By Dan Somers
With the amplification of social media, as well as the ease and increase in the ability for customers to complain, issues can quickly turn into operational and PR crises. Yet this is just the beginning; issues happen every day that cause customers to interact with contact centers. The intended customer experience can be impacted by taking up unallocated resources to deal with day-to-day issues.
Here we look at three ways machine learning can be applied within a contact center to unlock key data that can ensure the intended customer experience is achieved.
1. Sentiment Analysis 2.0
Typically, the richest and most actionable feedback generally has a negative sentiment. However, this can be buried within feedback that traditional sentiment analysis identifies as positive overall. As a result, key customer feedback that could drive positive business change is being missed.
A review such as “the food was brilliant, and I loved the atmosphere, but the service was terribly slow!” would have only one sentiment considered. Therefore, the actionable insight, “slow service,” is ignored, leading the business to miss out on a change that could potentially have turned this satisfied customer into a huge promoter.
The accuracy of items sold as sentiment analysis that are billed as 90 percent are sometimes as low as 40 percent accurate. The latest machine learning can identify multiple sentiments within text, so no valuable feedback is missed. What’s more, it can do so in near real time in an automated fashion.
2. Concepts, Not Keywords
Until now AI has not been advanced enough to deal with the subtleties of how different people voice different issues and how to make sure you’re not missing key insight as a result.
Existing analytics typically identify keywords within customer feedback. Not only does this fail to consider the myriad of ways different customers may describe different issues, but the overarching concept or message might be missed. This issue arises when a concept or feedback is implied instead of using explicit keywords. We need to understand what is driving that keyword or sentiment and not merely act on the word itself. This driver can get ignored without machine learning.
For example, a hotel chain may pick up keywords such as clean, dirty, noisy, but the driver behind these keywords might be unconnected to the issue itself. The reference to noise might be external to the hotel, or dirty could refer to a specific area of the hotel that could be easily resolved if the full picture was known.
For example, a restaurant customer stating, “By the time my meal finally arrived, the food was cold,” may be flagged as “cold food,” when in fact the driver was “slow service.” Therefore, the appropriate action is to increase speed of service. Machine learning can provide the missing link between multiple words and patterns, giving a much clearer picture of the full concept behind a piece of customer feedback, not just keywords in a silo.
3. Early Warning and Root Cause
Call center crises resulting in high customer churn or dissatisfaction can often be prevented with enough early warning. Unfortunately, current tools cannot identify negative sentiment patterns in text feedback early enough or accurately enough to allow preventative measures to be put in place.
By way of an example, digital communications company O2 had a specific issue in May 2018 with their Priority Offers promotional activity. The allocation of tickets for a popular music event at the O2 Arena for its customers was reduced, and this caused a huge influx of enquiries to their contact center.
Interestingly, this correlated precisely with an increase in customers complaining that they “couldn’t get a response” from customer service as well as “took too long” and “poor customer service.”
Furthermore, there were three categories of churn identified from the public data:
- Customers saying they were going to leave the provider
- Customers saying they could not make a purchase because of an issue
- Customers who made a public recommendation not to use this provider
There is a clear correlation on these three items between all forms of churn and the issues noted above.
Keyword and sentiment analysis had been applied and was not able to discern any of these insights. All it could do was pick up known keywords and generate a sentiment score. It would require an analyst to discern why there were increases or decreases in satisfaction, and this would not be an effective early-warning system.
By adopting machine learning, the company could discern in real time the topics people were talking about instead of just keywords. Fixes could be applied immediately, thus preventing more lost bookings, and they could divert more customer care representatives to the call center to deal with the increase in calls, thus lowering response times. The crisis was entirely preventable.
Using machine learning to implement an early warning system can tangibly reduce customer churn, increase customer lifetime value, and improve customer satisfaction.
Dan Somers is the CEO at Warwick Analytics, providing a machine learning platform for text and voice of customer data.