Can R&D Lab Chemists DOUBLE the Speed of New Product Development?

New Product Development Strategist

Mukul Mehta

Yes, Hope is not a Strategy.

In late Nineties, using my 20 years US industrial experience, I started tackling this problem.

How do I develop a simple predictive tool? An effective strategy?

If you are a professional mathematician, statistician or deep data analytics nerd, you will not like a simplistic tool stripped of all the razzle-dazzle and flexibility you are so accustomed to. Don't read further. You will be disappointed.

But, it may be just the right tool for the lab R&D Chemists/Scientists who hate math. Easy to learn, easy to use and super fast almost like Waymo.

As my friend, Jim Lucas of DuPont said if you are a very busy consultant, trying to get the job done and make a first production run, what would you do? Here is my answer.

Let us start with a simple flow chart of a typical Chemical R&D project.  

Simple Flow Chart of a Typical R&D Project

Flow Chart of a R&D project

First few steps may not take much time and effort. Steps 5 through 8, Recipe Development, Process Improvement, Application Development, and Scale-up take a long time and are very expensive.

While we try to simplify these steps, we need to address needs of four diverse stakeholders: the Business, the Scientist, Manufacturing, and the Customer. Here is a short list. 

Meet Four Diverse Needs Simultaneously. Simply.

  • Business Needs
  • Scientist Needs
  • Manuf. Needs
  • Customer Needs
  • Reduce time to market by 20 to 50%?
  • Develop custom products quickly?
  • Develop better products faster?
  • Improve R&D productivity?
  • Retain and rapidly reuse knowledge?
  • Ease technology transfer


On any industrial research project, you need to answer all these questions.

 Being a hero in R&D may not be enough. You also need to and want to develop high credibility with your peers in Manufacturing. For that you need to package your know-how  and know-why  and turn it over to manufacturing in production mode; where people with less expertise, can produce the same high quality product again and again, easily.

If you can do that, you are liked by R&D, loved by Manufacturing. And, you will get lots of accolades from Management.

What strategy should we use? How should we begin and address the needs of the four diverse stakeholders? How do we develop a simple Waymo like Predictive tool that can?

Hope as a strategy won't work. We need a new strategy. No a strategy won't cut it. It will be too vague, too difficult to implement. We need to simplify, simplify to bare essentials. Convert it to a new tool, that is:

  • Easy to use,
  • Minimizes new jargon
  • It is mistake-proof, uses poke-yoke
  • Can be learned in a few hours.
  • Encourages rapid iterative learning.

Before I share the tool, let me talk about the underlying principle. It is a MANTRA for real success in new product development:

Fail fast, fail frequently, fail frugally, but fruitfully.

Type the mantra in big bold letters, like a banner. Hang it on your wall. This is contrary to what you may have heard, but it is the real truth. Edison experimented with the light bulb filament materials over 400 times and failed. His answer, "I know what does not work."

Mantra of R&D

Mantra of R&D

Why do we want to “Fail fast, fail frequently, fail frugally, but fruitfully?”

The answer is simple. Technological problem solving is very difficult; it takes a lot of time and effort, far more than one realizes. When we estimate completion times, we tend to be optimists!

The brute force Edisonian approach won’t work.

You have to be smart. Very smart.

Occasionally you may get lucky. Your intuition is just right and you run the right experiment in the R&D Lab. But usually your intuition becomes right after many many failures, hard searches, and a lot of questioning and thinking. As Steve Job's would say "... over time the dots get connected." A picture emerges.

So if intuition or guessing won't help, is there a better way?

Yes there is. 

We need to build a Predictive Tool like Waymo founded on solid industry proven science with a high degree of predictability.  

Let us study with an example.

Example: Develop a Great Cheesecake

Imagine we would like to develop an optimum recipe for a cheesecake to replace Sara Lee's cheesecake available in the frozen food section of the American grocery store.

Let's brainstorm and make a list of the recipe and process variables that may be important for baking such a cheesecake. Here is a good list to start.

Developing a New Cheesecake Product

        INPUTs

  1. Eggs to Cream-cheese ratio
  2. Sugar to Cream-cheese ratio
  3. Flavoring to Cream-cheese ratio
  4. Mixing Time
  5. Mixing Speed
  6. Ambient Temperature
  7. Baking Temperature
  8. Baking Time

       OUTPUTs

  1. Percent Moisture
  2. Particle Size (texture)
  3. Color
  4. Bulk Density (height of the cake)
  5. Weight
  6. Cost


I listed eight recipe and process variables. A pastry chef with considerable expertise in baking cakes will easily add a few variables to our list. But, let us assume for now that this list is adequate.

Let us assume you, the scientist want to consider each input variable at two settings, e. g., Mixing Time of one or two minutes, use  two or three Eggs. 

Here are a few quiz questions:

Question 1: How many experiments can you run, if you only use two settings for each variable?

Your Answer 1: ________________ 


Question 2: When was the last time you counted the total number of possible experiments, for your project?

