Well – that depends.
Simulations like TeamFlow seem to indicate that a WIP-limit that is slightly below the number of people in the team is ‘optimal.’
On WIP-Limit size
But then again, there are so many constraints possible. If you want to encourage pairing in a software development project (teamsize/2) – 1 might be better. If you want to allow for some slack perhaps (teamsize/2) + 1 might get you there. And of course the handling of commitment points – the points, where work enters the controlled part of the system – also matters a lot. Do you start to limit your work in progress only after the analysis has been done? Or do you start at a the conception of an idea? Do you have different limits for different types of works? And how about policies when you the exceed those limits?
On finding right numbers
For me it is not so easy to answer the question about the right limits. That’s why I find it important not focus exclusively on WIP-Limits when it comes to implementing Kanban, but also to embrace the other principles. Especially “Agree to pursue improvement through evolutionary change” together with the practice to “Improve Collaboratively, Evolve Experimentally.”
This also means, that one should follow the scientific method formulate the hypothesis about the way you want to influence the system (or process), designe an experiment to verify or falsify that assumption, execute the experiment and only thereafter implement the change. Or design a new experiment if the hypothesis was falsified.
till next time