I think Taguchi methods would greatly benefit the #tech industry.

The methods describe using loss functions to measure results in $ terms. That really helps calibrate bets. https://en.m.wikipedia.org/wiki/Taguchi_methods
Genichi Taguchi showed how the cost of errors can be incorporated into measurement.

This is useful for #devops. Taguchi like Deming learned that retros don't lead to improved process. Error budgets provide a long term view that reduces stress!
They do this by creating expectations that allow management to prioritize and plan investment. No investment? No change in expectations.

Measuring in $ terms is critical because it's the only way to compare the work to revenue and hence get prioritization.
Retros don't work. I know that's super controversial but the two top reliability statisticians of all time came to that conclusion. Why?

Simple. The failure already occured, it's logically too late to understand what happened. A house of mirrors.
Oh, you found the cause? That's not the cause, so fixing the apparent root cause won't help. Instead, it takes time away from reliably and repeatedly diagnosing the cause.
By measuring errors/alerts/failures per unit of time, applying costs (yes this is a justifiable guess/estimate), setting error budgets, prioritizing fixes ahead of errors, high reliability and low stress can be achieved!
Errors and failures are best handled at the systems engineering / design level.
You can follow @statwonk.
Tip: mention @twtextapp on a Twitter thread with the keyword “unroll” to get a link to it.

Latest Threads Unrolled: