Early next year, Zendesk will be releasing a new tool that's currently in beta that applies machine learning and big data analysis in the — wait for it! — customer-service realm.
I know, I know. CRM vendors and pure play customer service providers do invest in the service segment of the customer experience. Contact center technologies, especially cloud-based applications, for example, routinely see enhancements and upgrades and fundraising.
It could be argued, however, that the best technology, the most cutting edge research, such as machine learning and artificial intelligence has been reserved for the revenue drivers of sales and to a lesser extent, marketing.
It could also be argued that the investments that go into the service space are also geared to better connecting with customers in order to increase sales, even as they address the service issue. As it turns out, according to a new survey commissioned by Verint, customers hate that.
Certainly these are arguments that Jason Maynard, director of Data Analytics and Products at Zendesk, can and does make.
"Customer service in general is underserved in terms of vendors apply the more advanced machine learning techniques," he told CMSWire. "Meanwhile the industry routinely sees marketing tech apply and refine predictive analytics to do lead scoring or recommend a product."
Watching the Meter
The product Zendesk will be making available in early 2016 is called Satisfaction Prediction. Like the name suggests, Zendesk says it can predict in real time, while a customer is on the phone with a customer service rep, or exchanging instant messages with him or her, how likely that customer will be satisfied when the call is over.
This information is not withheld from the rep, as the app shows via a meter how well the call is doing.
Satisfaction Prediction analyzes customer signals (more on that in a moment) via a machine learning model is developed (ditto) to generate a zero to 100 score for the rep each time a customer service ticket is created or updated.
A rep, watching the meter plummet below 50, can escalate a conversation on the spot. A rules engine permits other sorts of interventions -- a service team leader might, for example, be signaled when an ongoing conversation is dropping below 50 allowing him to listen in.
Customer Satisfaction? Not Necessarily
Now, none of this is to say that customers will walk away supremely and uniformly satisfied after a talk with a rep equipped with Satisfaction Prediction.
The app doesn’t offer advice on how to salvage a call, for example. Also, and this is very key: the 28 beta users are only just beginning to use the product. Zendesk could not offer any early use cases or stats to show its efficacy.
And ultimately, of course, the company's policies in place for addressing customer complaints and issues are what will rule the customer's satisfaction. No matter how empathetic a rep is, made even more so by this tool, if a company's rules do not allow a certain action or refund the customer wants then too bad.
That said, Satisfaction Prediction is clearly a vanguard in the use of deeper analytic uses than just predictive analytics.
It developed the product against two training datasets: the surveys customers fill out after a service call and the history of all discussions customers have had with a particular company.
Again comes the caveat: these training datasets seem rather limited, especially the voluntary surveys; certainly they are no match for the massive datasets developed by universities or governments. But Maynard argued that, in many ways, they are better suited for this kind of use.
Customers of Zendesk customers have a higher-than-average response rate to these brief surveys of 28 percent, he said.
Also any Zendesk customer using the product must have 1,000 finished surveys to make the feature live.
The App Identified Barf As a Signal
The combo of machine learning against these two training datasets and the company's customer service records produces a list of signals that point to a deteriorating call from a customer of that particular company, Maynard said.
"Certain terms are particular to an industry and when they are used they could suggest a breakdown or problem of some sort."
And of course there is the usual range of words and phrases that all customers pepper their conversations with when they are unhappy. Examples include "want refund" and "ASAP".
And my personal favorite, the word "barf." Strictly speaking, barf was identified as a signal of an unhappy customer at only one company, a transportation provider. Which immediately got me to daydreaming about what a service rep’s job would be like for that particular company.
Title image by Octavio Fossatti.