The Data Heads

One of the really important aspects of the Data Cost / Value Matrix is how reducing the cost of data can help you get more value from data.

This came up again the other day as we were debriefing following a recent deployment for a new customer. This customer is about to undergo a major cultural transformation because we helped them automate real-time data collection and analysis. In the process, they will shift from a Product Control to a Process Control way of looking at the world. Ultimately this will reduce costs, increase customer satisfaction, and (we hope) drive new business.

This customer makes electronic devices. Their products are complex pieces of equipment, and at a couple places on the production line, the various component parts and the final assembly pass through an automated test stand. This test stand runs the unit though dozens - even hundreds of electrical tests. A unit has to pass all of those tests to be released to their customer.

Sounds like a winner, right? One hundred percent inspection. Nothing bad ever escapes to the end user. And we have the data to prove it. Boy, do we have data. Entire databases of rows and columns of numbers. Every one of them in spec.

Except it doesn’t work that way. Some units pass all the tests with flying colors - and then fail out of the box or early in the life of the product. Here is what the data for one test parameter looks like plotted on a histogram (this isn’t actually the customer’s data - it’s just representative):

In this example, everything is well within spec - it should all be good.

When you look at this data in time series on a control chart it looks very different:

Viewed this way, this data tells a completely different story. If you look closely you can see a couple of signals that the process is unstable, even though everything meets spec. In a complex environment with dozens of critical - and interconnected - variables, these unstable values often directly correlate to the early life and out of the box failures.

Learning to rely on the control chart is the shift from Product Control to Process Control. It is often a tough shift to make - especially when everything is in spec.

What makes it even harder, however, is the sheer volume of data they need to sift through to find those signals. Hundreds of test parameters, hundreds of units tested every day. Automation is the only way you can keep up with the data. By automating the data collection and analysis process you can empower people to make that cultural change.

What about you? How can reducing data costs make it easier to get more value from your data? Use the ShareThis button below to mark this page, or leave a comment, tweet me, schedule a conversation, or call 800-958-2709.

Dr. Neil Polhemus, CTO at StatPoint Technologies (and publisher of StatGraphics) contributed a great article in the current issue of Quality Magazine about selecting statistical software for Six Sigma. In it he lists four criteria for selecting the right statistical package:

1. How strong a background in statistics does the typical operator have?
2. What types of data are operators most likely to encounter?
3. If data are mined for information, how easily can multiple approaches with multiple options be tried?
4. How easy is it to create a report or presentation that can be shared with other colleagues?

I’ll answer each of these questions for GainSeeker Suite before the end of this post, but first I want to surface an unspoken assumption in the article and add a couple of criteria that I think Dr. Polhemus missed in his list.

The unspoken assumption is that one statistical package will serve all the needs of a Six Sigma deployment. My experience is that there are at least two broad categories of statistical software, and each has their place in Six Sigma.

Two Categories of Statistical Software

One category of statistical software is Advanced Statistical Analysis tools. Dr. Polhemus’ article outlines criteria for this group. Products in this category include StatGraphics, Minitab, JMP and others. These systems were developed (originally) for statisticians. Often Black Belts (BBs) and Master Black Belts (MBBs) depend on these packages for in-depth work in the Analysis phase. These systems are less useful for selecting projects. For the most part they operate poorly in the Control Phase. Put another way, these tools are of less use to the Champions and Business leaders who charter projects, and also of less use to Green Belts (GBs) and Process Owners who inherit and live with a project when it is completed.

The other category of statistical software is what I call Real-time Enterprise SPC Solutions. It will come as no surprise that GainSeeker Suite falls into this second group. This category comes out of the Real-Time Statistical Process Control (SPC) world. These packages are designed for ongoing data collection and analysis in a continuous improvement (kaizen) environment. These tools are, first and foremost, a tool for process owners and Green Belts. They are also tools for Champions and Sponsors (business leaders) who are chartering projects and driving business performance.

While there is some overlap between the two categories, they are more complimentary than competitive. In fact, they should readily share data. Data should be especially portable from a Real-time Enterprise SPC solution to the Advanced Statistics Solution. That way BBs and MBBs can readily tap into the enterprise data sources to support their efforts.

So here are the additional criteria that you should look for when you’re selecting a Real-time Enterprise SPC System.

Criteria 5: What does it take to get new data into the system?

