Brad Doerr

Extracting Insights from Your Messy Data

Blog Post created by Brad Doerr Employee on Jan 30, 2018

If I close my eyes, lean back in my chair and take a breath, I can almost hear the echoes of kids’ voices. Tri-fold cardboard displays and colorful ribbons are splayed across rows of folding tables. I’m at my 4th grade science fair and I’ve just won first place for my solar-powered hot dog cooker.

Solar Energy School ProjectA lot of things have changed since then, but the scientific method used for my science project still stands: make an observation, form a hypothesis, conduct an experiment, capture and analyze the data and draw conclusions. Sounds simple enough, right?


In my previous blog, I talk about not being afraid to take your first step in data analytics (DA) for testing and measurement (T&M) and I outline tips on getting started. But then what?

If you want that coveted blue ribbon, read on.





Our Ambitions May Be Tripping Us Up

From where I sit in data analytics for T&M, I often see my customers – brilliant scientists, engineers, architects, project managers – get hung up towards the end of the scientific method process. After observing, hypothesizing and conducting the test, they struggle with managing and analyzing the data and drawing confident conclusions.


It’s easy to understand why: we’re all doing everything we can to get to market fast – to get to market first – that we lose the rigor we once had when we started engineering.


Some steps are taken to collect and analyze data, but they are misguided. Engineers use complex Excel spreadsheets, pivot tables, custom MS Access databases and other home-grown tools. While these tools appear good on the surface, they ultimately don’t deliver. Not only do they come up short but the tools can only be used once, and usually poorly.


Despite the best intentions, most don’t have the resources, time or expertise to deal with data or data analysis in an effective way. And it’s not just about dealing with data from one step in the product lifecycle. DA for T&M happens throughout the lifecycle:

Design -> Validation and characterization -> Release to manufacturing -> Full-scale production

So how do you find the right tools to help with your messy data?


More Data, More Confidence

It’s a basic concept: by making accurate measurements (and lots of them), the more confident you’ll be in your conclusion.


I wrote in my previous blog about how you must have your Design of Experiments (DOE) planned before you do anything. What questions are you trying to answer?


If you build a car, you’re not just going to ask, “Does it turn on?” You’ll want to know, what is its maximum speed? How will we measure speed? What kind of gas mileage does it get? How will we measure gas mileage? Will people survive in a fender bender? How many measurements do we need to ensure confidence in our eventual conclusions?


Your DOE can’t be set in stone. It needs to be fluid and you need a tool that allows you to change your DOE on the fly and not spend days – or even weeks – going back and forth with your own personal (and expensive!) IT database architecture team to get things just right. And you certainly don’t want your Team Lead spending weeks on end buried in her cubicle designing a system that only she understands or that blows up with one little change in your DOE plan.


Figure out what questions you have (and how you’ll get answers), and understand you’ll probably have more down the road.


Get Your Hands Off the Data

Imagine: you walk into a bakery and it smells of fresh cinnamon rolls. The baker rushes out to help you. Mouth watering, you point through the glass case at the squishy, doughy roll. The baker takes your $2 with his flour-dusted hands and returns a grimy quarter. He hands you your roll with his bare hands and rushes back to continue his work.

Is something wrong with this picture? Yes! The baker should hire a clerk – let’s call her a retail expert – to focus on helping customers. And the baker should focus on what he does best: bake.


It’s not much different in our world. If your designer is responsible for building cars, she shouldn’t be graphing data from her T&M. She needs to focus on what she does best and not waste valuable time formatting and processing data.


Keep your experts doing what they do expertly.


Make a Decision, but Make It with Confidence!

Analyzing big dataEventually, you need to make a decision about your product.


Is it time to build a set of prototypes? 


Can you launch to market? 


What will be your production yield? 


You can’t make a decision with confidence unless you have good data and insight from this data.


There are hundreds of information sources to help you determine how to achieve 99% confidence based on measured data. Online articles, data entry tools and tables help aggregate data and determine how much you need to get to that desired level of confidence. But first...

  1. Figure out the questions you want answered (remember your DOE!) before you begin collecting and analyzing data, and...
  2. Don’t overlook your testing environment. Would my solar-powered hot dog cooker from 4th grade perform the same on a cloudy and sunny day? Of course not. Your results won’t align if you’re not considering what’s going on in your testing environment.

Once you have your questions figured out and your testing environment set, it’s time to find your data analytics tool. 


A good tool will easily adapt with you. 

A poor tool will cause headaches, delays in your design and, even worse, cost a lot of money.


Product Lifecycles are Global and Your Data Should Be, Too

You have a globally distributed team, so your data analytics tool can’t be located on Some Guy’s laptop. And your data can’t be understood/analyzed/charted by just Some Guy who might be spending half of his time on LinkedIn looking for that next big engineering gig, when he should be designing driverless cars. It’s even worse if Some Guy is your top designer and he’s spending 50% of his time wrangling data (possibly poorly).


Data and decisions need to be communicated across your distributed organization throughout the product’s lifecycle. You need a team collaboration tool that is understood by multiple people, across your organization, and that extends throughout the entire lifecycle.


Here’s a real-life scenario that might make you sweat. Susan, a VP of validation engineering, wants to fabricate a prototype of an important new design. Tom, her top engineer, has been managing the vast amount of test data and analysis with his own system. She asks him if they’re ready to fabricate the design as he’s the only person on this highly skilled team who has the ability to analyze the data, and he says he sees no concerns. Long story short? Tom is an expert engineer, but not an expert at all facets of the design – and the others on the team were not involved in the analysis. This mistake cost the company $1 million and three months as they built something that didn’t fully work.


Imagine if all your company’s presentations had to be created using Adobe Photoshop. The hardship this would cause in terms of time, productivity, and expertise! Your data analysis tool needs to be as intuitive and easy to use as PowerPoint is when putting together a presentation.


Don’t Reinvent the Wheel

If you have butterflies in your gut when talking to your manager about a conclusion you’ve come to, it’s time to look into your data analytics tool. Explore what’s out there and test it with a small project. Report back and let me know how it goes.


In my next blog, we’ll dive a little deeper into decision-making with data analytics and figure out how to move from validation to manufacturing.