Building analyses in layers

There's a saying that to an engineer, π = 3.

Is this true?

According to this Quora answer, yes:

The interesting thing is that for mathematics, even "pi = 3.141593" isn't precise enough. There's always a way to make pi more precise.

Yet, for most practical applications, just a few digits are enough.

I say this because most of us work with numbers.

We're building models and doing analyses to decide what to recommend to our clients, or what our team should work on, or what investments to make and initiatives to pursue.

The problem that I see is that we should be building models where pi = 3 way more often.

Let me explain.

The "sophistication trade-off" of models and analyses

We're trained to think that the more precise and sophisticated the analysis, the better.

In fact, in the "precision-oriented" mindset, a back-of-the-napkin calculation is wrong. It doesn't consider all of these 30 other variables that would change the math.

But we're no mathematicians here. We're more like engineers. Not as concerned with the precision of our deductions, as we are with the practical effects of our conclusions.

What the "precise" people leave aside are the HUGE costs of precision: time and clarity.

Time because analyses and quantitative models scale exponentially with precision.

A back-of-the-napkin calculation may take you 5-10 minutes to make. A slightly more sophisticated model, a few hours. A highly sophisticated model, days. And the most precise one, several weeks.

Clarity because, let's face it, no one else is going to understand the model you took several weeks to build unless you show them the simpler versions first.

It's no wonder that most senior executives, even in fields like finance, use back-of-the-napkin analyses all the time: they're quick to build and easy for others to understand.

But of course, sometimes we need the complexity… So what to do?

How to thrive in the "sophistication trade-off" using LAYERS

Here's my tip that can help you save hours and hours of work and gain respect from senior people at the same time:

Build your analyses in layers.

Before you jump in and create a 2-week "perfect" model that's incredibly precise and has no imperfections (except for the fact that you're late and that information isn't that useful anymore)...

… Build a "back-of-the-napkin" version first.

Then, share with others.

If it's useful, build a slightly more precise version of that analysis.

Did the answer change by much? Is there any variable you left out that you think could change the direction of the outcome?

If so, build a slightly more complex version. And on, and on, and on.

This way, you get two benefits:

1. You can walk other people through your rationale in layers, which makes it easier for them to understand it.

2. You won't spend weeks on something that would never work (and could've taken you 5 minutes)

Just like an engineer has to choose how many digits of pi are enough for their calculation, you too need to figure out how complex (and precise) your analysis has to be for you to achieve your outcome.

And ideally, you won't spend one more minute making it more precise.

Sure, there are situations where having higher precision will give you an advantage. But most times, it'll just give you more work. And that work uses time that you could be using for something more useful than making your model 1% more precise.

You have to use your feeling (and your analytical skills) to know when to stop. When to say it's good enough.

What I don't advise you to do is to be so fear-driven that you always go for maximum complexity/precision from the get-go. This is working dumb.

And here, here we work smarter than that.