Your go-to tech blog

From Rules of Thumb to Flow Data: How Teams Now Set Initial WIP Limits in Kanban

The board has 23 cards in “In progress” and the team has five developers. Every standup sounds busy. Nothing has reached “Done” in nine working days, and the product owner has started asking, carefully, whether the estimates were wrong. The team lead has read enough to know the answer is a WIP limit. What she does not know is what number to write at the top of the column on Monday, and every source she checks gives a different rule of thumb.

That is the most common way teams meet the question of how to set an initial WIP limit: already in trouble, with a board full of half-finished work and a vague memory that the limit should be “team size plus one” or “about two per person”.

Those rules are not useless, but practice has been moving away from them. More teams now start from what their own board says, check the number against a simple law of flow and treat the first limit as a hypothesis to tighten over a few weeks.

What a WIP limit actually limits, and why the number matters

A work in progress limit caps the number of items allowed in a stage, or across the whole board, at one time. When the cap is reached, nobody starts anything new; people help finish what is already there. The slogan is “stop starting, start finishing”.

The reason the number matters comes from Little’s Law, a result from queueing theory proved by John Little in 1961. Applied to a Kanban board, it relates three long-run averages:

Average cycle time equals average WIP divided by average throughput.

If the team above carries 23 items in progress and finishes about 4 items a week, the average item takes 23 divided by 4, or roughly 5.75 weeks, from start to finish. Cut WIP to 8 without changing throughput, and the average falls to 2 weeks. Nobody has to work faster. Items simply wait less.

The law has conditions. It holds for averages over a period in which the system is roughly stable: arrivals and departures balance, items that start eventually finish, and the average age of items in progress is not growing. A team whose board keeps swelling does not satisfy those conditions, which is exactly why its cycle times feel unpredictable. A WIP limit is the tool that brings the board back into the stable state where the law applies and forecasts become possible.

This is also why the trend has shifted. The community Kanban Guide first published in 2020 frames Kanban around four flow measures: work in progress, throughput, work item age and cycle time. Once a team tracks those, the first WIP limit can come from data the team already has rather than from a rule written for some other team.

Three ways to choose the first number

There are three common methods. They often give different answers, and the comparison is informative in itself.

Method 1: rules of thumb based on team size. The classic suggestions are team size, team size plus or minus one, or 1.5 times team size for the whole in-progress area. For the five developers, that gives somewhere between 4 and 8. The advantage is speed. The weakness is that the rule knows nothing about how work actually flows on this team, whether people pair, or how large the items are.

Method 2: measure the WIP you already carry. Count the items in progress every day for two weeks, or reconstruct the counts from the tool’s history. Suppose the daily counts for the team average 21, with a low of 17.

Set the initial limit at or just below the current average, here 20, and plan to lower it in steps. This sounds unambitious, and that is the point: a limit far below current WIP forces the team to stop starting for days or weeks while it drains the backlog, and many teams abandon the limit before that happens.

Method 3: work backwards from a target cycle time with Little’s Law. Look at throughput over the last 6 to 8 weeks. Say the team finished 32 items in 8 weeks, about 4 per week. The product owner wants most items finished within 2 weeks. Rearranging the law, average WIP equals throughput multiplied by average cycle time: 4 times 2, or 8. That is the WIP at which a 2-week average becomes plausible, assuming throughput holds.

In practice, combine them. Method 2 tells you where you are, method 3 tells you where you are heading, and method 1 is a sanity check. For this team the plan becomes: start at 20 (just below the current 21), lower by 2 to 3 items every week while watching throughput and item age, and aim to reach roughly 8 to 10 within a month or so. If throughput drops sharply as the limit falls, stop tightening and look for the bottleneck the limit has exposed.

One caution about method 3. Throughput measured while WIP was very high is often depressed by context switching, and it may rise as WIP falls. That is good news, but it means the target WIP should be recalculated after a few weeks with the new throughput.

A decision tree for choosing the method and the scope

Two choices need to be made together: which method sets the number, and what the limit applies to. Work through these questions in order.

