Affinity estimation and the bucket system: size a big backlog fast

Affinity estimation and the bucket system are two fast ways to put relative sizes on a large number of backlog items. In affinity estimation, the team silently arranges items from smallest to largest, then groups them into sizes. In the bucket system, the team calibrates a scale on a few items together, then each person places their share of the rest into labelled buckets such as 1, 2, 3, 5, 8 and 13. Both trade a little precision for a lot of speed.

Use them when 40, 100 or 300 items need a size for a roadmap or a release plan, and save planning poker for the items near the top of the backlog, where the conversation matters most. Mountain Goat Software puts a typical affinity session at around 60 items an hour, against a coaching target of about 20 an hour for planning poker.

When to use affinity estimation or the bucket system

Both techniques shine when the alternative is days of planning poker. Typical moments:

• A new product backlog. After a story-writing workshop, you have 80 cards and no sizes.
• An inherited backlog. A team takes over a product with hundreds of unestimated items and needs to know roughly how much is there.
• Release or roadmap planning. You need a forecast for a quarter’s worth of work, not precise numbers for the next sprint.
• A fresh start on scale. After a big change in the team or the product, a quick pass over the backlog resets everyone’s sense of size.

They are the wrong tool for a handful of sprint-ready items, where the discussion is the point, and for items too vague to compare, which need splitting or clarifying first.

The pace figures come from Mountain Goat Software (affinity estimation and planning poker) and Berteig Consulting’s description of the bucket system. For how these fit alongside other methods, see our comparison of agile estimation techniques.

Note Speed may cost less accuracy than you’d expect. In a 2023 study, Marko Poženel and colleagues had eight student teams use all three methods on a university project. Once the teams were familiar with the methods, the bucket system and affinity estimation took on average half as long as planning poker, with no statistically significant difference in accuracy. It was a small study of students, so treat it as encouraging rather than conclusive.

Before you start: cards, scale and space

• One card per item. A key and a short title are enough, for example “SHOP-118 Filter orders by delivery date”. Long descriptions slow the sorting, so keep the detail one question away.
• A scale. Decide the values before you start. Your planning poker deck’s numbers, such as 1, 2, 3, 5, 8, 13, 20 and 40, keep the results compatible with later sessions; T-shirt sizes work if you want to stay rough.
• A space. A long wall or table in a room, or an online whiteboard with a lane for each size and a “to sort” pile.
• A timebox. An hour or 90 minutes is enough for most sessions. Say it up front, so nobody spends ten minutes on one card.
• A “too big” area. Epics and other oversized items go here instead of stretching the scale. An epic sitting among small stories squeezes everything else into the bottom few columns. Split them first, or size them in a separate pass.

How to run affinity estimation, step by step

There are several variations. This is the classic silent-sorting version, with a simple rule from Mountain Goat Software for stopping cards from bouncing around.

1. Explain the rules (5 minutes). The developers sort; the product owner answers questions but doesn’t place cards; nobody talks during the sort.
2. Silent sort (20 to 30 minutes). Put one card in the middle of the wall. Everyone takes cards from the pile and places each one to the left if it’s smaller, to the right if it’s bigger, or level with a card of similar size. Anyone can move a card that’s already on the wall.
3. Catch the restless cards. A card can be placed once and moved once. If anyone wants to move it a second time, it comes off the wall and goes into a discussion pile.
4. Add the scale (10 minutes). Place your size labels along the line where the natural breaks fall, smallest on the left. Expect a short discussion about where one size ends and the next begins.
5. Challenge (10 to 15 minutes). The product owner and the team walk the wall. Anyone can question a placement and the team may move a few cards, but this isn’t a second sort.
6. Record. Write each size into your backlog tool, and photograph or export the wall before anyone tidies it away.

The variant Mountain Goat Software describes first puts the size labels up before anyone starts, and people place each item straight under a value. It’s quicker to set up; the trade-off is that people think in numbers from the first card rather than in pure comparisons.

How to run the bucket system, step by step

The bucket system, as described by Berteig Consulting, suits 5 to 25 people and from about 40 to 500 items, and takes 30 to 90 minutes. Its trick is to calibrate the scale together, then divide the work.

