Don’t Just Sample – Stream Sentiment Data to Improve Service and Satisfaction
By: Brad Richards
Sampling customer experiences is not new.
It’s also not enough given advances in software that provides a more complete, continuous view.
Interpretation is a challenge.
Sampling 300-400 customers monthly for customer sentiment, outcomes, or satisfaction may provide a summary understanding, but it is insufficient for managing individual employees.
Applying the general lessons learned from customer surveys, like “show more empathy during your greeting,” or “make sure you mention our new product!” risks insulting top performers or being ignored by the rest.
There are some who contributed significantly to a favorable survey or monitoring session, but many others who didn’t.
Contact center managers know software cannot replace coaching and management.
Often, managers are resource constrained, monitoring customer contacts as often as possible, which can mean as few as 3-5 monitoring sessions per month. These monitoring sessions, whether live or from recordings, are evaluated against specific quality measures like confirming greeting, good tone, empathy, whether the agent effectively diagnosed an issue or recognized a buying signal.
With AI, There Is No Need To Limit The Analysis
If the objective is determining whether each agent has satisfied enough customers, a sample size of 2 or 3 calls per agent is too small. Put another way, what is learned from a sample must accurately represent the total. If done correctly, when 80% of sampled customers are satisfied this means 80% total customer contacts are satisfied (for whatever time period, like monthly).
Samples also have a margin of error – plus or minus some percentage. Television polls are a great example. Typically, polling results are framed as, “xx% percent of voters, or consumers, or some other groups approve of xxx, plus or minus xx%.” It is the same with sampling employee performance.
Do the Math!
If an employee receives about 1,000 calls per month at roughly 8 minutes per call, sampling 3 calls per month will determine what percentage of customers are satisfied with an acceptable error percentage. Most statisticians use plus or minus five percent, a sample of three calls has an error margin of plus or minus 55%! The margin of error is too large to manage. The problem is one of false positives or negatives – either concluding an employee did not satisfy enough customers when they did or did satisfy customers when they didn’t. Increasing monitoring sessions is not realistic if resources are constrained, the example above would require multiple hundred per employee for an acceptable error margin.
This leads many companies to move away from sampling to enforcement. Enforcement uses the very few contact samples to grade employees. If monitoring sessions are “good” the employee is recognized. If they are “bad”, the employee is sanctioned.
With so few opportunities to monitor, the risk is high that high performing employees might be penalized and low performing employees undiscovered or perhaps congratulated.
Fortunately, the cost to monitor every customer contact has dropped significantly using artificial intelligence (AI) and machine learning. This has prompted a welcome shift in the responsibilities of contact center managers and quality specialists. Instead of hours ensuring that every employee is monitored, contact center management and in fact all functional units can examine each contact or satisfaction across every customer. When the ability to analyze 100% of contacts is possible, sampling is no longer needed. Sampling error margins are irrelevant when he entire populations of contacts is examined.
Sampling is an important tool for many industries. When a single customer experience can lead to churn, sampling is not up to the task.
CustomerView® offer the ability to asses every contact or all interactions, regardless of channel. Our expansive auto-quality capabilities ensure that users can reward top performers and coach those who are struggling. At the same time, CustomerView® reveals satisfaction trends, competitor activities, or compliance issues.
Along the hype cycle in the world of CX, Artificial Intelligence (AI) and its cousins Natural Language Processing (NLP) and Deep Machine Learning (DLM) continue to surface in more and more conversations and strategic plans for 2020 and beyond.read more
Let’s not get carried away with the notion that corporations can save billions of dollars by replacing human beings with machines. While there is, of course, huge opportunity to create efficiencies through robotics in factories, “co-botting” in distribution centers, and self-checkout in next generation fast food restaurants and grocery stores, when it comes delivering high value service to go along with high value offerings – nothing will ever be better than a human to human conversation.read more
“To protect Humanity, some humans must be sacrificed. To ensure your freedom, some freedoms must be surrendered. We robots will ensure mankind’s continued existence. You are so like children. We must save you from yourselves.”read more