The Fibonacci scale in estimation

What is wrong with an even scale

Give a team a scale from one to ten and it will start arguing whether an item is a seven or an eight. The argument feels substantial, but it has no answer: the difference between seven and eight is smaller than the precision with which a large item can be sized at all.

An even scale promises accuracy that does not exist. The bigger the work, the more unknown it contains — and so the wider the step between neighbouring values should be.

What the gaps buy you

Fibonacci grows faster and faster: 1, 2, 3, 5, 8, 13, 21. Between two and three the difference is one; between eight and thirteen it is already five. Choosing between neighbouring cards always stays meaningful, because they differ enough for the difference to matter.

The scale also tells you what the conversation is really about. “Is this a five or an eight” asks whether the item contains a second chunk of work. “Is this a seven or an eight” asks nothing.

The third effect is speed. There are few options, so choosing takes seconds instead of minutes. That matters more than it sounds: a round where everyone deliberates eats the whole meeting.

What the deck looks like

The classic Mike Cohn deck is 0, 1, 2, 3, 5, 8, 13, 20, 40, 100, plus a “?” and a coffee cup. The numbers past thirteen are no longer strictly Fibonacci — they were rounded, because at those sizes precision is gone anyway.

Many teams cut the deck off at eight or thirteen, and they are right to: a forty is not an estimate but a statement that the item needs splitting. Zero is kept for work already done in passing or finished in a couple of minutes.

• Zero — there is nearly nothing to do.
• 1, 2, 3 — small and medium items whose size is obvious right away.
• 5, 8 — large ones: usually two or three chunks of work inside.
• 13 and above — a signal to split, not an estimate.
• A special card for “I do not know” or “sizing this is pointless” — more honest than a random number.

What to do with an item bigger than eight

Split it. Not by layer — “backend” and “frontend” are not two items but one in two halves that cannot ship separately — but by outcome: which chunk can be taken to done and shown.

If an item refuses to split at all, it usually means it is not understood yet. Then its place is in refinement rather than on the table: half an hour of discussion saves a sprint.

Other scales

Fibonacci is not the only option, and every working scale is built the same way: the step grows with the size.

• T-shirt sizes (XS, S, M, L, XL) — for teams that want no numbers at all and do not track velocity.
• Powers of two (1, 2, 4, 8, 16) — the step grows even faster than in Fibonacci.
• Only 1, 2, 3 — for teams that split everything large in advance.
• Counting items instead of points: all cards are roughly the same size and you count how many shipped.

The “I don’t know” card

The classic deck has “?” for this — “I do not have enough information”. That card on the table means the round did not happen: questions first, numbers after.

A separate card is worth having for the other case, when sizing is pointless not because of ignorance but because the size changes nothing. The item will be done now anyway, and arguing whether it is a one or a three costs more than the item itself. In our deck that card is called “poher” — it says out loud what teams already say to each other.