Principle 6: Variability (Item 10)
All systems and processes exhibit variability, which impacts
on predictability and performance.
All processes are complex networks of inter-connected activities.
Some activities have clear relationships with other activities. With
others, the relationship is obscure. Most systems are made up of many
processes, each with many activities that interact and inter-relate
in complex ways.
The total variation of the system comprises all of the variation
from all its processes. Variation accumulates from activity to activity
and from process to process.
The most important intervention points are in the early stages
of the system because reducing variation there will reduce variation
overall.
Consider this simple system that combines the output from processes,
A, B & C, which are each a series of activities.
 Each activity in this
system has variation in what it produces. This variation comes from
machines, materials, people, methods, measurements, response times,
reactions, moods, weather, supplies etc.
In each case, the next activity in line receives this variable input,
adds its own variation and passes the `accumulated' variation to the
next activity. That is, each activity accumulates variation from itself
and all prior activities. Each process accumulates variation from
every activity in the process. The whole system accumulates variation
from all its processes.
In his book The Goal, Goldratt uses three examples to illustrate
how the variations accumulate in dependent events in a process.
The first example is a line of boy scouts walking
on a hike. The scouts are analogous of dependent events in a process.
Each person cannot walk the path until the
person in front of them has walked it. The first in line must process
the trail – to walk it – before the second can set foot
on it. The entire troop must walk the trail before it is completed.
If one scout is slow, that slowness accumulates. The scouts behind
that person bank up and will not be able to complete the task (walked
trail) in the required time (specifications). If one scout is consistently
slow because of an injury, overweight or too much equipment, that
is a common cause of variation for the process. However, other variations
will occur as each person occasionally stops or stumbles.
Suppose the scouts are arranged in line from
naturally fastest walker to naturally slowest walker. If a scout stumbles,
he will not able to catch up (without running), and that time lost
will accumulate behind them. You could reduce the accumulated variation
by arranging the line from slowest to fastest. However, such an arrangement
may not be practicable.
The second example is of a statistical fluctuation
game. Arrange five bowls in a line (A, B, C, D, E). You also need
a die (one half of a pair of dice) and some matches. The idea is to
pass matches from one bowl to the next. You use the die to decide
how many matches to pass, but – and this is the sticking point
– you can only pass matches if you have them. On average (without
that rule), you should be able to move 3.5 matches per cycle or 35
matches in 10 cycles. In practice, you will move far fewer.
Cycle 1. A throws a two and puts two matches
in bowl A (down 1.5 from the expected average). B throws a four but
can only get two matches from A (so down 1.5 from the expected average).
C throws a four but can only get two matches from B (so also down
1.5 from the expected average). D and E throw ones (down 2.5 from
the expected average). So we get one instead of our expected 3.5.
But don't worry. It will average out. Just wait and see.
Cycle 2. A throws a six and puts six matches
in bowl A (up 2.5 from the expected average). B also throws a six
and moves six matches to bowl B (up 2.5 from the expected average).
That's looking better. C throws a three so can only get three matches
from B (down 0.5 on the expected average). D throws a four and takes
Cs three from this cycle and the one left from cycle 1 (up 0.5 from
the expected average). E throws a three and moves three (down 0.5
from the expected average). So at the end of cycle two we have four
matches when we should have seven.
Keep going? It just gets worse. You will find
that matches move through the system not in a manageable flow, but
in waves. Confirming the rule that:
In a linear dependency of two or more variables, the fluctuations
down the line will fluctuate around the maximum
deviation established by any preceding variables.
There is an excellent simulation of this 'dice game' at www.ganesha.com. (Don't forget to come back.)
Goldratt's third example goes on to show us that the maximum deviation
of a preceding operation will become the starting point of a subsequent
operation.
An order of 100 products is needed in five
hours. Two machines are needed to make the product. The first machine
passes parts to the second in batches at the end of each hour. Both
machines can do a maximum of 25 parts per hour so it should be easy.
Shouldn't it? The first machine makes 19, 21, 28 and 32 parts respectively
in each hour. The second machine can therefore only make 19 parts
in its first hour and 21 in its second hour because that is all
it gets from the other machine. It makes 25 parts for each of
the last two hours easily enough. However, because of the early backlog,
it ends up 10 parts short.
What if those were files instead of widgets.
What if you worked in an insurance office and your job was to process
claims. On average, you can deal with 25 claims an hour. At the end
of an eight-hour day, on average, you can process 200 files. Does
that mean the person downstream from you can also process an average
of 200 claims a day as well? No, it does not! Many things happen to
you during your day: interruptions, meetings, training, lunch, coffee
breaks. Many things happen to your files: missing data, time on the
telephone checking or confirming, or people are late getting back
to you. As a result, in some hours, you get almost no work done, but
then you make up for it during other  hours. Therefore, you complete 1,000 claims in a five
day week, or 25 an hour on average. And the people downstream? Well,
its true they do not do anything much in the hour or so following
those hours when you are not doing much. But it works out in the end,
doesn't it? No it does not. Every thing downstream is impacted by
the fluctuations upstream. Everything shows variation and this variation
accumulates.
We have mentioned some of the usual causes of variation. What of
the others.
Because of multiple dependencies between activities at the systems
level, the real picture can be very complex. Processes inside real
companies are usually not as discrete as the simple process shown
so far. In practice, Processes A, B and C usually influence each other.
In reality, everything is connected to everything else. Which is why
simplistic solutions do not work.
 For example, for the
output characteristic `service courtesy', Process A could be `recruitment
and training', Process B `communication' and Process C `recognition'.
We have several customer complaints about Debbie, who appears to lack courtesy. She appears not to like
the job and tends to get upset easily. After we chew her out, she
gets better for a few days, but then she slips back to her old ways.
She resents being underpaid what she does. Actually, she wasn't really
suitable for the job originally, but she was the only applicant. We
had a rush on when she joined, so she missed most of the orientation
program. Her supervisor has not been able to do much on-the-job training
for her. Her skills at handling angry customers are not good. She
gets angry and resentful when they complain.
She has been complaining to us that the
procedure manuals she has to follow are badly written and do not accurately
describe the job she does. Her supervisor does not appear to like
her, and she does not like him. He has real problems also. He used
to do Debbie's job. We promoted him into a vacancy. He was not an
ideal choice and does not understand that he needs to coach and not
shout.
 Where does the variation
in `courtesy' originate? Can you find a single cause that can
be fixed? The old thinking would be to "fix Debbie". Every
element in this system is contributing to variation in `courtesy'.
The activities and processes relate in complex ways and the amount
of variation increases as we move through the activity/process sequence
and with time. And as we saw from Goldratt's work, they accumulate
at the rate of the maximum deviation – not the minimum. Which
begins to explain why Murphy's Law works so well ("If anything
can go wrong it will, and at the worst possible time.")
Do you see the strong links back to Principle 4 (`To Improve the
Outcome, Improve the System')?
|