Advanced Statistical Analysis Packages begin with an assumption that data are in a file, in rows and columns. In this view, data are static: generated once, analyzed in some way and then saved in a folder somewhere. Real-Time Enterprise SPC Packages assume that we are tapping into a live stream of data. Each new data point contributes to our understanding of the process. (Some of the Advanced Stats Packages are beginning to recognize this, but their core competency is in analyzing a static data set.)

The ability to readily incorporate new data is what makes Real-time Enterprise software so effective in the Control Phase. Users set up automatic data collection once and then monitor the results for exceptions.

Keep in mind too that the system should collect data at all levels of the organization. Good systems will make it easy to collect and manage data from the shop floor to the executive suite. This makes it easy to capture high levels of business metrics which can be used to help prioritize projects.

When selecting a statistical package, be sure to ask:

• Can the system tap into any data source, including front-line process owners, gages, a wide variety of text files and databases, PLCs, PDAs, cell phones, and so forth?
• Can the data entry process be controlled so that only valid data can be entered into the system, in a reliable and repeatable way?
• Can data be collected automatically and without human intervention?
• Is it easy to set up and manage these data collection processes to meet all the various needs across my business?

Criteria 6: Does the system automatically test new data for real-time process shifts?

Real-time doesn’t just refer to the process of connecting to data sources and readily incorporating new data. It also refers to statistically evaluating all new data for expected variation. This is an essential tool for understanding processes. If the system does detect a change or shift, it needs to automatically communicate that to people and systems that can do something about it.

When selecting a statistical package, be sure to ask:

• Does the system automatically detect process changes using appropriate statistical tools?
• Does the system automatically let me know there is a shift through email, pagers, on-screen displays, or other appropriate means?

Criteria 7: Are Data Stored in a Robust Relational Database?

The word “Enterprise” in our category name (Real-time Enterprise SPC Solutions) tells us that we’re not looking for a point solution. There are some fine packages out there that do SPC with Excel spreadsheets. But these programs can create a data management nightmare when you are managing all the data in your business (not to mention the risk of defects being introduced in a spreadsheet environment).

An enterprise system builds a data warehouse in a relational database. This makes it possible to tap into a rich data set for selecting and prioritizing new projects. It also makes it easier to share data (and best practices) across the organization.

When selecting a statistical package, be sure to ask:

• Are data stored in a robust relational database structure with a flexible hierarchy?
• How fast are retrievals on large data sets?
• How easy is it to group or segment data?

Criteria 8: How Easy is it to Slice and Dice the Data?

A good Real-time Enterprise SPC System will collect data at multiple levels of the organization. It shouldn’t be confined to the down and dirty shopfloor data.

By capturing this data - along with information about the data - think of it as demographic information - you can slice and dice the data to find opportunities to improve the system.

At high levels it might mean viewing Overall Equipment Effectiveness (OEE) by Line, and then drilling down into various machines or sliced across all shifts. In a transactional environment it might mean tracking cycle times across all offices, or within offices by customer service rep. Being able to easily slice and dice the data makes it easier to understand the relationship of all the parts.

When selecting a statistical package, be sure to ask:

• How easy is it to drill into various subsets of the data?
• Are automatic analysis wizards available to help prioritize and focus your attention on the critical variables?
• Can you data be rolled up into dashboards and other high level summary views for easy monitoring?
• Can data be easily tagged with demographic information?

These additional four criteria are a good starting point for rounding out your tool box of statistical software for Six Sigma.

Additional information

For more information, check out these white papers and case studies:

• Freeing Six Sigma from the ‘Data Shuffle’
• Data grouping and drill down
• Walking the talk - How Micropump built a comprehensive data system to empower their people with data
• Aberdeen Lean Six Sigma Benchmark Report
• The Role of Real-Time Data in Improving Profitability and Customer Satisfaction
• Building a Six Sigma Measurement System in Financial Services

How GainSeeker performs against Dr. Polhemus’ criteria

I promised at the start of this post that I’d address how GainSeeker Suite performs against Dr. Polhemus’ criteria.

1. How strong a background in statistics does the typical operator have?

GainSeeker Suite serves a wide population of users, and assumes that the typical user has little or no background in statistics. Furthermore the system assumes that the user has many other tasks to perform besides statistical analysis.

2. What types of data are operators most likely to encounter?

GainSeeker Suite is targeted for engineering and manufacturing applications. The product is in use in some purely transactional environments too. The system isn’t particularly robust for managing survey data, but does very well with cycle times and defect/error tracking. The system is not designed for the R&D community.

