INTRODUCTION -- You can design an acceptance sampling plan to signal when to undertake a total physical inventory count. Compared to taking a total inventory count at a fixed period (annually), sampling can provide you with more accurate record counts, cause less disruption of operations, and lower costs. The sampling plan involves a relatively small number of items at frequent periods.

The sampling plan itself is developed based on the Operating Characteristic Curve (OC-Curve) as a criterion to ensure that the sample size is sufficient and that the accept/reject decision rule matches your operating standard for minimum accuracy (ie. 95%, 98% etc.).

With sampling you do not have to physically count the items for every part number every year, as long as the estimated accuracy of recorded counts is sufficient and the statistical margin of error of the estimate is sufficiently small. I know that in some leading companies, the accountants support this sampling practice.

By counting small samples frequently rather than performing total inventory counts annually, you can detect and correct record-keeping system problems early - using control charts.

USE OF CONTROL CHARTS -- I favor two types of control charts because there are two purposes:

1) Acceptance control chart, where the limits are equal to the decision limits of an acceptance sampling plan. This chart triggers the accept/reject decision.

2) Standard "three-sigma" control chart for the average and standard deviation, targeted and using the time-sequence rules for detecting special causes. Use this chart to analyze the inventory system for continuous improvement.

3) Use the sampling plan software of H & H Servicco Corp. to ensure that you use the right sample size for these charts to achieve your accuracy goals.

THE RIGHT SAMPLE SIZE -- The OC-Curve approach to sampling uses as a criterion the probability of acceptance of populations of various accuracy levels. There are several strategies to attain maximum precision for decision with minimum sample size:

To minimize sample size and the amount of effort, use sequential sampling. Start with a fixed-n plan and after some experience, switch to sequential for further gains)

Define your test statistic as a variable rather than an attribute. This will further minimize sample size and effort.

There is more discussion on these strategies and others on www.samplingplans.com/modern3.htm

TYPICAL LEVELS OF ACCURACY -- I found these example targets published in the literature:

1) 95% accuracy of computer count versus physical count, by items, within a given location. Leviton Manufacturing Company, P&IM Journal Q2, 1990.

2) 98% accuracy, Litton Integrated Systems Technology, P&IM Journal, Q4 1985

CHOOSING ACCURACY GOALS FOR YOUR LOCATION -- As a starting point, you should analyze your records of past total inventories to see what levels of percent accuracy you are dealing with. This could be interesting.

Take into account the main motivations: cost reductions, smoother operations, improved accuracy of book value for better planning, for accounting purposes, etc.

Once you establish an accuracy level, you must map it onto the sampling criteria. For example, if the accuracy goal is 95%, choose between these three interpretations, which specify three different OC-Curves:

Examples: (all 95% accuracy)
A) AQL=95%, AC=93%, RQL=91% AQL plan like mil-std-105

B) AQL=97%, AC=95%, RQL=93% Ac is the decision limit, easy to explain.

C) AQL=99%, AC=97%, RQL=95% Very likely to reject if accuracy is 95%.

Some people prefer to just choose a sample size and acceptance number, and then calculate the OC-Curve, AQL, and RQL from that. In either case, the software programs of H & H Servicco Corp. will do the calculations that make this easy.

CHOOSING A TEST STATISTIC -- Consider what measure of error to use for the sampling plan: error count, percent accuracy, dollar percent error.

You might use use error count converted to error in reorder time. This would be a variables measure, which does not require a large sample size, compared to attribute measures.

For engineering purposes, you would be interested in the absolute value of the errors. For accounting and tax purposes, the accountants would be interested in the overall value of the inventory - without absolute values.

SELECTION OF SAMPLES -- You could classify the population of part numbers into groups that have different inventory procedures. Each classification could be treated as a separate population.

SOFTWARE TO DESIGN SAMPLING PLANS -- H & H Servicco Corp. has software to develop sampling plans based on the OC-Curve: (Sampling Plans for Attributes-TP105, and Sampling Plans for Variables-TP414).

Additionally, the Audit Sample Planner software has the capability selecting a series of random samples of items, where each sample does not contain an item of any other sample. It prints these item numbers on a data sheet for physical recording.

SUMMARY OF BENEFITS -- The sampling approach can lead to sizable cost reductions. You conduct a total inventory only when the inventory book counts are sufficiently inaccurate to justify it. Combined with the application of control charts to the small but more frequent samples, you will discover special causes and improve the accuracy of the inventory recording process. And the smaller samples are less disruptive of operations.