What a WIP limit is and how to find your team’s
A short guide for teams that work all day and feel they get little done. It explains what work in progress is, what limiting it is for, and how to choose your team’s number, with data from the game Mosca del WIP.
What WIP is and what a WIP limit is
WIP stands for work in progress, which is what a team has already started and not yet finished. A WIP limit is the maximum number of things in progress at once that the team agrees to respect, written on its board.
The idea comes from kanban. The limit goes above the In progress column, and when that column is full nobody starts anything new until what is already in there is finished or unblocked. Every card on the board counts, not every email or every small to-do.
In the game Mosca del WIP (mosca means fly in Spanish) you pick that number before the day starts, from 1 to 10. The fly works alone, and what happens to a team is for you to measure.
Why being busy isn’t the same as making progress
Because there are two ways to be efficient. One is keeping everyone busy, resource efficiency. The other is having each thing move without waiting, flow efficiency. With many things open, people are busy all day and each thing spends almost all its time waiting for its turn.
In the game, with 9 tasks at once the fly was busy almost all day and finished half as many as with 2.
| What happened in the day | 2 tasks at once | 9 tasks at once |
|---|---|---|
| Tasks finished | 18 | 9 |
| Time per task, from start to finish | 1 h | 8 h 54 min |
| Minutes of every hour that really get work done | 50 | 25 |
| Share of the time each task is moving | 47% | 9% |
That last row is flow efficiency, the share of the time an open thing actually moves forward. In the game, with 2 the fly ended the day Calm and with 9, Overwhelmed. A limit helps you finish more, keeps work moving instead of waiting, and means less overload. Your team’s own test will show how much it gains.
How to choose your team’s WIP limit
Start with a little less than what you have open today. Count the things in progress on the board, agree on a somewhat lower number, uncomfortable but possible, and try it for two weeks. There is no right number for every team, because it depends mostly on how much your work waits on other people.
The game doesn’t give a number for teams. In the simulation, for a fly working alone, the best limit landed between 2 and 4 in 16 of the 17 cases with waits that we tested, and at 1 when nothing gets stuck. A team, unlike the fly, can unblock what is waiting, and that is why its limit is its own.
Why it depends on waits
Waits push the limit up, because if something gets stuck it helps to have another thing at hand, and the cost of switching from one thing to another pushes it down. In the simulation, when the fly waits a lot on others, the best limit goes up, and when she waits little, it goes down. In a team, the first step is to go after the waits, not to raise the limit.
Reducing waits lets you lower the limit. Raising the limit doesn’t reduce the waits.
The two-week test
Write the limit on the In progress column and note on each card the day it started and the day it finished. After two weeks, compare how many things got finished and how long each one took with the two previous weeks, and decide whether to keep, lower or raise the limit. What’s started takes time to clear. In the game, when the fly drops from 10 tasks to 2 at 12:00, she only gets back to two open at 14:35. The 60-minute workshop takes the team through these steps with the game live. Its material is in Spanish for now.
What to do when something is left waiting
Go after the wait instead of opening something else. Help unblock what is stuck, get the missing answer or finish something already started. If the team opens something new every time something waits, the limit stops existing.
DORA, Google’s research program on software teams, advises not relaxing the limit when people have nothing to do, and using that time to unblock what’s waiting (DORA, Work in process limits). It helps to agree on it before it happens, and in the workshop the team writes it on its sheet.
What about urgent tasks?
In the game, the “Send an urgent task” button sends one to the fly. The urgent task jumps over the limit, one at a time, and she does it first.
In the simulation, with limit 2, each urgent task a day cost one regular task, from 18.0 finished with no urgent tasks to 17.1 with one a day. Raising the limit doesn’t help the urgent task, because with more things half-done it waits too. The full numbers are on page 11 of the report and in the second data table.
In a team it also helps to let urgent work in one at a time and over the limit, with its own card on the board. If it arrives every day, that is something to look at in the two-week review. suggestion
Little’s law, in plain words
Little’s law says that the time each thing takes equals how many are open divided by how many get finished per unit of time. At the same pace, fewer things open means less time for each one. If a team finishes 5 things a week and has 10 open, each one takes 2 weeks on average, and with 20 open it takes 4.