1. Lay out the buckets. Use cards labelled 0, 1, 2, 3, 4, 5, 8, 13, 20, 30, 50, 100 and 200, spread along a wall or a long table with room beneath each one.
2. Anchor the scale. Pick an item at random and put it in the 8 bucket. It’s an arbitrary reference point; if the item is obviously tiny or huge, pick another.
3. Calibrate together. Pick a second item at random and discuss it as a group, sizing it against the first. If it’s about half the size, it goes in the 4 bucket. Everyone should agree. Repeat with a third item.
4. Divide and conquer. Share out the remaining items equally. Each person places their items in buckets on their own, without discussion. Anyone who can’t judge an item passes it to someone else.
5. Sanity check. Everyone walks the buckets, looking for items they know or care about. If something looks misplaced, they raise it with the group. The facilitator keeps these discussions few and short.
6. Record the numbers. Each item takes its bucket’s number, written into the backlog tool.

Berteig’s buckets include 4 and 30, which many planning poker decks lack. If you plan to follow up with planning poker, use your deck’s values as the buckets instead, so an item can move between techniques without a translation step.

To picture the scale: nine people sizing a 180-item backlog for a banking app’s redesign calibrate on three items together, then take 19 or 20 items each. The slow, discussed part of the session covers three items plus whatever the sanity check turns up.

Why silent sorting works

The silence feels odd the first time, and it’s deliberate. Mountain Goat Software gives the main reason: it keeps the meeting moving and stops the first person’s explanation from anchoring everyone else. If the tech lead says “that’s trivial, it’s one endpoint”, the cards that follow will cluster around that view whether or not it’s right.

Silence also evens out influence. A quiet developer moving a card says as much as the most senior person in the room, and the cards carry the signal: a card nobody moves is agreed, a card moved once shows mild disagreement, and a card that would need a second move needs a conversation.

A few facilitation tips:

• Allow written questions. If someone doesn’t understand a card, they write the question on a sticky note or in the chat, and the product owner answers in writing too.
• Watch for card wars. Two people moving the same card back and forth is your cue to pull it into the discussion pile, not to referee on the spot.
• Keep the discussion pile small. If a third of the wall ends up there, the items probably weren’t ready, and more refinement will help more than more estimation.

Following up with planning poker

A big-batch technique gives you a sized backlog, not a finished one. Plan the follow-up:

• The discussion pile goes to planning poker. These are the items the team understands differently, which is exactly what planning poker is for. They don’t need to be done the same day.
• The next sprint or two get a second look. When items reach the top of the backlog, check them in refinement. If the team’s understanding has changed, estimate again with planning poker; if not, keep the number.
• Big buckets mean split, not estimate. Items in the 20, 40 or 100 bucket are too uncertain to plan with. Split them before they come anywhere near a sprint.
• Agreed items become yardsticks. Items that settled quickly make good reference stories for the planning poker sessions that follow.

You don’t need to re-estimate the whole backlog with planning poker. The point of the fast pass is that most items are sized well enough for ordering and forecasting, and only a few deserve the slower treatment.

On planningpoker.org When you follow up with planning poker in sessions on a Laya board or a Jira board, a free Laya account adds Reference Issues and Matching Issues, which show items your team has already estimated, so new ones can be compared with agreed sizes. If your sort used T-shirt sizes, there’s also a T-shirt deck mapped to numbers (XS=1, S=2, M=3, L=5, XL=8, XXL=13).

Remote variants

Both techniques work online with the same rules. Mountain Goat Software notes that remote teams can run affinity estimation on an online whiteboard or in a backlog tool. A few adjustments help:

• Prepare the board. Create a lane for each size, or one long open line for the silent sort, plus a “to sort” area holding every card, a discussion pile and a “too big” area.
• Take turns in order. On a shared board, five people dragging cards at once creates chaos. Post a turn order in the chat and let each person place or move one or two cards per turn, or give each person a batch, bucket-system style.
• Keep the silence, keep the call. Stay on the call with microphones muted during sorting, so questions can still go to the product owner in the chat.
• Show a timer. It replaces the sense of progress a room gives you, where everyone can see the pile shrinking.
• Turn cameras on for the review. The challenge and the sanity check are conversations, and they go better when people can see each other.

A distributed team can even run the sort asynchronously: leave the board open for a day, ask everyone to place and move cards in their own time, then hold a short live review of the discussion pile. Our guide to remote planning poker covers the call etiquette that applies here too.