Maybe I'm not making myself clear enough. So lets do it this way: If you have a 1000 question true/false exam. How big of a sample size, of the 1000 questions would you need to accurately represent the whole 1000 question exam? 30? 100? 500? Help me out fellas

Now that just made it a whole lot worse.

If your fail rate is constant and a truly random sample of 30 out of 1000 is taken, 30 should give quite acceptable estimate of the fail rate. Now how big a sample size your boss needs depends on what he/she finds an acceptable fail rate.

If he feels that only 1% of the calls may “fail” then he can probably take even less than 30. If he feels that your fail rate should be around 50%, he will need to take more samples to be just as sure.

For instance: You're about to buy a truckload of very cheap apples, 500 of them. The salesperson tells you 1 out of every 100 apples are rotten, you then pick up and inspect 10 at random. If you find 5 out of 10 to be rotten you will doubt his claim. You will certainly not believe him if he says that you “by chance” took the 5 rotten ones.

Now if he told you 40 out of 100 are rotten, and you again found 5 (out of 10) you could not as easily have refuted his claim and you would need to inspect more.

The same rational holds for your call centre fail rate, if standards are high they only need to sample very little to find fault with certainty. So the sample size of 30 probably has everything to do with what your boss feels is an acceptable fail rate.

Nobody on this forum can tell you if your boss is wrong if you don’t know the fail rate standard.

How can you tell if the apple-salesman is wrong without knowing his claim?

vinux actually said it all in the first reply:

“It is actually depending on your limit of margin of error”