It is a theorem John Little published in 1961, and it holds for any stable system (Little, 2011). In the game it holds on its own, with 1 hour per task with 2 and almost 9 hours with 9. That fewer things also get finished with many open comes from the cost of switching tasks and from stress, which are assumptions of the game.
Common mistakes when setting a limit
The most frequent is raising the limit as soon as someone is left waiting. The wait is still there, with more open things on top.
- Setting a number and never looking at it again. Put the review in a meeting you already have.
- Counting every email or every small to-do. The limit is about the cards on the board.
- Looking for the right number before trying. One a bit lower than today’s, tried for two weeks, teaches more.
- Expecting results on day one. The old things come out first, and only afterwards the new pace shows.
- Using the limit to measure people. If it is used for that, people hide work.
What the studies with real people say
No study with real people gives a number of things at once. The ones we found point to the same thing. Past a certain point, with more things open each one takes longer, and sometimes fewer get finished.
- Judges in Italy. Those who by lottery carried more open trials at once took longer to close them, with the same quality (Coviello, Ichino and Persico, 2014 and 2015).
- Recruiters. A little parallel work helped, but past a point they finished less (Aral, Brynjolfsson and Van Alstyne, 2012).
- Software teams. Limiting work in progress is one of the habits of high-performing teams, especially together with a visible board, and on its own it predicts little (DORA and Forsgren, Humble and Kim, 2018).
- A team with a board and limits. At a Norwegian software company, the time to fix a bug dropped from 12 to 5 days. It is a single case (Sjøberg, Johnsen and Solberg, 2012).
Only the one with the judges comes close to an experiment. The others show that things go together, without proving the cause, and the report, p. 8, tells each one in more detail.
The game’s data, from 1 to 10 tasks
In the game, with 2 tasks at once the fly finishes more than with any other number, and up to 3 she ends the day Calm. From 4 on her stress goes up, and with 9 she finishes half as many as with 2.
| Tasks at once | Finished per day | Time per task | Productive, of every hour | Idle, per day | Cigarettes | Coffees | Stress at day’s end |
|---|---|---|---|---|---|---|---|
| 1 | 15.9 | 34 min | 44 min | 134 min | 0 | 6 | Calm |
| 2 · finishes the most | 18.0 | 1 h 00 min | 50 min | 20 min | 0 | 5 | Calm |
| 3 | 16.9 | 1 h 36 min | 47 min | 2 min | 0 | 4 | Calm |
| 4 | 15.4 | 2 h 20 min | 43 min | 0 min | 1 | 3 | Busy |
| 5 | 14.0 | 3 h 14 min | 39 min | 0 min | 3 | 1 | Busy |
| 6 | 12.3 | 4 h 24 min | 34 min | 0 min | 6 | 0 | Stressed |
| 7 | 11.2 | 5 h 40 min | 31 min | 0 min | 8 | 0 | Stressed |
| 8 | 10.0 | 7 h 11 min | 28 min | 0 min | 11 | 0 | Overwhelmed |
| 9 · overload | 9.1 | 8 h 54 min | 25 min | 0 min | 13 | 0 | Overwhelmed |
| 10 | 8.2 | 10 h 34 min | 23 min | 0 min | 15 | 0 | Overwhelmed |
In the game, average of 200 days per number of tasks (simulation, seed 2024). Time per task in office hours, from start to finish. Shaded, 2 (green, finishes the most) and 9 (red, overload).
The sign in the game’s office says “up to 3” because it measures stress. The table measures what gets finished, and there 2 wins by one task a day, which comes from the cost of switching tasks, an assumption of ours.
The data are free to use under a CC BY 4.0 license, crediting Felipe Fernández and mosca.deliverycraft.ai. They come from a single fly and don’t measure people or teams.
- Download the data (CSV, in English)Ten rows, one per number of tasks.
- Download the data (CSV, in Spanish)The same rows.
- Download the urgent tasks data (CSV, in English)Fifty rows, by urgent tasks a day and limit.
- Download the urgent tasks data (CSV, in Spanish)The same rows.
Where each rule of the game comes from is in Appendix B of the report, p. 14.