Why teams estimate relatively — and why they hide their cards
Ask a developer how long a task takes and you get a guess dressed up as a number. Ask a team how big a task is compared to one they finished last month and you get something usable. That difference is the whole foundation of agile estimation.
Humans are famously bad at predicting absolute durations. We forget interruptions, assume the happy path, and confuse effort with elapsed time. But we are surprisingly good at comparison: shown two stories, a team will agree within seconds which one is bigger, and usually roughly by how much. Story points lean into that strength — they are a unitless, relative measure of problem size, converted into time only indirectly, through the team's actual measured velocity.
"About 8 points" is an honest statement. "6.25 hours" is a fabrication with two decimal places.
The anchoring problem
Relative estimation has one natural enemy: the first number spoken out loud. Once someone — especially someone senior — says "that's a three, right?", every following estimate gravitates toward three. The effect is well documented, robust, and completely immune to good intentions. You cannot un-hear an anchor.
The fix is procedural, not psychological: everyone commits to an estimate before anyone reveals one. That is the entire trick behind planning poker. Cards are chosen simultaneously and hidden; the reveal happens at once; the interesting conversation — why does Ana say 2 and Ben say 13? — happens after honest positions are on the table, not instead of them.
The outliers are the payoff. When the lowest and highest estimate explain themselves, one of two things surfaces: a misunderstanding about the story, or information half the team didn't have. Both are worth more than the number itself. An estimation round that ends in a short argument between the 2 and the 13 has done its job even if no estimate is recorded.
What this means for tooling
Any tool that shows votes as they arrive rebuilds the anchor it was supposed to remove. So when I built BESTIMATE, the secrecy rule went into the server, not the interface: votes are simply never transmitted to anyone before the reveal. Not hidden by CSS — absent from the wire. The estimates people give when nobody is watching are different, and they are better.