Standardization of everything, or maybe not

Β·

2 min read

In lean continuous improvement, standardization is the 4th step in the 5s methodology. This helps ensure that there are responsibilities and procedures to perform a process. The result is a consistent process outputting a consistent, and hopefully quality, product.

  • Sort

  • Set in order

  • Shine

  • Standardize

  • Sustain

The advantages of a standardized process can seem pretty clear, people know what to do to output the same product.

But taken too far, or applied in some processes or areas, can actually cause harm.

One downside I have seen is people can lose the reasons why a process is done in a certain order, or why there is a rule we follow. The rule was incorporated into the process but not the logic or reasoning behind it. Just push the button, but we don’t know why. There is definitely a place for that.

Another reason is learning. Sometimes I need to struggle through a process to learn the best method.

Maybe we don’t know the best method when we standardized the process. Standardization is asking for the end of change and innovation, we are demanding it. Does that impact the use of the tool?

πŸ’‘
Sometimes I need to be a cook and follow the recipe. πŸ‘¨πŸ»β€πŸ³
πŸ’‘
Sometimes I need to be the chemist and experiment. πŸ§‘β€πŸ”¬

I think standardization can definitely be applied to data analysis. I have started to follow a general guideline in setting up exploratory data analysis, or when building a machine learning model. It helps to follow some general steps.

I have also learned there is a fair bit of experimentation in machine learning, and over standardization of my process can be at odds with the need to experiment. Often in machine learning I need to improve a model performance. Not everything will work, and depending on the dataset or problem I’m solving for, I will turn to different tools and methods.

Make a small change

Evaluate the change

Start again

I still want the general process standard enough to follow but leave enough flexibility to experiment to find better solutions. I don't need to over apply a tool to death.

Just like training a machine learning model for too long at some point can lead to overfitting, over applying standardization by not building in steps or space for experimentation can prevent my processes from improving.

Β