Best Analytics Approach: Holmes vs. Columbus


Imagine you are a quality manager at a solar cell manufacturing line, and you have just received a note from your testing group that the last batch produced had an alarming rate of low-efficiency product. Given the implications and impact of such an event, you need to get into full action to understand the reasons for the higher failure rate and how to fix it for the current and future batches. Meanwhile, the clock is running, because each incremental failure is adding to the millions of lost revenue already incurred. What do you do?

Now imagine you are a treasure hunter who has just received a grant from a private investor to find lost treasures in the Pacific Ocean. You'll get a special bonus for finding the Japanese general Tomoyuki Yamashita's ship, which had precious treasures looted from Burma. The grant is enough to help sustain you and your team in the sea for 30 days. What do you do?

Although these are two very different situations, the paths to the solution -- or, should we say, the paths to the efficient solution -- would be similar.

As a treasure hunter, you could take a “Christopher Columbus, the explorer” approach: setting sail with a crew and a submarine in tow and starting to look for the gold. If you go the explorer route, you are guaranteed the views (nice corals, beautiful wildlife, emerald-green waters), but the chances of finding gold in 30 days are slim. Essentially, you are not directing your effort toward finding gold. Your actions of exploration are independent of what you are looking for. You'd take the same action when tasked with looking for killer whales.

Or you could take a “Sherlock Holmes, the detective” approach and identify potential areas for shipwrecks in general and where Yamashita’s ship might have sunk in particular. How do you identify top candidates for the shipwreck's location? Can you look at historical trade routes and records of wrecks and then eliminate some locations using depth information and possibly records of Yamashita’s post-World War II retreat? By using clues and facts, you can start identifying potential areas to explore.

Once you have identified a dozen-plus potential locations, you can prioritize the top three areas and then use your submarine or deep-sea divers to go explore. You will likely find your gold in a much shorter time. You will either succeed fast or fail faster and then restrategize to attack the problem again.

Going back to our quality manager in the solar cell line with a major malfunction, the approach to identifying and fixing the problem is no different than the treasure hunt. You could take the Columbus approach and start collecting data points, hoping to find the causes of the failure. But, as you can imagine, with multiple assembly lines, each having multitudes of processes and equipment, your chances of finding the problem area quickly are going to be slim. It could be because of a single valve malfunction, but imagine the probability of finding it among the millions of things you could inspect!

That changes if you take the Holmes approach and identify clues guided by where the failures happened. Are all the lines producing unusually faulty products? When did the problem start? Where exactly are the faults in the product, and to what processes/equipment do they correspond? You get the gist!

The important thing to note is you don’t need to know all the answers to the guided questions you are asking. You can construct a solid hypothesis based on what you know and then use the hypothesis to unravel the potential problem candidates. With this approach, you can find quickly that high temperature in line 10 caused the faulty construction. Process recipe changes or malfunctioning hardware, such as heat exchangers, often cause temperature issues. By identifying the top things to consider, you can easily narrow down the candidates and find the cause of the problem -- in this case, a heat exchanger's faulty valve.

I can’t imagine a treasure hunter worth any salt going the Columbus route, but I have seen enough explorers of data in the business world! Efficient managers and analysts use the guided Holmes approach to direct their efforts and look for answers that are relevant to the problems at hand, thereby finding gold nuggets and delivering a financial impact to the organization. In this case, you, the smart manager, can follow the clever Holmes approach and save your company millions by identifying and replacing the faulty valve in a few days!

To learn more about the detective approach to analytics, download this whitepaper on Aryng's five-step analytics framework for moving from data to decision. We elaborate on how hypothesis-driven analysis helped us identify $120,000 for a $1 million winery in just two hours!

Point / Counterpoint, President & CEO, Aryng.com

Counterpoint: Train for Data Visualization Skills

Chances are you've already got good data visualization experts on staff, even if you don't know it yet.

Customer-Centric Banking Analytics Scares Me

Banks have tons of customer data at their disposal; unfortunately not all will use it scrupulously.


Re: A matter of degree
  • 4/22/2012 12:19:38 PM
NO RATINGS

That sounds like a successful implementation. Have they implemented identical analytics to other business units since then?

Re: A matter of degree
  • 4/20/2012 10:17:11 PM
NO RATINGS

Dear all,

I agree with everyone that less is more. From my exposure to organizational goals and objectives, strategic plans provide direction to medium term and tactical initiatives which in turn provide direction to organizational goals and metrics.

Critical few goals (e.g., Profitability, Cost, Quality & Safety, and Growth) provide laser-like focus to the organization. Within the organization every employee has the same goals but the matter and degree (% of contribution) would differ.

This in my past experience has helped the organization to keep a standard, simple, and focused pursuit towards organizational alignment.