3. If data are mined for information, how easily can multiple approaches with multiple options be tried?

GainSeeker is an interactive system. Users do not need to write programs that are submitted for execution. Having said that, it is not intended to function as an advanced statistical tool. Instead it readily ports data to other software systems including advanced statistics packages. Of course Gainseeker is an excellent tool for automatically updating databases and analysis.

4. How easy is it to create a report or presentation that can be shared with other colleagues?

GainSeeker pays particular attention to sharing data, analysis and reports with various user communities. Users can access data over the web, including mobile devices, and output can readily be delivered by email. In addition, GainSeeker includes a new Enterprise Dashboard module that provides easy-to-understand role-based summary knowledge.

What do you think? Are there other Critical to Quality Characteristics that we haven’t mentioned?

A couple weeks ago I published an audio interview with Jay Bronec about his ‘Ah-Ha’ Moment when he realized that he was spending valuable company time doing non-value-added work. In that interview he described how he is automating his company’s (QualiFine) KPIs by integrating our CRM and Web data using GainSeeker.

Today I followed up with him to see how that project was coming. He took me on a webex tour of his project and I was impressed. He is using GainSeeker Suite to mine data and analyze his target market. Then he ports that over to Minitab for some advanced regression analysis that predicts class size based on how many people have registered for the class and how many days are left to sign people up. It is very cool.

But after we talked for a while, I turned on the tape recorder and asked him a question:

“Jay, whats the value you’re offering your customers? Why does it matter to people if they implement GainSeeker and get training in an economy like we have today? Why would anybody want to spend money on that today when things are crashing around us?”

For a little over seven minutes we discuss the Data Cost / Data Value Matrix and how it applies to saving money in an uncertain economy. We touch on dashboards, CMM (Coordination Measuring Machine) data collection, and how real-time data can catch problems before you waste a day’s production.

Follow this link to hear some great insights into the value of real-time automated data.

Recently Larry Coburn, the Sr. VP of Operations at Crown Audio, gave a presentation at the Aberdeen Manufacturing Excellence Summit. We recorded his talk, and you can watch it on-line.

I really enjoyed Larry’s talk because he doesn’t pull any punches. He understands how manufacturing drives the economy, and he has a powerful antidote to all the economic gloom and doom that we’re hearing:

I don’t like being a victim. When I hear manufacturing guys talk about being victims, I say, “It’s your choice. You can be a victim if you want, but I don’t recommend it.” I think you should always work to not be a victim. Do what you got to do and be a leader.

Woven in the presentation is one insight that you don’t want to miss. Larry tells the story of touring a competitor’s site (a laugh out loud experience - this guy has nerve) and learning that they’ve invested $8Million in a new factory that produced a solid product in a fraction of the time it took them to make their product. He brought that news back to his staff and challenged them to implement lean and quality improvements without an $8Million investment.

We just can’t give up. And how we gonna do it without eight million dollars? Because if we take away the demand (for product) that eight million is a boat anchor they got to pay, not us.

His team’s efforts resulted in a 90% improvement in Fabrication Lead Time, a 90% improvement in Circuit Board Assembly Lead Time, a 85% improvement in Efficiency, and a 92% reduction in absenteeism. All without an $8 Million capital expenditure.

During the presentation you’ll hear Larry talk about the importance of having the Right Data, in the Right Format, at the Right Time.

Nothing’s worse than working on two‑week-old data trying to make some good decision’s on it. It just drives you crazy. You can’t get anywhere.

In another conversation he told me:

We’ve always had tons of data, but not enough knowledge. GainSeeker gives us the knowledge to eliminate the finger pointing and focus on solving the problems. Plus it gives us the data in real-time, so now we walk into a production meeting with an up-to-the-minute yield report, with a prioritized list of what we need to work on.

If you’d like more information about Crown’s journey, you can also check out this feature story in Industry Week Magazine.

In my last post I described an interesting conversation with a customer about his company’s pilot deployment of GainSeeker Suite. You may recall that because of staff turnover, this plant was collecting data but not doing anything with it. The company was feeling pressure from a significant increase in raw material costs, and because nobody in the business knew how to use GainSeeker (because of the staff turnover) GainSeeker was not helping them reduce costs.

I had sat down with the corporate staff guy and the plant quality manager. We had started to review some of the data she had been collecting and used GainSeeker’s Analysis Wizard to drill down on the data and found that Shift A had the highest variance among three shifts, and the six or eight operators on that shift had very different results. (Click on the chart to expand to full size.)

