T-shirt Sizing

A quarterly planning session arrives with something like forty candidates on the table. Assigning each one a number takes three hours even at five minutes apiece, and after two hours the participants' attention goes before the list does. The numbers on the later items aren't produced by the same standard as the earlier ones.

What the time bought was small, too. Items at this stage are chunks nobody has broken down yet, so unknowns are still mixed in. Arguing for twenty minutes over whether one line reading "improve payments" is a 5 or an 8 usually ends with neither once the work starts. That the margin of error on the same task depends on when you estimate it is the cone of uncertainty, and before anything begins the actual effort can run to four times the estimate.

Meanwhile the answer that session actually needed was singular: can we fit all of this into the quarter? Just seeing that eight large things are on the list answers it, and no amount of extra precision changes that answer.

So a method with a deliberately coarse scale took hold. Accuracy is given up, and in return forty items get surveyed in an hour — an hour that is enough to make the next decision. Precision that changes no decision is precision nobody needed.

The sequence starts with picking a reference item for each grade. Six steps in all.

1. Choose the references from work already finished. Last quarter's invite-link improvement is M, replacing the payment provider is L. Without references, everyone picks grades from a different private yardstick.
2. Keep four grades. S, M, L, XL is common; go past five and the arguments about boundaries come back. Coarseness is the point.
3. Write a rough duration band next to each grade: S a few days, M one to two weeks, L three to six weeks, XL beyond that. The table is for sharing a sense of scale among participants, not for conversion — try to distinguish six weeks from eight and there's no reason to be using this method.
4. Participants reveal grades simultaneously. That prevents everyone drifting toward whoever spoke first, and when someone says S against someone else's L, both explain why. Missing information usually surfaces in that exchange: one person was picturing only the screen while the other was counting a data migration. The grade was never the point of the exercise; the surfaced information was.
5. Treat XL as a signal rather than a grade. Don't work to pin its size; split it on the spot. An item that large means something inside it is still undecided.
6. Don't use the results as a sum. Instead of converting grades to numbers and adding, read how many items are L or above. Ten of them and the fact that the quarter can't hold all of it is visible in the room. Counting is also something everyone present can verify.

Right before work starts, you measure again. These grades are for quarterly planning; moving down to cycle planning means splitting the items and re-sizing them on a finer scale.

It fits where the items are numerous, large, and still carrying unknowns: quarterly or half-year planning, filtering roadmap candidates, an area nobody has touched before. None of those needs a precise number. What they need is whether it fits and what to look at first.

Five limits.

1. It can't be summed. Grades carry order but no defined interval, so there's no way to know whether two Ms equal one L. Producing a total means converting to numbers, and that conversion removes the method's advantage.
2. The grades bunch. Left alone, more than half of everything lands on M. Reference items per grade are what reduce that, and even then M stays the largest column.
3. It doesn't travel outside the team. Our M and the next team's M aren't the same quantity, so combining two teams' grades into a company plan produces something misaligned.
4. It can't carry detailed planning. What goes into this cycle has to be known in days, and that requires smaller items and a finer scale, where planning poker — revealing a card per item — is the technique that fits.
5. Attach dates to grades and they become commitments. Once a table reading "L is six weeks" hardens, an estimate has turned into a deadline. An estimate can be wrong; a deadline can only be missed. So the duration bands stay inside the team for calibration and never become the schedule shown outside.

"Forty items — t-shirts, not poker. We have to finish inside an hour."

Choosing the technique from the item count.

"This one's XL. Don't grade it, split it."

Treating XL as a signal.

"Nine Ls. You're going to pick five of them for the quarter."

Reading item counts instead of a total.

Pagedeck ran a quarterly plan this way. Mia brought 36 candidates and Theo and Noa graded them with her in an hour and a half. The result was fourteen S, twelve M, seven L, three XL.

Splitting the three XLs on the spot produced six more items across L and M. Grace then looked at thirteen items at L or above and decided the quarter would take four. Nothing was ever summed; the count of large items made the decision.

The most common reading is that this is sloppy estimating. The scale is coarse and it takes little time, so it looks that way. But coarse and careless are different properties, and the reduced accuracy here is a choice. On items nobody has broken down, a precise number is unlikely to match reality, and the time spent chasing the match exceeds the accuracy gained. Treat it as sloppiness, use a fine scale in this seat, and you spend three hours to be wrong by the same margin anyway.

The other frequent move is converting grades to numbers and adding them. Set S to 1, M to 3, and L to 8 and the arithmetic works, so it's a natural attempt. But grades are a scale with no interval defined in the first place. The numbers substituted in are values the team invented, and summing them produces a total with nothing behind it. If a total is what the occasion needs, the right move is to use story points from the start rather than this method.