#23 Quantifying Quality - (Pebblebed) Keith Adams | Slush 2022

Quantifying quality is a psychological concept, challenging to measure directly. Insights on managing quality, effective metrics, and the limitations of feedback in achieving "good" experiences.

Key takeaways
  • Quantifying quality is challenging because it’s a psychological concept that can’t be directly measured.
  • The only kind of good experience is when nothing bad happens, and all bad things fail to materialize.
  • Management discipline is necessary to ensure quality.
  • Using a bug database as an approach to assessing health is not effective.
  • It’s more useful to focus on the global health of a system, rather than specific metrics.
  • The lower three charts (performance, quality, and security) don’t align, and need to be treated separately.
  • The lower one (performance) is most sensitive to changes.
  • Zeroing in on the sweet spot of “fast enough” is subjective and depends on the system.
  • A slightly better question to ask is “is this event of sufficient magnitude to tell me something meaningful?”
  • The masses of customers, who are never going to bother to tell you why they churned, are lost to feedback.
  • Timely feedback is crucial for learning, and users are unlikely to report problems unless they’re significantly impacted.
  • Quality is a narrow definition: software behaves as intended.
  • The numbers game: multiplying slow and 0.99 gets you 0.99, and 0.98 to 0.99 is more significant.
  • The nines system is a simple way to express quality as a percentage of uptime or responsiveness.
  • Goodhart’s law: aiming for a target can lead to perverting its meaning.
  • The alpha and one minus alpha part is an editorial decision about how to weight the two components.
  • Post-product-market fit, quality gets decoupled from the user’s experience.
  • Fudging numbers can lead to a pessimistic impression of service quality.
  • Feedback is often obtained from a biased sample, like customer support tickets, which underreport actual problems.
  • Directly measuring successes and failures is difficult, but we can do a better-than-chance job of estimating problems.
  • Measuring quality is a psychological concept, so Goodhart’s law applies. *apter