Vicky Singh

Hi Kiran This is a very relevant discussion going on. Here are my drops in the ocean: 1. The first element is to decide which parameter is being measure. If a rating on say a five-point scale is being given, it will be ordinal data and hence fall under Attribute R&R. If the measure is only good/bad, then its binomial data and hence also fall under Attribute R&R. However, if scores (numbers) are being given in a parameter based on the evaluation, then the overall score will fall under Variable R&R. 2. Let us take only one element of the call monitoring i.e., the quality analyst is checking whether the resolution took place or not based on the conversation. So this is pass/fail and hence binomial. 3. For the above scenario, you should take minimum 30 calls. Lower samples reduce the resolution. e.g., with 10 samples, every sample going wrong impacts overall score by 10%. Typically 2 trials per sample per analyst and 3 analysts would be taken. Here one must note that if there are less than three analysts for that team, only those should be taken. For larger numbers, a thumb rule of 20% is taken. Also, please note, as the number of analysts will increase, the overall score will have higher opportunity of going down. 4. Let say you have taken 30 calls, 2 trials each and 3 analysts. Since this is a R&R exercise, the same sample should be available for the analysts for evaluation. This could be achieved by recording such calls and then playing the call for each reading of an analyst. It should be attempted that the calls are played at some gap or specific info in the call should be removed which could create bias from memory rather than fresh evaluation. 5. Each analyst listens to a call and marks pass/fail as per his/her evaluation. 6. Once the complete data is captured, it is analysed through software like excel or Minitab. 7. The challenge here could be to understand the Standard. This typically is achieved through a) having an expert who is considered as the final authority from a knowledge perspective checking with the customer, which most people avoid c) post the data collection, all analysts come together and deliberate till they reach a consensus on the result (pass/fail), which is then treated as the standard It is not mandatory to have a standard. However, this helps in accuracy check i.e., calibration of the system. Trust this helps. Regards