Once we had this chart displayed on the screen, I right-clicked on the chart and then selected the ‘Control Limit Legend’ option. That displayed a list of the 8 different operators, along with the mean and range (with related control limits) of the data for each operator.

Here is how we interpreted this table, along with the chart:

It is clear that one of these operators (710) has a very different process. When you look at the control chart for this operator it is much more stable than the other operators, and when you look at the average for each of the operators, Operator 710 is running at about 173.4g compared to as high as 174.9 for some of the others. (See the yellow highlighted cells in the table). That’s a shift of about 1.5g.

Now here is something you need to know: the critical dimension is weight. Weight is critical because a minimum weight has to be met, but anything heavier than the minimum is given away - the company doesn’t get paid for it. So getting as close as possible to the minimum will reduce material costs - substantially.

How much?

We went out to the internet and found a site with typical raw material prices for this commodity. At the volume they were running, the difference between Operator 710 and Operator 214 came out to $457 per day. This is a 24/7 operation, so the annual cost savings between the two adds up to $166,861. And this only one line. This plant ran nine lines. So across the plant the potential savings of over $1.5 Million.

Who was the comedian who said “A million here. A million there. Pretty soon we’re talking real money?”

The other thing that will be obvious to you if you click on the chart is how much more stable the process is in Operator 710’s hands. Operator 214 would be foolish to try to adjust his average down because with the variation he is running, he’d be below specification too often.

Operator 710, on the other hand, could shift his process closer to the lower specification without jeopardizing quality.

So actually the impact could be even greater because the lower specification is 167.6g. If the process is tightly controlled with minimum variation, you can shift it towards the lower spec, reduce material consumption by as much as half a million a year!

Here is another way to visualize what they’re trying to do:

So is there money to be made here?

Looks like a safe bet to me.

Not long ago I had an interesting conversation with a customer, a corporate quality systems guy from a multi-plant corporation.

We had installed GainSeeker as a pilot project at one of his plants, and he and I met to plan deployment to another division. Unfortunately a couple weeks after we launched the pilot, corporate had completely upset the management apple cart at the plant and brought in (among others) a new quality manager. We met the QM and I’m sorry to say she had that deer-in-the-headlights look:bright and capable but completely overwhelmed.

In short, GainSeeker was installed but not being used except to collect a bunch of data. Not to put too fine a point on it, the QM’s training reminded me of the Dilbert cartoon where Dilbert says to the new hire: “This is your mouse. Move it around if you see anyone coming. And remember if you ask any questions you’re bothering me.”

My corporate contact bemoaned the lack of evidence to take back to his boss, which led us back into a conversation that we had touched on many times before: “Why are you deploying GainSeeker in the first place? Where is the money?”

As we talked it became clear that material costs were going through the roof, and the company needed to find ways to control and reduce those costs.

Eventually we went back to the QM’s office and her computer where we spent about 10 minutes digging into the data that they were alreading collecting.

What I found surprised all three of us. Here is a control chart of the data, our starting point. (Click on the chart to see it full-sized in a new window.)

A quick glance at this chart shows some fundamental instabilities. (And I wouldn’t recommend putting specs on the chart, but that is another topic.) But looked at this way, the chart doesn’t help us get to the underlying issue - reducing material cost.

I wanted to see if GainSeeker’s Analysis Wizard could help us see inside the data to identify any underlying causes.

In a few seconds I had learned that Shift A had the highest variance among three shifts. Not only that, but the wizard automatically drilled into that shift and pinpointed clock number (operator) as the number one driver of variation on that shift. You can see that in the following chart.

If you click on this chart it will open in a new tab or window, and you’ll see that the data for Operator 710 looks quite different from the others.

So how does this translate to the bottom line? I’m out of time for now, but stay tuned: that will be the subject of my next post.

Speaking of staying tuned…

Are you new to staying tuned to blogs? The best way to stay tuned is to set up an RSS Reader. There are lots of good sources on the web that explain how to do that. Here is one, and here are several more on video.

Andrew Downard had a great post last week on the iSixSigma blog about the maturation of Six Sigma. He argues that Six Sigma has an “Act II” problem. Act II is all about institutionalizing Six Sigma so that it actually delivers on it’s promise. Downard argues that the skill set required for Act II is very different than what is required for Act I.

