Taking the Guesswork out of Performance Budgets
summary
Describes a repeatable process that uses Python/Jupyter and scikit‑learn to pick a single “hero metric” via a random forest classifier, coalesce page and business signals into visit‑level rows, then derive a performance budget from a CDF of that hero metric (example budget: 5.37 seconds to gain ~10% fewer bounces). Shows how to segment by device/country, rank problem clusters by impact and size, and use treemaps or XY rankings to prioritize fixes so teams focus on the biggest real user wins.
AI-written summary · about
chapters (24)
- 0:00Intro
- 3:20What is a closed one
- 8:20How do I get started
- 11:35Business metrics
- 12:19Visit metrics
- 13:24Refine metrics
- 15:28Data science machine learning
- 16:35Random forest classifiers
- 18:02Random forest classifier considerations
- 21:03Stop subjective bickering
- 21:39How do we get the performance budget
- 23:42Sample code
- 24:12Make it iterative
- 25:53Segment by user population
- 27:26Solving problems
- 27:57Monitoring vs optimizing
- 29:09Analyzing performance data
- 29:32Single dimension analyses
- 31:05Ranking by multiple factors
- 32:41What to fix first
- 33:28Focus on performance
- 33:44How a change in performance will change your business
- 34:29Resources
- 34:53The Big Takeaway
related talks
48mRevisiting Performance Budgets
48mPerformance Budgets: the What, Why and How
57mAn Opinionated Guide to Web Performance Budgets
06mPerformance is about more than time series charts
24mHow To Manage Performance Like a Product
36mStatistical Methods and Tools for System Performance Analysis
12mPerformance In Context
44mMaking Performance Allies