Search Network

Improving and Streamlining Quality Assurance

30 November 2011
print version share on facebook share on linkedin

Executive Summary

In the outsourcing industry, being able to provide quality assurance often means listening to samples from many thousands of customer interactions. To automate this process, whereby typical interactions are flagged, is a challenging problem, especially since the interactions are in the form of recorded conversations. This project delivers a set of representative calls for managers to review. The outcome is that the best set of calls is presented to the manager in a faster time and more accurately reflecting the operators' performance.

Challenge overview

The company receives over 25,000 calls per week on behalf of one client. The quality of these calls needs to be ensured for each operator, through their manager listening to a sample. The challenge is in finding a sample of three representative calls automatically. The first issue is to characterise each call and then to select from these descriptions, a set of calls which best represents the range that one operator may have taken. This project first automates the characterisation of call centre engagements (voice recordings) and then applies clustering techniques to identify the three calls.

Implementation of the initiative

The project involved close interaction with the management of the outsourcing company, who regularly have to make these reviews. Evaluations were designed around managers comparing selected representative calls from the system, together with a random selection. If they recognised the former calls as being good, then the clustering algorithms were correct.

Sound files were the source of all the data. These were broken down initially into segments for each of the two speakers and then characterised in terms of pitch, silence, overlapping speech, as well as overriding parameters such as duration and hold times.

Fig 1. Three clusters of similar types of call (including outliers).

Results and achievements

By applying the appropriate AI techniques from classification and clustering, we were able to make a significant impact on the company's business processes and the quality of assurance they were able to offer their clients. In the business of outsourcing, quality is often the differentiator and having this backed up by solid mathematics, enhances this even more.

This project was recognised in 2010 with an it@cork leadership award in Research &Innovation.

Lessons learned and applicability

The area of outsourcing is a rich one for AI techniques, especially when trying to model operator interactions. This project showed that an AI approach was the right one to deliver the quality assurance the company sought. Dr Dara Curran,
Research Scientist,
University College Cork

Attached Documents

PDF icon Irish Mathematics and Industry (Irish_Mathematics_and_Industry.pdf | 1.42 MB)