The Estimation Myth
Why we are so bad at estimating complex work, and what to do instead.
Key Takeaways
- Relative sizing is more accurate than hours.
- Uncertainty is highest at the start of work.
- Forecasting based on data beats guessing.
The Uncertainty Principle
In knowledge work, we are often asked for precise estimates for unique problems. This is a logical fallacy. Every new feature is a 'discovery' process, not a 'construction' process. We aren't building a bridge using a known blueprint; we are exploring a territory that has never been mapped before.The Cone of Uncertainty
At the beginning of a project, we know the least about it. Yet, this is when we are asked to make the most important estimates. Research shows that initial estimates can be off by a factor of 4x (400% variance). As we build and learn, the 'Cone' narrows, and our accuracy improves.📹 Video Resource: The Cone of Uncertainty Explained
Why we fail: The Planning Fallacy
Cognitive psychology (Kahneman & Tversky) proved that humans are systematically over-optimistic. We envision the 'happy path' and ignore the 100 things that could go wrong. We estimate the effort of the work, but we ignore the effort of the 'interruptions', 'handoffs', and 'unexpected bugs'.The Breakthrough: Relative Sizing
Stop guessing hours. Start comparing sizes. We don't know exactly how many pounds a dog weighs, but we can easily see if it's 'Small' (Chihuahua), 'Medium' (Beagle), or 'Extra Large' (Great Dane). This is the foundation of Story Points.Relative estimation is up to 80% faster and more accurate than trying to estimate in absolute hours because our brains are hardwired for comparison, not calculation.
— The Law of Relative Sizing
📹 Video Resource: Agile Estimation: The Dog Analogy
The Math (Law of Large Numbers)
Aggregating many small, imprecise estimates leads to much more stable results than one large, 'precise' estimate. Mathematically, the relative error decreases as you add more items to the bucket.The Law of Aggregation:
Total Variance = 1 / sqrt(N) (The more items you break your work into, the more your individual errors cancel each other out.)