Job #1: Performance analysis for better customer service

Analytics reports with dashboard

The story begins

This story is the first analytic assignment in the job simulation series powered by Forage and PwC Switzerland. The aim is to foster digital awareness, emotional intelligence, and creativity to fully participate in the digital future workplace. As a business intelligence analyst – a Digital Accelerator, I received the requirement below to study a case about a call center’s performance. Many questions are raised about the team’s performance, yet the most important one is: What do customers really want?

Such a typical request from a business partner! Claire started wondering about her team’s performance and the nature of her customers. She came up with 100 different ideas about what metrics should be measured and envisioned a story related to the trends of her clients and agents. With questions in mind, she reached for a visualized vision and believed in the power of looking at multiple facets while glancing through a dashboard. Let’s see the dataset I was given to work with.

The dataset is reasonably structured, but we can always review and utilize it for more efficient analysis. So, my starting point is to learn about the characteristics and the points that need to be modified:

1. The primary key is “Call Id”, revealing the unique set of values in each row (by each call, not by each customer, by day, or by any means of measure).
2. There are columns related to time that can be used for time analysis: “Date” (in the form of full date) and “Time” (in the form of the timestamp where the call started).
3. There are numeric columns (containing discrete data): Speed of answer (in seconds) and Talk duration (in Time format). My first thought is how they should be transformed into scalar values with the same unit type (in seconds or minutes – decimal value).
4. Satisfaction rating: Should it be considered a categorical variable only, or should it also be seen as a scalar value (like an average satisfaction rating of 3.5 for Becky, perhaps?)
5. So, I have categorical and numeric variables and a prospecting time analysis. But how can I approach it effectively?

Transform data

So why should we need a specific and separate table for date while using Power BI for analyzing a dataset?

First, it ensures continuous date ranges, even if your transactional data has gaps, preventing misinterpretations in trends and aggregations. For example, we can see monthly trends throughout a year while minimizing the impact of 1-2 months omitted from the dataset.

Secondly, it allows you to define standard fiscal years, quarters, and other custom time periods, aligning your analysis with specific business needs. For example, create a new time period – Day of week unit that has not appeared in the data set yet to explore new facets of business performance. In my Date table below, I used Format (with “dddd” ) to transform the date column into the Day of the week. These transforms ease how we can use different filters. Reference link

Thirdly, a date table enables the creation of sophisticated time-intelligence calculations, like year-to-date, month-over-month growth, and moving averages, which are otherwise difficult to implement reliably.

For further specific cases, you can do yourself a favor to have a glance at this topic.

Be careful with loading the Excel dataset while using the Power Query tool from Power BI: in this case, the automatic type detection did not do anything wrong, but the outcome did not match what I expected for a column of (potential) scalar value type.

When working with data that includes multiple columns with different units (e.g., minutes and seconds), it’s essential to consolidate these into a single, consistent unit to ensure clarity and ease of analysis. Therefore, we are going to transform this “timestamp” into meaningful time (in seconds).

1. Duplicate the original “Duration” column to preserve the raw data.
2. Extract the minutes and seconds into separate columns or variables.
3. Convert the extracted minutes and seconds into a single unit (e.g., total seconds) for consistency, using formula: time duration = minute(s) x 60 + second(s)

Data analysis approach: From Descriptive to Diagnostic

Using the fishbone diagram, I followed two types of analysis that matched with finding trends of Agents’ performance and Customers. The required dashboard would be the final one, after analyzing different aspects of the Call center story (through Distribution of several categorical variables, Time analysis, Outliers seeking, Impact analysis, and Comparison analysis).

Starting with Descriptive Analysis, the diagram highlights the importance of understanding the distribution of calls through various lenses such as time units (shifts, daily, monthly), staff involvement, and resolution status (Yes/No). This analysis helps in identifying patterns and trends over different time frames and staffing scenarios, which is crucial for operational adjustments.