  1. Does the team have at least 6 weeks of history in a tool that records when items start and finish? If no, use method 1 for a starting number, begin recording daily WIP immediately and revisit after 4 weeks. If yes, continue.
  2. Is current average WIP more than twice what method 3 suggests? If yes, start from method 2 and step down gradually; a sudden cut to the target will stall the board. If no, start close to the method 3 number.
  3. Do items move through clearly separate specialist stages, for example development, code review and testing done by different people? If yes, set per-column limits, starting each at its current average, because the bottleneck is likely to be in one stage. If no, a single limit across the whole in-progress area is simpler and usually enough.
  4. Does one stage, such as code review or testing, regularly pile up? If yes, give that stage the tightest limit relative to its current WIP, so the pressure lands where the queue forms.
  5. Is the work a mix of very different item types, such as small bugs and multi-week features? If yes, consider separate swimlanes with their own limits, or a combined limit plus a small expedite lane. If no, one limit is fine.
  6. Do people regularly work on more than 2 items each at once? If yes, add a soft personal guideline of 1 or 2 active items per person, on top of the board limit. If no, skip personal limits.

Most teams end up with a whole-board or two-stage limit set from method 2, with a target from method 3 written next to it so everyone can see where the steps are heading.

The first four weeks with a limit

Week one is the uncomfortable part. When the column hits its limit, someone will have a free afternoon and nothing new to pull. The rule is to look right to left across the board: can you help review, test, unblock or pair on an item closer to done? Teams that skip this conversation simply raise the limit, and the exercise ends.

Agree in advance what happens when the limit is reached, and write it on the board. A typical policy: no new item starts while the column is full; the person who frees up first checks blocked items, then review, then testing; anyone who wants to exceed the limit raises it at the next standup, and the team decides together.

Track three things weekly. Throughput, to confirm the team is not finishing less. Work item age, meaning how long each item currently in progress has been there, because a falling limit should make the oldest items move first. And how often the limit was hit. If the limit is never hit, it is too high to change behaviour. If it is hit every day and throughput has not dropped, it is doing its job.

Adjust in small steps. Lowering by 1 to 3 items at a time, once a week or once every two weeks, gives the team time to learn new habits such as swarming on a single item. Raise the limit only with a reason, for example a new team member joining, and never as a reaction to one bad week.

By the end of four weeks the team in the opening example might reasonably expect WIP around 10 and noticeably shorter cycle times, though the exact result depends on the bottlenecks the limit exposes. Those bottlenecks are the real output of the exercise. A WIP limit does not fix a slow test environment or a single overloaded reviewer; it makes them visible enough that someone has to.

Where WIP limits are heading, and how they come up in interviews

The direction of travel is clear enough. Teams are moving from fixed rules of thumb toward limits derived from their own flow data, reviewed regularly, and paired with service level expectations such as “85% of items finish within 10 days”.

Tools increasingly show cycle time scatterplots and item ageing charts by default, which makes method 2 and method 3 available to teams that would never have built the charts by hand. The limit itself is becoming less of a fixed number and more of a dial adjusted against measured flow. The counter-trend is worth noting too: some teams drop explicit numeric limits once their habits are established and manage by work item age alone, pulling the oldest item forward every day.

The question comes up regularly in interviews for Scrum Master, agile coach, delivery manager and engineering manager roles: “How would you introduce WIP limits to a team that has never used them?” or “How do you choose a WIP limit?” A strong answer has four parts: measure current WIP and throughput first; explain Little’s Law in one sentence and use it to set a target; start near current WIP and tighten in steps rather than imposing a sharp cut; and describe the team policy for when the limit is hit.

Interviewers often follow up with “What if the team pushes back?” The good answer is that pushback in week one is expected, and that the response is to look at what the limit exposed, not to abandon it. Naming the conditions under which Little’s Law holds, briefly, signals you understand it beyond the formula.

Questions teams ask about initial WIP limits

Should blocked items count against the WIP limit? Yes. Keeping blocked items inside the limit creates pressure to unblock them. If they are excluded, the board fills up with stuck work that nobody chases, and the limit stops reflecting real load.

Do WIP limits work for a single person? They do. Personal Kanban, a term popularised by Jim Benson and Tonianne DeMaria Barry, applies the same idea to one person’s tasks, and a limit of 2 or 3 active items is a common starting point.

What about urgent work that arrives when the limit is full? Create an expedite lane with its own limit, usually 1. Anything in it can bypass the main limit, but if the lane is used more than once or twice a week, the definition of urgent is too loose.

Can a Scrum team use WIP limits? Yes. Scrum does not prohibit them, and a limit on in-progress items within the sprint often stops the pattern of everything being 80% done on the last day.

If your team has at least six weeks of history, set the first limit slightly below your current average WIP, calculate the Little’s Law target from your throughput and step towards it by 1 to 3 items a week.

If you have no history, start with team size as the whole-board limit and begin recording daily WIP on day one. In either case, do not set a limit at all until the team has agreed what it will do when the limit is hit, because that agreement, more than the number, is what changes how work flows.