Kkanbanwall

Why Kanban works · chapter 4 of 5 · 6 min

The slow step

Every workflow has one step that sets its pace. This chapter names it, shows the three ways to make it go faster, and how to keep it from ever standing idle.

  1. Overview
  2. 1
  3. 2
  4. 3
  5. 4
  6. 5

The workflow is the small one from the last chapter: Specify takes a day, Implement takes a week, Validate takes a day, one person on each.

Busy people, or finished work

Everything so far cuts against an instinct that runs deep: nobody should be idle. It is why Doing had seven cards. Somebody finished something, so they started something else, because standing still looks like waste.

But the board is paid for finished work, and busy people are only a means to it. The pile on the last chapter's first board was made entirely of people being usefully busy, and it bought nothing: one item a week before, one item a week after, and a wait that grew every day. Kanban's answer, and the whole of the Toyota tradition it comes from, is to optimise the process rather than the person: an hour of someone's time is cheap, and a day the workflow loses is not.

Which turns the usual question inside out. The question is "is the thing that sets our pace ever waiting?", and if a person has to stand idle for the answer to be no, it is a bargain.

The constraint

A workflow runs at the pace of its slowest step, whatever the others do. Ours finishes one item a week, because Implement takes a week, and the quick steps around it cannot help. If you let them run anyway, they do not add throughput; they add queue, and by Little's Law the queue is paid for in response time. That was the pile, and now it has a name: Implement is the constraint.

Goldratt called the fix "elevating the constraint", and there are three ways to do it:

Five items in the long step at once. Five times the throughput, and five times its WIP.

Day 0 week 1 · no finished so far
Backlog
Specify
1 d
WIP 1
Implementthe constraint
5 d
WIP 5
Validate
1 d
WIP 1
done

an item, filling as it is worked · waiting for someone · the people on the step, lit while they have work

5.0items a weekthroughput
7.0in progressaverage WIP
1.4weeks to finish oneresponse time
100%of people's daysat work

7.0 in progress = 5.0 a week × 1.4 weeks Little's Law, measured — not asserted. It balances because it must.

What you doWhat it costsThroughputWIPResponse
Nothing changed—1.0/week2.22.2 weeks
Five people on it4 more people5.0/week7.01.4 weeks
Broken into five steps4 more people, 5 hand-offs5.0/week7.01.4 weeks
Made faster: 2 days, same personnothing added2.5/week2.51.0 weeksbest

Five people on the long step, or the long step cut into five, come to exactly the same numbers (five a week, seven in progress, 1.4 weeks to answer), but one of them needs four more people and the other needs five hand-offs. And making the step itself faster wins, with nobody added at all: less throughput than five people, and still the shortest response time of the four, because it buys speed without buying WIP. That is why a team's best investment is usually the build, the tests and the deploy.

Drum, buffer and rope

Every idea so far assumed each step takes exactly as long as it says. Drop that: an item in the long step now takes anywhere from two days to eight, which is what real work does. The board immediately stops reaching five a week, and the reason is worth saying plainly. A day the drum spends idle is gone for good. Nothing downstream can make it up, because the drum is what everything downstream is waiting for.

Drum
The slowest step, here Implement. Its pace is the board's pace, so the job of every other step is to keep it beating. The name is the point: everyone else marches to it.
Buffer
A little work waiting in front of the drum, so a slow day upstream never leaves it with nothing to pick up. It costs a slightly longer answer and buys days the drum would otherwise lose. On a board it is simply a WIP limit a size or two above the number of people.
Rope
Limits on the faster steps, holding them to the drum's pace so they cannot pile work up in front of it. Specify could produce five a week; the rope is what stops it, because that work would only wait.

So the aim is to get as close to the drum's pace as you can while carrying as little work as you can: the drum's pace is a ceiling nothing elsewhere can raise. Watch the drum on the board below: it turns red on the days it is short of work, and the badge beside the date counts how many of its days have gone that way.

Work varies by half either way now. With no slack in front of it the drum runs dry on nearly a fifth of its days, and those days are gone: the team finishes 82% of what the drum could have done.

Day 0 week 1 · no finished so far
Backlog
Specify
rope · held to 1
1 d
WIP 1
Implementdrum
the pace of the whole board
5 d
no buffer left · the next slow day upstream stops it
WIP 5
Validate
rope · held to 1
1 d
WIP 1
done

an item, filling as it is worked · waiting for someone · the people on the step, lit while they have work

4.1items a weekthroughput
6.7in progressaverage WIP
1.6weeks to finish oneresponse time
84%of people's daysat work
82%of what the drum could do5.0 a week at best
18%of the drum's dayswith nothing to work on

6.7 in progress = 4.1 a week × 1.6 weeks Little's Law, measured — not asserted. The workflow is still filling up, so it does not balance yet.

LimitsOf the drumDrum idleWIPResponse
No buffer82%18%6.71.6 weeks
Buffer the drumthe compromise90%9%8.41.9 weeks
Buffer everything91%6%10.72.3 weeks
No limits at all92%7%12.62.7 weeks

So which one should a team pick? The second: a buffer on the drum and a rope on everything else. It gets 90% of what the drum can do, within two points of the best any of these manage, while carrying three or four fewer items than the alternatives and answering a week sooner. The last two points of throughput are not worth what they cost.

And what that lower WIP buys is worth naming. Work in progress is your turning circle. Everything started and unfinished is a commitment: a customer told "it is in progress", a branch to rebase, a context somebody has to reload. When the world changes (the urgent bug, the customer who matters, the idea that turns out to be wrong), a team carrying two items can turn inside a week. A team carrying twenty-five either waits weeks for its own queue to drain, or abandons work it has already paid for. Low WIP is the ability to change your mind without throwing anything away.

David J. Anderson tells this as a walk: a man, his young daughter and a big dog. The daughter sets the pace: she is the drum. The tether on her wrist has enough slack that she is never pulled along: that is the buffer. The dog's leash keeps him from running off: that is the rope. Everyone gets home together, at the daughter's pace.

Little's Law is John D. C. Little's (1961). The theory of constraints and drum-buffer-rope are Eliyahu M. Goldratt's, from The Goal. The walk with the dog is David J. Anderson's. Flow efficiency is from Modig and Åhlström's This Is Lean. The simulations are ours, and the arithmetic is theirs.

0562