Diagnostic Analysis focuses on pinpointing issues by comparing agents’ performance and identifying outliers. In this case, I desire to understand the differences in the pattern of agents’ performance and how they quantitatively compare to each other in several filters. We may not want to know every detail of each agent, however, we prefer impactful details aligned with each agent to deliver actionable solutions.

Descriptive analysis

By cross-filtering through the time unit (Month) and other variables, we can see where the better-than-average call rating was distributed. For example, some insights were highlighted from the dashboard filtered in March with Rating 4&5, including:

1. Approximately half the number of calls are rated 4 and 5
2. Answered is more important than Resolved: For some calls, customers feel better when our agents reach out and put an effort to receive the cases even before any solutions are recommended
3. Among topics, Technical Support seems to be highly appreciated by customers
4. In March, Becky, Stewart, and Jim maintained a good level of satisfaction from the customers, while Greg and Joe went a bit short in high quality calls. We can delve into this performance and see if there is a trend through months among agents.

Through time, especially with different time units, we can figure out the pattern of each agent’s performance and further insights,

Following each agent’s performance, it seems that there is a trend that needs more attention: The more an agent has high rating calls on a day, the more they would abandon more calls on the same day. Let’s check through cases, where the call performance of different agents (Greg, Martha, and Stewart) is filtered by the call topic “Contract”.

Diagnostic analysis

I always have a preference for Waterfall charts, especially if the dataset has different categories and at least a numeric value (which aggregating values such as “average” or “count” make sense) that needs to be analyzed throughout a specific period. For instance, according to the below dashboard, the average call rating decreased from January to February, mainly from the Morning call, in the topic of Admin or Technical support. It seems to be better in March, when some topics, mainly Contract-related or Technical support, and agents such as Becky and Jim, influenced the team’s average rating not to go down much further.

However, not everyone would follow the team’s trend. For instance, let’s see what Stewart brings to the team with his expertise and pattern. After disastrous performance in February, he improved both the number of calls he received and his customers’ satisfaction rating. Yet his expertise shines best when he performs well at the topic of Technical and Admin support in February. Compared with the topic trend above, you can easily see this difference.

And no one is perfect! Becky, a rising star in March, has her weakness in the topic Technical Support. When visualizing by this waterfall chart, we can clearly see a weird situation: Becky lost most of her rating in February by Technical support, and gained her best improvement by the same topic a month later. Why? It seems that she was more focused on quality than quantity of calls in March, or perhaps, if we can somehow delve into the content of her answers and expertise, we would know more about the reason.

To this stage, I found 4 numeric variables that can be used to compare the agents’ performance: The satisfaction rating (which was mostly used as a categorical variable in the descriptive analysis), the speed of answer, the call duration, and the number of calls. Therefore, I used the scatter plots, with the idea that each pair of numeric values would demonstrate the relative position of each agent to the others. If you are a football analytics fan, this kind of chart is usually used to demonstrate the focused aspect of a footballer’s performance throughout a period, or even after a match.

Let’s see if the average satisfaction rating can segment into different types of patterns among agents.

1. We have different success patterns. If we combine the factors in charts above, we will see that Martha is slower to answer calls and spends less time on each call, but her efficiency and diligence are not inferior to Dan’s.
2. Conversely, to improve performance, we have to adjust for each agent accordingly. For example, Becky is very eager to answer and is very diligent in consulting many customers, but she is either not focused, or is not strong in some consulting topics (which was suspected in the breakdown analysis above).
3. Joe, on the other hand, clearly needs stronger supports from the supervisor: he consults customers less than other, is extremely slow to answer to call, and does not tend to spend more time on consulting customer than others.

The final dashboard is a good way to have an overall view of all agents’ performance. However, it would be a journey of discovery through analytics steps and multiple charts of different aspects that reward us with meaningful yet detailed insights to drive the performance to the next level.

Conclusion

So, there are many analyses and stories that can be found from just a few dashboards to help the call center’s manager to drive the performance of her agents. The visualization ensures her multi-faceted approach and tailors her solution to each agent, with the proper level of detail that constructs actionable solutions. Hope this dashboard can facilitate her work and the teams!