Planning poker vs T-shirt sizing

Planning poker describes how a team estimates: everyone picks a card privately, all the cards are revealed at the same moment, and the gap between the highest and lowest becomes the conversation. Nothing in that procedure specifies what is printed on the card.

T-shirt sizing describes what is printed on the card: XS through XXL instead of numbers. It says nothing about how the team arrives at a size. A team can T-shirt size by having the loudest person announce a letter, which is the failure mode planning poker exists to prevent, and it is just as available with letters as with numbers.

So “planning poker or T-shirt sizing” is a category error, and running planning poker with a T-shirt deck is not a compromise between two approaches. It is the normal way to combine them. What people actually mean when they ask the question is whether their scale should be numeric or ordinal, and that is a real decision with real trade-offs.

The main practical advantage of T-shirt sizes is one nobody puts on a slide: they are useless to a stakeholder who wants a date. A number is one arithmetic step away from a promise. If a team says 8 and a manager knows the sprint carries 40 points across two weeks, the conversion to “two days” happens silently, in someone else’s head, without any of the uncertainty the 8 was supposed to carry.

An L does not survive that trip. There is no honest way to divide a fortnight by a letter, so the question comes back to the team as “when do you think this will be done?” — which is the conversation that should have happened anyway, with the uncertainty still attached to it.

The same property is the reason sizes are good early. An estimate on unrefined work deserves to look approximate. Writing 13 on a story nobody has decomposed lends it a confidence it has not earned; writing XL says exactly as much as is actually known.

Almost every team that adopts T-shirt sizes eventually wants a velocity, and the obvious move is a mapping: XS=1, S=2, M=3, L=5, XL=8, XXL=13. It is tidy, it lines up with the Fibonacci deck, and it removes the entire benefit.

Once the mapping is published, M means 3. People stop asking whether a story is bigger than the reference and start asking which number they are willing to defend, and the letter becomes decoration over a numeric estimate. You now have Fibonacci with a translation layer, plus a new problem: the letters imply ranges, the numbers imply points, and the two get summed as though they were the same claim. Half a sprint of Ms shows up in a report as an exact figure that no one ever intended.

SprintsPoker does hold a numeric weight behind each size, because otherwise it cannot show an average at all. Treat that average as a rough centre of gravity for the round, not as a score to plan a sprint against. If you want numbers you can commit to, re-estimate the split stories with a numeric deck rather than converting letters you assigned when you knew less.

Backlog roughing or sprint commitment? Sorting a roadmap, comparing epics, deciding what is worth refining at all — letters. Work going into a sprint that someone will be held to — numbers.

New team or calibrated team? A team with no shared history has nothing to anchor a 5 to, and its early numbers will be noise dressed as data. Start on sizes, run a few sprints, then pick a finished story as a reference and switch. A team that already has a reference story and a stable velocity gains very little from moving to letters.

Who reads the output? If the estimates leave the team and land in a portfolio review, letters protect you from being held to arithmetic you never performed. If the estimates stay inside the team and feed a forecast, numbers are simply more useful.

How much time do you have? Forty backlog items in an hour is a T-shirt exercise. Eight refined stories in an hour is a Fibonacci one.

The arrangement that works for most teams is sequential rather than either/or. A new epic gets a T-shirt size in a short session with the product owner present, purely to answer whether it is worth doing and roughly where it sits against everything else on the roadmap. Nothing is committed and nothing is summed.

Anything that survives that pass gets decomposed. The resulting stories are estimated with a Fibonacci deck in the normal refinement session, and those numbers are the ones that enter the sprint and the velocity history. Crucially, the earlier letter is not an input to the later number. If an epic sized L decomposes into 34 points, the epic was wrong, and nobody should be arguing the stories down to match it.

Both passes are planning poker. Only the deck changes. What the numbers should mean once you get there is covered in the story point estimation guide.