Measurement
We have no data. Where do we start?
The most common starting position in process work, and the one every method assumes you are not in. A week of deliberately cheap measurement gets you further than a data project.
Almost every technique in this field opens by assuming you have data. Cycle times, arrival rates, a distribution of handling times. The position most people are actually in is that the process runs, it runs badly, nobody wrote any of it down, and the system it runs on records the end and not the middle.
The instinct at that point is to fix the data first: instrument the process, add tracking, wait a quarter. That is usually the wrong call, because it costs a quarter and because a process nobody has mapped will get instrumented in the wrong places.
Here is what a week of cheap measurement gets you instead.
Day one: follow one item
Not a sample. One. Pick something typical, follow it from the moment it arrives to the moment it leaves, and write down every point at which it changed hands or waited.
You will get roughly the shape of the process, the number of handovers, and one very large number: the difference between how long it was worked on and how long it took. That ratio is almost always shocking the first time and it is the finding that gets everybody’s attention. Ninety seven percent of that form’s life is waiting is exactly this exercise, done once, on one claim.
Cost: a few hours. Value: the map, and the argument for doing the rest.
Days two and three: count arrivals
Not a study. A tally sheet, or a query against whatever recorded that something came in, by hour of day and day of week, for as long back as the record goes.
This is usually the easiest data in the building to get, because arriving is the one event almost every system does record. It gives you the demand pattern, which drives everything downstream, and it very often explains the complaint on its own: work arrives in a burst on Monday, or after six in the evening, or in the last three days of the month.
Cost: an afternoon. Value: the arrival pattern, which no amount of process improvement can be planned without.
Day four: time the slowest step, thirty times
Just the one step, the one everybody already believes is the problem. Thirty observations, timed the same way, every one recorded including the ridiculous outlier.
Thirty is enough to see the shape, which is the thing that matters, and it is few enough to do in a day. You are not looking for the average. You are looking for the spread and for whether the plot has one hump or two, which is what your step does not take eight minutes is about.
Cost: a day. Value: the variability, which is the other half of every queueing result.
Day five: count what is waiting, twice a day
Walk the process morning and afternoon and count what is standing between each pair of steps. Do it for a week if you can, twice a day, on a piece of paper.
This is the crudest measurement on the list and it is the one that most reliably finds the constraint. The pile is not always in front of the constraint, but the pile that is bigger every day is, and two counts a day for a week tells you which is which.
Cost: twenty minutes a day. Value: where the constraint actually is, as opposed to where everybody says it is.
What this deliberately does not give you
The limits matter here, because the next question is usually whether this is enough to model on.
It is not enough to predict with. Thirty observations of one step and a week of counts give you the shape of the problem, not a validated model. Anyone quoting a five percent improvement from this data is overreaching.
It has no seasonality in it. A week of arrivals in September says nothing about December. If the process has an annual shape, this will mislead you about the peak, and you will only find that out at the peak.
And it is biased by being watched. People work differently when timed. It is a real effect, it is usually smaller than the effect you are looking for, and pretending it away helps nobody.
What to do with it
Do the five days before commissioning anything. They cost a week and they are what tells you which quarter-long instrumentation project is worth running, if any.
Write the numbers on a wall, not in a document. The audience for this is the people in the process, and their reaction is the second finding. They will tell you which of your numbers is wrong, and they will usually be right.
And resist the pull towards precision. The gap between no data and rough data is enormous. The gap between rough data and good data is real but much smaller, and most processes are so far from optimized that the rough numbers point at the same actions the good ones would.