As Cathy rode to work this morning, she knew it was going to be a strange day. She reminded herself that once you decide to start a business you sign up for adventures, but today would be one for the record books. Today her ‘new hire’ was starting. She had on-boarded dozens of employees over the years but they all had one thing in common: they were alive.
This adventure started a month or so ago, when her Tech Lead came into her office. “Cathy, I am getting swamped, and the new roll out may slip.” As she started to respond, Steve continued “It is not the project, that is going fine, but I am losing time dealing with questions from our current customer base. I’m spending hours on the phone and video chats, and messenger windows answering stupid questions that are covered in our documentation.” Steve went on, “The docs are great, but our younger customers don’t want to read – they want instant interactive help, and it is killing me. You need to get me some help!”
In discussions with the team it looked like the options were limited, hire a full time person to cover the help desk (expensive and where would we put them?) or out-source the help desk to an off site contractor (still costly and how do we assure quality). Then someone jokingly suggested “Too bad we can’t get a smart chat-bot to handle the work!” And that led to today’s on-boarding.
They had worked with the vendor over the last week and were ready for a live test. When Cathy got to the office, she grabbed a cup of tea and met her team in the conference room (except for Steve, he was on a video chat on the help line). Cathy dialed in to the company’s help line.
<Xenocol Industries, this is CaiT. How can I help you today?> If you listened for it you could tell that the voice was synthetic, but it was clear, unaccented, and pleasant. Cathy looked at the team. “Ummm, yeah, I, ummmm,” Cathy was a little tongue tied talking to an Artificial Intelligence, “I wanted to know if your system can prioritize multiple contact emails for our clients.”
<Okay, I should be able to help with that, who am I speaking with?>
“This is Cathy Smith”
<Thanks Cathy, yes, you can have an unlimited number of contact emails per client, and they can be both prioritized and also grouped by functional area.>
The conversation went smoothly through a few more technical questions about the product. Finally, on of the techs said “Holy Cr*p! This thing is sick!”
<I’m sorry I didn’t hear that clearly, could you say it again?>
“HOLY CR*P! Where’s Steve he needs to hear this!”
There was a short pause,
<Perhaps I should transfer you over to one of our senior technicians, please hold.>
A few seconds later,
“Hello, this is Steve, how can I help you Cathy?”
“Steve? Is that you? What happened?”
“Cathy? Whoa! I had just gotten off the call I was on, and my screen lit up, I had a complete transcript of a help call, and this voice said that she needed to transfer an upset customer to a senior tech, and was I available. So I said yes, and then I was talking to you. Was that the AI Business Agent?”
Over the next hour they tested the system: chat windows, voice calls, even video calls. The AIBA had a non-descript avatar that had a great professional demeanor.
Every call was answered quickly, accurately, and only when the questions got too technical would the call be transferred over to Steve. He figured that this would cut his tech support workload by 80%, and since it worked 24/7 it would really improve moral over the weekends. Steve was sold. So was Cathy.
Welcome to the Future
So is this all science fiction? Not really – every piece of this is commercially available technology including the video avatars. Much of it is being put to work today. If not by you, then probably by your competitors. The biggest hangup is “ChatBot Engineering” designing the complex knowledge bases needed to import the knowledge. A recent “build a bot” tutorial based on the IBM Watson chat API required 181KB of structured data (think 60 pages of code and data) to handle questions like “Where did you go to school?” Another problem is language processing. While systems are great at finding the address of a restaurant, anything much deeper and the accuracy drops. A recent study suggested that the top engines from the big players were hitting 60% to 70% accuracy of simple intentions extraction. Like hearing “I want a nearby Mexican restaurant” and missing the fact that they were looking for a restaurant 1/3 of the time.
What is also currently missing is any form of ‘understanding.’ The systems rely on a human carefully encoding the information, the information flow, and the complex interweaving of language and knowledge to make this work. It requires the ChatBot Engineer to design and implement a complete mapping of the knowledge base and technical language into a seamless canvas. And if it is not complete, or if there are inconsistencies the customer experience is (far) less than optimal.
There are a few companies, like GunderFish (shameless plug) that are working on integrated natural language understanding engines that enable the system to actually reason about a help session on the fly, rather than requiring thousands of lines of pre-programmed question / response structures.
Estimates are that in the next 3-4 years these systems will be so common place that they will become the defacto standard for the majority of customer interaction.
So start planning now for your first AI Business Agent ‘new-hire’!
For more information get in touch with us, we live for this stuff!