Your Answer 2: ________________


Question 3: What is the chance that you will find the optimum if you run 20-40 experiments, selected arbitrarily? 

Your Answer 3? _______________


How about applying the newly learned mantra, "fail fast, fail frequently, fail frugally but fruitfully", repeatedly?

Title deleted?

How do we do that?

How do we fail quickly and learn FAST?

As Einstein would say, make it as simple as you can but no simler.

We need a really simple formula that can tell you the Minimum Number of RIGHT experiments needed to declare failure. ( Learn why you are failing , revise your thinking and move to greener pastures? )

What do you think, if you had this real simple formula would it help you in your R&D planning?

Notice I said RIGHT Experiments. If you run the RIGHT experiments, you learn FAST.

RIGHT experiments help you learn FAST. To identify the RIGHT experiments, you need to ask the RIGHT questions. In the table above for Four Diverse Needs, I listed twenty right questions. That is just the start. We can dig deeper and list a dozen more.

What is the simple formula? I will tell you in a while. (Taylor Series)

Title deleted ?

Did you know that there is a proven scientific formula for finding the set of the RIGHT experiments for Lab R&D?  ( Want proof: Start with a multivariate Taylor Series, Use Newton-Raphson approximation, and apply the Kiefer-Wolfowitz theorem.)  In most cases the answer will be for the first iteration run 16 or fewer experiments. The KW will also tell you which ones!

Here is an interesting thought, a concept to mull over. "Just because you do NOT know the laws of science, you are not immune to these laws." Laws of science are merciless. Violate them at your own peril.

 If we leverage the RIGHT questions, use the TWO Scientific FORMULAS we can quite often find the optimum recipe/process  in 10 to 30 experiments, two to four weeks work, and answer all the questions that Manufacturing may ask!

Remember I told you I was lucky. 

In 1969, while at Graduate School in USA studying Chemical Engineering, I read about an interesting seminar announcement: “New Techniques for Industrial Product Development.” A visiting faculty member was going to talk about a newer approach to product development and optimization. I was intrigued. I attended the seminar and it changed my focus, my career and professional life. The visiting faculty member introduced us to the new predictive science, and how to use this approach to reduce laboratory experimental effort by 50% and more. 

My first reaction was, “Wow, I can apply these methods to most of my chemical technology related problems.” Well, I enrolled and picked up another graduate degree, in applied statistics with a thesis in the D-Optimal Designs, the scientific formula for finding the RIGHT experiments that will help you find the maximum information at minimum cost for your R&D project. As a part of my graduate thesis work, I studied from cover to cover Fedorov's new book,  Theory of Optimal Experiments! (lucky break, hun)

I got lucky again. I joined BF Goodrich Chemical Technical Center  in Cleveland, OH.  They had just started to promote COED - Computer Optimized Experimental Designs on CompuServe timeshare computing platform. Pretty soon DuPont, Dow and others started using the same technology.

We tried to teach this new science to R&D scientists. To our surprise we failed. we heard that story repeated again and again in various large R&D organizations.

We wondered, what we were doing wrong. Maybe the R&D scientists were just not interested.

We struggled and struggled.

As Pogo said: "We had found our enemy. It was us," the teachers. We were too close to the subject to look at the problem objectively enough.

As Daniel Kahneman, the Nobel Prize winner, would say much later: "It is easy to fool anyone. The easiest person to fool is yourself." Maybe we were fooling ourselves. 

Why? Why? Why? What were we missing?

It took me a few years to recognize the many issues:

  • the formidable mathematical/statistical jargon.
  • The ugly math.
  • The needs of busy lab chemists/scientists who are math averse. Who were never exposed to this science/methods/tools during their schooling years. I had a Ph. D. chemist from Harvard lament they were never exposed to this thinking. He wished they did.
  • The complexity of simplifying a big catalog of methods developed from 1930s to today.  
  • Windows software menu driven user interface that hid everything from a novice frustrated user.

I had an idea.

A typical R&D scientist works on three or four projects per year. If I were an R&D Chemist, I would need to use this new technology 3-4 times a year. This frequency of use is too low, too low to learn any subject and develop even modicum proficiency.

Windows menu driven user interface that hides the methods in sub menus does not help either.

Around that time I was using a Roadmap analogy for project management.

  1. Where am I going?
  2. Where am I?
  3. How do I get there?

INSERT READ MORE ...??

Using this analogy, I solved the problem of Windows hidden menu commands. My team and I developed a new version of the experimental design software specifically for Lab R&D scientists. 

New User Interface, Replacing Windows Menus

Replacing Menu Driven User Interface


Will the "Roadmap Analogy" work? Will this new user interface work where each command is clearly visible and commands are deliberately arranged to read from top to bottom in each column? 

At that time competitive Experimental Design software was being sold for $1000 - $1200 per user license. I started selling our software at a very high price point deliberately,  I wanted to fail fast.   A single user license priced at $7500 and a corporate license for $100,000. Two day training included.