My language for this is the difference between Explorers and Settlers. In US history the explorers came from outside, mapped the terrain, and then moved on to the next place. Settlers came in and put up a cabin, plowed the earth, and settled in to make a difference. The actors in Act I are Explorers. The Settlers don’t show up until Act II.

But there is another way to look at this. If you do a Google Image Search for Six Sigma DMAIC, and then tally the images by type you get a very interesting perspective on Six Sigma. Of the first 40 images, about one half of them depict DMAIC as a linear process. Here is a typical graphic:

DMAIC as a Linear Process

The other half of the images, however, are cyclical. This one is typical:

DMAIC as a Cyclical Process

I wonder if how we picture DMAIC indicates whether we’re an Explorer or a Settler, or in Downard’s terms, whether we’re in Act I or Act II.

I think it may also lead us to different conclusions about how we view data and data systems. If you’re an Act I Linear Explorer, I’m guessing you’ll see data and data systems as something that you need to conquer. Data is a wild river that holds some great fish that you can eat, but essentially it is blocking you from your next destination.

If you’re an Act II Settler, that same data river is something that can help drive subsequent cycles of the DMAIC process. In Settler terms you can use it to water your crops, drive a new sawmill, and (once you’ve invented them) power electric lights and a toaster.

If you’re an Act I Explorer, tools like GainSeeker Suite won’t make much sense to you. Who wants to build a hydro-electric dam when you’re just trying to get across the river?

But if you’re an Act II Settler, being able to connect to disparate data sources, collect and alarm based on statistical process control rules, and analyze for root cause of process variation becomes essential.

Recently a colleague forwarded this email from a friend of his and asked me for my comments:

Please let me know what lean/6sigma book you would recommend. I’ve read a couple over the years but they are a bit dull and won’t fire up our management - is there one that is simple but convincing?

You’d be surprised how hard it is to get their heads out of the sand. Part of the problem is that we are so successful and dominant in our field.

e.g. At a major meeting last week, the exec responsible for customer satisfaction presented data to show that our competitor is the leader in terms of quality and customer satisfaction. He actually said “we don’t want to copy THEM because their margin is lower than ours so their quality is costing them money”. That’s how dumb and simplistic we are.

My response:

The problem you’re facing is way too common, and I don’t think there is any pill you can give your CEO to adjust his attitude. This isn’t what you want to hear, but until your CEO’s hair is on fire, I don’t think there is any book that will make a difference.

If you can find some smoke somewhere above the hairline, and then tie that back to quality and customer satisfaction, you might get somewhere. Sadly there is usually a huge disconnect between the ceo/finance function and quality/performance. Here is an true example:

I’ve been working with a customer (an electronics firm) on a data collection/analysis project. A bright Six Sigma Black Belt proved a direct correlation between the statistically significant variation on a handful of test results and out-of-box failures at the customer’s site. These were units that met specifications on well over a thousand tests, but still failed out of the box. Those OOB failures were threatening the contract: the customer was ready to pull the business.

We put together a proposal using SPC software to capture the signals so engineers would know immediately when they had a unit that was statistically different than the others (even though it still passed all the tests). It was a beautiful solution and everyone was really excited about it.

To sell that proposal to upper management my contact bypassed the (to me) obvious argument that identifing statistically significant variation would isolate defective units, which protects the customer from receiving bad product, thereby saving the contract.

(I can hear all the MBBs in the audience saying I’m not really using statistics right, but they were doing 100% testing here so pardon me.)

Instead my contact framed the proposal around faster product release cycles and the impact that would have on inventory turns. By having real-time data as the units were produced, he argued that they would be able to release production lots from WIP to Inventory 3 or 4 hours earlier. That was an argument that got the finance guy and the CEO excited.

I tell this story because it is an example of how a creative (read politically savvy) middle manager pushed through his agenda for quality process improvement using language that any  CEO understands. No book on SS or TQM or anything else would accomplish the same thing.

So look for smoke above the hairline. Something is keeping your CEO awake at night, and unless it is his girlfriend, you can probably tie it back to quality and customer sat.

Hope that helps, and best wishes.

Regards,

Evan

Even though the CEO and CFO didn’t care, in the first 10 days of the pilot project, GainSeeker real-time SPC software trapped two defects that the test process couldn’t catch. My contact reported: “Three defects in a year is enough to knock us out of ‘Preferred Vendor’ status. Anytime I stop two defects from getting to this customer, I’m a happy man.”