engineers and scientists.  So the quants are in constant pursuit of the ultimate model to describe all situations so that they can predict the movement of the market under any and all conditions.  This is an attitude that is very common in supply chain management too.  I think it is flawed from the start.

In a separate article in the WSJ titled “Did a Big Bet Help Trigger ‘Black Swan’ Stock Swoon?”, it is clear that what happened last week Thursday was not “human error”, but rather “model error” in the sense that there was an over reliance on computer models, which in turn drove market behaviour.

The non-quants have been fighting back for some time since the market crash in 2008 and the whole CDO mess.  A good example of this is Scott Patterson’s book “The Quants: How a New Breed of Math Whizzes Conquered Wall Street and Nearly Destroyed It”.  It is a fascinating read, and very instructive.  But also fairly predictable in the blame game.  What I found most interesting was a comment by a reader of a book review of “The Quants” in the Globe and Mail.  Interestingly the title of the book review is “Quants accept no blame for financial crisis”.  Can’t be more explicit than that.  The reader wrote that

“In finance, you have a lot of people in high positions who are surprisingly innumerate (MBAs and the like) – they didn’t really understand what the quants were doing but didn’t mind as long as they were making money. Let’s not forget who hired the quants in the first place! When you combine this lack of technical oversight with poor regulation, you have a toxic mix.”

I believe we have a very similar situation in manufacturing operations, particularly supply chain management.  Senior management doesn’t really understand the complexities of operations and rely too heavily on the quants.  As long as they see inventories go down and stock prices go up, all is well.

To go back to Lucchetti’s article in the WSJ, the first act in the blame game for the market behaviour last week Thursday was to focus on “human error”.  Clearly a first salvo from the quants.  Later in the article, Lucchetti quotes Jamie of White Cap Trading as stating that

“Markets are a mix of technology and human judgment. Thursday, we saw far too much technology and not enough (human) judgment.”

I could not agree more.  I think I am going to print out that statement in 94 pt font and put it in a frame on my wall.  I would like to see everyone in supply chain management follow my example.

All too often I see this same behaviour in supply chain management where optimization engines are thrown at a problem.  I do not have too much of an issue with the use of optimization engines.  What I struggle with is that there is a slavish belief that the results are accurate to the nth decimal.  There is no understanding of the likelihood of achieving this optimum nor the degree to which the model is inaccurate nor the degree by which the result is affected by inaccurate data.  What happened in the stock markets is a classic example of relying too much on machines in the pursuit of efficiency.  The parallel’s in the supply chain space where we rely too much on optimization, be that Lean or mathematical optimization.  Do you know the first sign of when the quants have taken over your supply chain?  It’s when you hear that your data isn’t clean enough after you have already spent millions implementing an ERP system and countless hours “cleaning” data.

I am not suggesting that we unplug the ERP and APS systems we have deployed over the past 20 years.  I think there is a huge amount of value that has been received from the use of these tools.  But they are tools.  Let us treat them in that manner.

As always, I look forward to a robust debate, perhaps including some of my erstwhile colleagues.

As we proposed then, our suggestion to i2 customers is to turn this time of uncertainty into a time of opportunity… to do something different and get more. 

i2 customers really need to evaluate their “pent up” and unsatisfied needs and look for a solution that can best address them, and in a most cost-effective, low-risk,  time-efficient way.  We try to avoid too much self-promotion on this blog, but in this case I want to ensure that everyone understands the distinction.

Casio is a perfect example of a customer of ours that installed i2 (and took 3 years to do it), only to find that the performance results achieved were far below targets, and that it failed to address changes in demand and supply effectively so users still had to deal with these manually. 

Several other customers, Toshiba as one example, evaluated i2 and decided to go with RapidResponse.  There is a good article here on Toshiba – one quote says it all – “The company looked at several other solutions (including i2), but found that nothing provided the capabilities that RapidResponse did – it was unique in the value proposition it could offer.”

We offer a one-to-many value proposition, providing a single integrated on-demand service that solves a variety of supply chain problems (which can replace upwards of seven i2 modules, incidentally).   There is a real difference between RapidResponse and traditional planning systems, namely:

RapidResponse

• Capitalized on human judgment
• 1000’s of daily users/experts
• Plan-Monitor-Respond
• Typical deployment – 14 weeks
• On-demand
• One solution – many applications

Legacy Planning Systems

• Black-box optimization
• 1-5 expert users
• Plan only (assumes static world)
• Lengthy – often promised  benefits never entirely realized
• On-premise (significant IT resources)
• Multiple planning tools + Excel hell + integration nightmares

Food for thought anyway….

Here is the abstract:

What is needed in today’s dispersed and loosely coupled supply chains is more collaboration, and less control; more coordination, and less optimization. Companies must have the ability to enable their front-line people to use their judgment to make fact based decisions which address the surprise and compromise inherent in today’s global and multi-tier supply chains. At the heart of delivering these capabilities is the technical architecture of the supply chain solutions.

This paper describes the integrated set of capabilities that is required to satisfy the business needs of the 21st century supply chain.

Let us know if you agree…

The part of the article which prompted me to respond was the section on the correct level of modeling. Having studied optimization techniques in an Operations Research graduate program and worked for i2 for many years, I was steeped in the belief that optimization was the answer. Over the years I have come to the conclusion that optimization can provide direction but not the ultimate answer. The reason is captured in the article under the headlines of:

• Complexity of the Underlying Model
• Data Quality
• “The Business has Changed”

Without accurate representation of the business (Complexity of the Underlying Model), it is not possible to generate an optimum. One may get close, but a small change in the assumptions or model can have a big effect on the results, especially at the detail level. Yet the level of information required to model the business accurately is massive (Data Quality). More importantly, it also changes constantly (The Business has Changed). Not only does the physical structure of the supply chain — suppliers, items, locations, customers – but so do the business rules and objectives.

The solution is to provide “enough” detail in the model that an optimization/search engine can get one into the region of the optimum and then have people evaluate the results in a collaborative manner in order to provide the human judgment and compromise necessary to adapt to the latest business conditions. This is a lot more efficient than trying to capture and modify constantly the business rules used to guide the optimization.

To take the paradigm of human judgment further, the speed of business has increased dramatically over the past two decades, when many of the APS solutions where designed and constructed. At that time it was sufficient to run and APS engine once a month or once a week. Coupled with the knowledge that the optimzation does not provide an exact answer (nor is the forecast 100% accurate), this leads to the requirement for a system in which people can create scenarios in a rapid and effective manner in order to evaluate the organizations response to reality. Human judgment and compromise is used to select the “best” option amongst the scenarios.

It would be remiss of me not to point out that I work for a vendor that provides an on-demand service with integrated demand-supply planning, monitoring and collaborative response capabilities.