Kanban WIP Limits: Choosing Numbers That Actually Hold

The Illusion of 'Perfect' Numbers

Setting Work In Progress (WIP) limits feels straightforward: when you hit the cap, you stop pulling. Yet, too often, these limits become decorative—a number on a virtual board that nobody dares move for fear of slowing the perceived pace. The common mistake organisations make is using WIP limits as a measure of capacity, when they should instead be used as a measure of system health.

Think of a recent case: a remote marketing team tracking asset localisation. They set a WIP of 4. However, three complex legal reviews entered the queue, causing the team to halt all active drafting (the process bottlenecks). The apparent 'limit' was fine, but the system was broken because the intake process (the dependency pipeline) was ignored.

Actionable Calibration Steps

To stop guessing, focus on data flow, not task count.

1. Measure Lead Time Variation: Track how long the average item takes to move from 'In Review' to 'Done' when the WIP limit is hit versus when it is comfortably below the limit. Huge swings signal a bottleneck, regardless of the number you picked.
2. Model Dependencies (The Remote Factor): If one resource group (e.g., Legal in London) is offline for 4 hours due to time zones, your WIP calculation must factor in the absolute working hours available, not just the number of people. Don't calculate capacity assuming 100% overlap.
3. Phased Implementation: Start conservatively. If the board has 5 columns, impose WIP limits on the two riskiest columns first, say, 'Client Review' and 'Technical Sign-off'. Let that stabilize at 80% adherence for two weeks before touching other queues.

Never treat WIP as a fixed rule; treat it as a hypothesis you test with real, observable flow data. Consider alternatives to strict board caps—perhaps integrating mandatory ‘Dependency Check’ steps before an item can enter a bottleneck column.

Takeaway

The goal of the limit is not to halt work, but to expose constraint. When the team feels the natural resistance of the limit, they are signalling exactly where process improvement must occur.

*