Human Call Centers Dead In Three Years

Machine learning will make human answered customer service call centers obsolete in three years or less — especially for English. Other human spoken languages might take a little less or longer but their days are also numbered.

Zeros and ones are taking over.

Initially the cost to run machine learning customer service depots will be too prohibitive compared to third world wages. However, the key to making machine learning cheaper will be to re-teach call center agents to program machine learning. Training algorithms to take 500 to 1000 calls per day will be the common benchmark to compare to their human counterparts.

Today, 200 calls inbound per day is nearly impossible unless each call averages 2.25 minutes and an agent takes exactly 30 minutes of break and 30 minutes of lunch in a typical 8.5 hour work day.

As countries such as India and Philippines make a bulk of their revenues from BPO (Business Process Outsourcing) regarding English speaking customer service and call centre duties — their economies need to retool to prepare for a machine learning future. Or else face an economic meltdown.

Companies such as HSBC who have nearly 10,000 employees that focus on customer service in three languages — English, Cantonese, and Mandarin placed in Guangzhou have an interesting journey over the next couple of years. The question is: what degree of automation, machine learning, and human connection should they use to create the perfect customer service platform.

Other sectors that will take a hit including headhunting and data science. Data scientists are currently working the the hardest to make themselves obsolete.

In a decade from now, machine learning will change everything around us. We will be reading books written by artificial intelligent programs, the new 900 numbers will be hotlines manned by machine learning bots, and we will be listening to non-human podcasting.

Unbelievable? We are already appreciating artwork created by computers’ dreamswhich are mashups of ideas gleaned from parsing the internet and existing artists work.

The biggest trick to transition to machine learning is training programs to be more human than human. Building in banter, jokes, empathy and understanding spoken emotional inflection. Maybe even doing A/B Testing to figure out what voice do certain customer personalities respond to.

Now, when you call customer service, you are usually at a place of vulnerability and desperation that when you get someone on the phone that you like, that you “feel” wants to help you, you ask them for their extension. What if in the future, you can do the same and ask for extensions that are pre-programmed algorithms that are tailored to your disposition, your vocal cadence, and your comfortability of knowing your past.

These programs will be trained to ask questions that if you answer and volunteer information it will know that these topics are not taboo but can be used as references — to build affinity between human and non-human — as long as the same voice linked to a name is remembered the human caller.

Incorrect ways to use machine learning is for a program over the phone to tell you everything they know about you in a single call — as they query vast data lakes of information about your past, present, and forecasts of your future behavior. These algorithms will have to use “key words” or “inside jokes” as gatekeepers to how much information they can they reveal in a normal conversation in order to not seem “creepy”.

Programming punch lines will also be the key to winning over humans in accepting machine learning. However, it has to feel organic. Microsoft is busy at work creating algorithms to do comedy captions and later use this same code for computational humor- telling jokes within interactive human scenarios.

When I co-produced a skit comedy show in Hong Kong called “Comedy Dim Sum” back in 2010, we produced a skit that introduced Kindle the Comic. The comedy is in how robotic the “comedian” spoke and delivered his (the voice’s) lines. People knew not to take him (the voice) seriously. In the future, this non-organic way to communicate comedy will not be tolerated.

All details will be re-created including background ambient noise of other conversations or having to talk to a manager (which will be alerting a human monitor of a situation and possibly ask for an “override”).

Also infrastructure will again need to be replaced, as CPUs that can do heavy lifting around machine learning will replace the common desktop PCs that the call center agents are using — such as Tesla servers (not to be confused by Elon Musk’s version). This means “living” machine learning servers will begin to be phased in and “dead” servers will be phased out. What infrastructure provider will rise as the new manufacturer of Sentient Servers as opposed to simply Unified Messaging boxes?

But the biggest challenge over the next half a decade for those countries who make a huge chunk of their GDP from Business Process Outsourcing is how quickly they can transition common phone courtesy agents who read written scripts to machine learning programmers?

Google is offering free online machine learning courses. As we are in the infancy of machine learning, the concepts are a little difficult to teach to non-computer science or statistical students. But the education companies that are able to simplify these topics into easy to understand modules for the common person will be the next big trend.

Over the next eighteen to twenty-four months, machine learning schools and institutes will be begin to pop up world wide — and call centers will start taking note. Retraining call center agents to program “personality” and monitor calls with the option for “human override” if a call goes poorly. That means one call center agent will be tasked with monitoring a dozen to twenty calls or more simultaneously.

Programming machines to be “likeable” and asked for by name or extension will all the rage. Think of call center agents treating their customized algorithms like their own personal Furby’s. Talking to, teaching, and making sure the programs are as human as possible in order to make themselves successful.

Once machine learning call centers are commonplace, having machine learning algorithms that capture the highest KPIs — lowest customer churn, highest retention, and sales growth in cross selling will be the ones that will dominate.

So by 2019 or 2020, when you ask for that call center agent by name or override the automation to put you in direct contact with your favorite agent via their extension because they were friendly, were always patient, or they asked how your mother was after her surgery because you talked about it during your last call with them, and suddenly you ended up upsizing your mobile phone plan after they resolved your issue, remember its all numbered:

Zeros and ones.