During the pandemic, organizations had to interact with their customers digitally. Contact centers provided the company’s “human face.”
Without face-to-face interactions, it is a lot harder to understand how your customers feel, since you cannot experience customer behavior directly.
Running a contact center is like steering a submarine: you need a periscope to see what is going on.
What Does A Contact Center Manager Use For A Periscope?Bigstock
Contact center managers have two tools—post-call customer satisfaction (CSAT) surveys and sentiment analysis.
CSAT surveys ask customers to react after their encounters with the company, prompting them to give a numerical score.
Sentiment analysis uses speech analytics to take customers’ “emotional temperature” during the conversation.
I believe that sentiment analysis is a better “periscope” than post-call surveys.
Post-Call CSAT MeasurementBigstock
How It Works
When the interaction ends, the automated survey asks the customer to give a numerical score. This measures how they feel about the interaction. Customers may also be asked to say why they gave this score.
Survey Wording Issues
One popular CSAT measurement is the net promoter score (NPS). Customers are asked how likely they are to recommend the company to their friends and relatives.
NPS’s strongest advocates believe asking how likely customers are to recommend the company is better than asking how happy they feel. It’s not clear how carefully respondents think about the question. They are asked to respond unexpectedly. They rarely have the time or the interest to consider the question carefully. Their response will most likely reflect their emotional state.
NPS’s scoring system may not match up with how customers think. NPS classifies anyone giving a score of 6 out of 10 or below as “detractors,” or people who will complain about the company. Customers giving 9 or 10 out of 10 are classified as promoters, or people who will tell everyone how good the company is. Those giving 7 or 8 are classified as “passive.” Respondents are unlikely to think in such depth. If their problem has been solved, they will give a high score, if it hasn’t, they will give a low score. Some respondents have even given a 7 or 8, because “they never give 10 on principle.”
About 3% of customers respond to post-call surveys. This is too small to be considered a representative sample. Where results show poor CSAT, this may reflect angry customers’ motivation to show their feelings or get “revenge” on the agent. It does not necessarily indicate how all customers feel.
Inconsistent customer reactions and low sample sizes make aggregating CSAT data a frustrating task. Inaccuracies potentially baked into each result are then compounded by the volume of results.
At a high level, ranking agents’ average CSAT or NPS scores can raise some red flags if an agent has a lower score than the team average. The same can apply to team or queue averages.
How It Works
This is a much newer technology than post-call surveys. Speech analytics software can be programmed to identify and indicate whether customers are expressing positive, neutral, or negative feelings.
It is trained to recognize such feelings based on samples where the speaker’s feelings are known. The system uses artificial intelligence (AI) to construct a picture of which combinations of phrases, pitch, pace, and volume match feelings that have been identified in a recording by the AI’s trainer. Where mismatches are discovered, the system can be further trained.
For sampling purposes, the sounds on a voice call can be split into each party on the call and analyzed separately. Sentiments can be identified even when both parties are speaking at once.
This is the major differentiator between sentiment analysis and post-call surveys. Sentiment analysis is usually applied to all calls. It can be applied to all parts of a call, showing users how customers’ feelings change throughout the call. The sample size is likely to equal the population being studied, so the statistical significance of the data cannot be denied.
Sentiment analysis reflects how the customer feels without having to process and respond to a question.
One limitation is that because there is no question, you cannot tell why the customer is angry. The root cause might be the agents’ behavior, the issue with the product, or be unrelated to the call at all.
Since the sample size is so much larger, there is more scope for aggregation and analysis. You need to build a set of benchmarks to establish what is “normal” for your population. If a water-utilities contact center handles issues relating to wastewater disposal, customer sentiment will be fairly negative as a matter of course.
League tables showing average sentiment by agent, team, or queue can quickly identify where improvements can be made. Comparing or correlating this with other data such as call length or first contact resolution, you can see how contact center operations affect customer perception. You can see what makes customers angry or happy, and then tune your offerings as a company accordingly.
Sentiment analysis clearly produces more data than a post-call survey, but it's usually more expensive to collect. Cloud computing is making speech analytics and sentiment analysis more affordable for smaller contact centers.
What do you use as a “periscope” on your contact center? How useful are the results? Do they match your expectations or are they surprising? I’d love to hear more!
Here are links to some other articles on NPS, customer feedback, and customer sentiment: