The success of a customer service organization depends on the performance of its agents!
The devil is in the details – so what’s in the data?
In this example I’m going to don the hat of an analyst trying to investigate this critical issue of our dipping customer service satisfaction ratings since the past few months. I have been given a subset of data pulled from our case management repository to analyze the last 10,000 cases that our agents resolved.
Apart from doing some ad-hoc exploration, I want to use the self-service predictive capability of Watson Analytics to understand what really drives Service Satisfaction and more importantly, how to improve on it.
Ask your question as though you were talking to an expert:
Once my data is uploaded, I want to start by looking at the spread of cases across its types. So I type my question in natural language right here on the main page and choose the most relevant insight.
Customer is king – so were they delighted?
I drag and drop Service Satisfaction on top of this visualization to better understand how we have performed across each of these four Case Areas. Minimizing ‘Unsatisfied’ ratings is a business imperative. At this juncture, it’s important to note that the performance of our service agents is measured based on their average call duration and the service satisfaction ratings on the cases they have resolved.
I replace the number of Cases with Case Call Duration and instantly realize that Access and Login case area had more number of cases but takes less call time whereas Hardware case area had less number of cases but takes relatively more time to resolve. This is a good start!
This easy to understand spiral visualization shows me the factors that drive Service Satisfaction and ranks them in order of their predictive strengths. Looks like the combination of ‘Agent Training Level’ and ‘Case Severity’ has the most significant bearing on customer ratings. I can see that the predictive strength of this combination is almost 60%. I can also see that Case Call Duration, Case Area, Case Priority and even Requester Seniority are other important factors. This is great insight and I can look at each of these factor combinations and their associated detailed insights to delve deeper.
With my curiosity piqued, I click on the detailed insights to know more about this top driver combination!
A well trained agent is like a well trained soldier:
Looking across Case Severity (from left to right), especially for the Critical cases, I see that most of the ‘Unsatisfied’ ratings is due to ‘No training’ while most of the ‘Highly satisfied’ ratings is the result of ‘Sufficient Training’.
Despite our recent aggressive hiring of new agents, we need to have a strategy in place and to map the right agents to the right case based on their training levels and the case severity. This will have a direct bearing on bringing ‘Unsatisfied’ ratings down!
I open a new discovery and type my next question to look at the predictive model for Service Satisfaction.
Here I first review the statistically derived decision rules for ‘Highly satisfied’ ratings and then those for the ‘Unsatisfied’ ratings. This further corroborates the pivotal role that Training plays towards our agents’ ability to resolve cases and delight our customers.
Bottom-line, while it’s great to have a handle on what’s happening, I now know the WHY behind what’s happening!
Brevity is the soul of wit:
I go ahead and ask a new question to look at the spread of avg Call Durations across Agent Training Levels and Case Severity and click on the most relevant insight. This visualization shows how the lack of training results in higher average call duration across case severities.
This insight further emphasizes how training can help cut Call Durations and hence minimize cost while expediting successful resolution of cases.
When it all comes together like a charm:
With my quick exploration done, I can bring it all together in a Display!
Great insights but where is the action plan?
In the second tab, while the scatter chart and the heat map further underlines what we discovered, I want to modify the Service Agents by Avg Call Duration tree map (on the right) to arrive at a priority list of those agents who should be immediately enrolled in the first phase of our training roll-out. The numbers in the tree map correspond to agent IDs. This is the action plan I want to derive from my insights!
To get to this priority list, I apply a bunch of filters as follows – Service Satisfaction is ‘unknown’ or ‘unsatisfied’, Agent Training Level is ‘no training’, Case Priority is high, Case Severity either ‘major’ or ‘critical’, Case Type being ‘issue’ and Requester Seniority being either ‘senior’ or ‘management’.
I then distribute this list of agents across Case Areas for better focus during the training program.
So to recap, I started with a simple dataset and an intent to get to the bottom of our dipping customer satisfaction ratings issue. Thanks to the cognitive smarts of Watson Analytics, I was able to quickly analyze the status quo, understand the latent cause of the problem and concluded with an action plan for its immediate resolution.
Check out the narrated video for this usecase – here.