Introduction: The hidden pulse of productivity

WIP and buffers: define before measuring

Work in process includes units inside the chosen boundary, whether being processed, waiting, or transported. A buffer is a planned reserve serving a purpose. A queue alone does not prove excessive inventory, high utilization, or poor planning.

Separate location, quantity, state, and function. One reserve can protect a machine and support transfers; those functions are not automatically mutually exclusive categories. Define how batches, rework, and incomplete material are counted.

What observations can measure

A visit can record queue presence, quantity, and resource state. The fraction of visits with a queue estimates time presence, not average part count unless parts are counted. Representative inventory counts can estimate average inventory using analysis suited to quantities, not automatically a binomial formula.

Observing workers measures worker states. Combining their activity proportion with a timed operation does not reconstruct an item's end-to-end flow time.

Little's law relates averages

Average WIP = effective throughput × average flow time.

The quantities need compatible boundaries, units, and temporal flow balance. Flow time includes waiting and processing. Little, 1961 establishes the relationship, not a minimum buffer or safety percentage. Sensors are not required: reliable manual records can measure arrivals, departures, and quantities.

Changing WIP can change flow time or output. Do not hold two quantities arbitrarily fixed and call the third optimal.

Sample and round example

Assume queue presence at one location is the target, with p₀=0.60, z=1.96, and e=0.05. The approximation n=z²p₀(1−p₀)/e² gives 368.7936, rounded upward to 369 independent observations. NIST describes the binomial model.

A round covering twenty-five points does not give twenty-five readings at each point. Fifteen rounds leave only fifteen observations per location. They may supply 375 records overall, but aggregation requires a defined population, weights, and correlation checks. It does not provide the preceding precision at every location.

Test the protection required

1. Identify the interruption to be buffered and the required output.
2. Measure arrivals, service, stops, quantities, and flow times.
3. Propose buffer location and size with explicit assumptions.
4. Trial changes while monitoring starvation, blocking, good output, and lead time.
5. Review effects across product mixes and volumes.

There is no universal buffer of 50% of upstream lead time or occupancy threshold establishing necessity. Persistent queues can have different causes.

Relationships not to assume

High WIP does not universally correlate inversely with OEE. OEE uses equipment and production data; human activity time is not a replacement. An assumed rework share does not establish an upstream quality failure: confirm unit states against records.

Distinguish inventory reduction, potential capacity, and monetary savings. Assess a trial that reduces queues against quality, safety, and delivery performance as well.