Murali Sastry

Re: Best Analytics Approach: Holmes vs. Columbus
  • 4/20/2012 9:53:07 PM
NO RATINGS

Piyanka,

I appreciate your complements.

I forgot 1 last step in the problem solving process.

After corrective action is verified, typically there is an activity widely known as "Lessons Learned" where the process owner shares lessons learned within the organization or if it is a global organization with other facilities to prevent others from creating the same defect.

This step would be very beneficial in any endeavor of the organization.

Thanks again for the opportunity.

Murali Sastry

Re: Best Analytics Approach: Holmes vs. Columbus
  • 4/19/2012 11:15:45 PM
NO RATINGS

Murali,

 

Very well laid out "structured" problem-solving to manufacturung defects.. Thanks for sharing.

 

Piyanka

Best Analytics Approach: Holmes vs. Columbus
  • 4/19/2012 10:48:55 PM
NO RATINGS

The defective product in a manufacturing process typically has tell-tale signs in addition to product identification. Using scientific approach to problem solving, the following could be collected in a matter of minutes to an hour:

1. Product produced on an assembly line (date code stamp) on product if available. 2. Review existing Failure Mode and Effects Analysis (FMEA) 3. Control plan for the line and product (shows inspection, and test protocol and criteria etc.) 4. Process flow diagram (process sequence with responsibilities, operation etc).

Based on the above, following actions are typically taken:

1. Containment Action (in suspect product lot) so suspect product is quarantined from rest of "good" production parts.

2. Root Cause Analysis-combination of brainstorming (with process experts), review of tell tale signs of product failure, talking to "parts" (identifying differences and causes of Best of the Best (BOB) and Worst of the Worst (WOW) products, Fish Bone Diagram, etc. could be drawn to move towards most likely cause and testing the cause by turning the problem "on" and "off" to finalize the root cause.

3. The above step is critical in leading towards problem resolution (permanent corrective action) that directly addresses the root cause identified above.

Typically the problem resolution happens in days to weeks depending on complexity of the problem, identification of causes, organization culture, infrastructure conducive to problem solving etc.

I like the article on Holmes Vs. Columbus. It is thought provoking and thanks to Piyanka.

Murali Sastry

Re: A matter of degree
  • 4/18/2012 11:12:36 PM
NO RATINGS

Piyanka,

A clear example of how sometimes less is more. It also illustrates your point perfectly. In this case, looking for the right solutuion for an unweildy problem was clearly the right approach.

Re: A matter of degree
  • 4/18/2012 7:44:02 AM
NO RATINGS
1 saves

"However, practically, I have seen, the better the tool, the higher the expectation/dependence on the "tool" showing the right answer"

Priynka, you are right. There are many tools for the same reason and the output only depends up on the input datas. Tools cannot create any outputs; it can only segregate the results based on inbuilt equations and decision blocks. So the data flow and decision approaches are more important.

Re: A matter of degree
  • 4/17/2012 6:49:32 PM
NO RATINGS

Piyanka's view is what I strongly believe in as well. If the tools were too perfect, nobody would hire an analyst. Most of the tools though advanced and user friendly, unfortunately lack "scope for customization". A perfect GUI can just get you what you need but not interpret the results for you. How would a marketing manager understand  'Rsquare' , 'chi-square' and 'validation misclassification rate'.

It is the analysts who need take the " Statistics to Business reommendations" path rather than relying on a tool with the threat for "Garbage in Garbage out" 

Re: A matter of degree
  • 4/17/2012 4:29:38 PM
NO RATINGS

Ah ha! Good point Piyanka. I do think some cuation is needed when introducing data visualization, for the very reasons you spell out. Great for those who "get it," but too much of a detraction for those who only think they get it. Thanks!

 

Re: A matter of degree
  • 4/17/2012 4:20:37 PM
NO RATINGS

Theoretically, you are right!

However, practically, I have seen, the better the tool, the higher the expectation/dependence on the "tool" showing the right answer.  But unfortunately, tool doesn't have the answer, the analyst with a proper method can find the answer. So my experience with great visual tools is it often distracts because it can visualize anything and everything and can thus allow a user to be lazy and dump everything into the tool and see what comes up.. a common mistake I see is folks would get super excited by a deviation from trend and would spend a lot of time on it before realizing that the trend affects very very tiny percentage of the population (like a very small country in emerging markets) and thus has literally no impact on the business (for example: on the overall global business) even though visually, one sees a clear deviation from the mean.. So in my experience

1. good tool + good skills  = great business results

2. good tool <> great business results

 

Page 1 / 2   >   >>
INFORMATION RESOURCES
ANALYTICS IN ACTION
CARTERTOONS
VIEW ALL +
QUICK POLL
VIEW ALL +