To my pleasant surprise I signed up seven large, Fortune 1000 companies within a year. Their scientists started using it and said "... this is easy enough, very useful and will help us do our job better."

It was a good start, but, underneath each button there was still some jargon and some mathematical decision making. 

How do I change it and use the language  of Lab R&D? Lab R&D scientist who does not know statistics or mathematics? How do I overcome the "jargon", the "not invented here syndrome?" Five years later we found the answer. 

Six Step Process

Define project
Easily speed up new product development

++++++++++

It is a simple step-wise system.

Step 1: Define the Project/Problem.

Provide a brief description in 5 - 10 lines of simple text. This description brings everyone a basic understanding of the project, its goals and objectives. 

"Develop a better cheesecake than Sara Lee's frozen cheesecake" is a good start. though the description does not explain what it means to be better? Higher Moisture content? Better texture(smaller particle size?), Cake height? Weight? Cost? color? and so on? Sooner or later we need to provide an operational definition that key stakeholders can agree on.

Step 2: Establish Reproducibility.

This is easy. Just rerun an experiment several times, and compute the average and standard Deviation for each product attribute. Note we need to rerun the experiment several times, testing a sample several times will not do.

Step 3: Plan Experiments

This is a key step. We need to identify and select a small set of experiments that will help us get the maximum information at minimum cost. Since during early stages of a project failure is very likely, we need to keep the effort to a minimum, conserve the resources for a 2nd trial and a 3rd trial. How many experiments do we need for a meaningful failure, which results in good learning, good insights.

Step 4: Build Models, Analyze data. Predict Performance.

With a minimal data set how do we build good models? Make meaningful predictions that might give us a clue of the next steps. If you fail, it should tell you why you are failing. It might even point you into a better direction, greener pastures. 

Step 5: Optimize.

If it is a chemical reaction, optimum is usually reasonably well defined. Higher conversion, higher purity. Keep impurities to a really low level. If it is a polymer, plastic or food product more thoughts need to go into what to optimize. What is an acceptable product?

Step 6: Play What if...

In early stages failure is more likely and so we need to ask, if we did not get the optimum, how do we tweak the product performance? Does that give us a reasonable solution? If not what next?

While the steps are simple enough, the thought process is not, it is highly structured, ensuring high meaningful learning at each step. Almost guaranteeing fastest overall learning with minimal effort. 

I could write extensively about each step, tell you stories of abject failure because one or more steps were skipped. Eventually the "simple omissions" were identified with lots of unhappy discussions and people. 

While the steps when described appear to be simple enough, they still involve 18 mini-steps(buttons) and perhaps as many more are hidden.

And while an experienced consultant can identify and list all possible failure points, the solution is still a bit fragile in inexperienced scientist's hands. 

It needs poka-yoke, mistake-proofing!

Mistake proofing, or its Japanese equivalent poka-yoke (pronounced PO-ka yo-KAY), is the use of any automatic device or method that either makes it impossible for an error to occur or makes the error immediately obvious once it has occurred. (https://asq.org/quality-resources/mistake-proofing)

Now I am retired and pursuing this as my hobby and passion. I just signed up as a volunteer Adjunct Professor at my alma mater, Institute of Chemical Technology, Mumbai. I hope to share my knowledge and expertise, teach 10,000 R&D Lab Chemists and/or students over next three years and give away my know-how and know-why.

Now, I am away from corporate R&D,  away from influence of my techie peers. It gave me a chance to rethink from the ground up and simplify.   How do I develop This NEW Predictive Tool? A real low cost Waymo for R&D Scientists? With newer technologies? Better? Faster? Cheaper?

As Einstein would stay, "Make it as simple as you can, but no simpler."

I have.

I have simplified it as much as I can to make it easier for new Lab R&D users to learn. 

Here is a quick 60 second silent demo.


Silent Video Demo

Maybe you can help me make it even simpler. 

I need early adopters. Are you willing to participate as an early adopter? 

The tool is you guessed it, FastR&D version 3.0.

Why use FastR&D v 3.0 in Lab R&D?

Two DOE Examples from Industry:

Let me illustrate the power of DOE with two examples I worked on.

  • A Fridel-Craft alkylation project involving mono, di and tri alkylated amines with several byproducts. With 9 experiments (yes, 9!)in a three liter laboratory reactor involving Reaction Ratio, Temperature and Catalyst Concentration, we helped Manufacturing improve productivity from 550 lbs/hour to 770 lbs/hour.
  • Scaled up a Recycle Process from an 1-inch laboratory reactor to 18-foot diameter, 70+ feet tall, fluid-bed reactor producing 750,000 pounds per day, increased raw material efficiency from 96% to 98% with 20 well designed DOE and confirmatory experiments at PLANT SCALE!

If you use these techniques on five problems within one year, you will easily double your success rate!

The Waymo like tool for new product development?

You guessed it, FastR&D version 3.0.

Maybe you can help me make it even simpler. 

I need early adopters.

Are you willing to participate as an early adopter?  join me before June 15 and take advantage of this early adopter special.

>