Artificial intelligence in customer service & contact centers
Yuriy Mikitchenko, MBA | 1. November 2018
Managing a contact center is a difficult feat. If you haven’t been in a contact center or call center, picture this to put it into perspective: There’s an office filled with hundreds of agents and managers, fielding tens of thousands of calls, emails and chats from millions of customers. Every day.
Taking it further, the agents are typically new to the professional job market, have little to no professional training, and their average tenure is fourteen months. The agents are under pressure to solve customer issues or requests quickly, while handling commonly angry or upset customers.
To add to the challenge, customer service departments strive to give customers the utmost experience possible.
That’s only some of what we learn from our interactions with contact center leaders at the 9th annual Optimizing Contact Centers summit in Barcelona. We also explored more ways to apply artificial intelligence models in their business units to improve the customer experience while lessening the workload on contact center agents.
While there are several applications of machine learning models in customer service departments, we’ve reinforced the obvious conclusion that customer experience improvements are amplified when multiple models start working together.
Simplify the IVR
Customers will always call – a key takeaway from the OCC summit. The leader of the contact center for a prominent Northern European bank noted that they did not experience an immediate decline in customer calls after they introduced live chat to their website (although an AI bot would automate most chats.) Hence, the volume of customer contacts increased after they deployed online chat, while the number of customer service agents remained the same (more on chat further.)
So, let’s start with the IVR (interactive voice response.) Well, as a customer, I don’t like too many options in when I call customer service, and customer service leaders understand that, so they keep them simple – three, but up to about ten, options. The goal is to get the customer to the correct agent for the issue the customer has as fast as possible.
However, based on the work MindTitan has done with customer service centers, nearly 50% of the time a customer spends on the phone is with the IVR and being transferred between agents to get to the person that can handle the issue (or is placed on hold so the agent can ask the right person how to solve a particular problem.)
Why not get rid of all options and just ask the customer why they’re calling?
AI-assisted IVR simply asks the customer, “What can I help you with?”
As the customer describes the issue, the machine learning model listens and transcribes the call. Since the model is taught using previously recorded calls, it knows which words and phrases are associated with specific customer issues – or reasons they’re calling. Based on the classification, the call is routed to the right agent the first time.
And the model only gets better as it categorizes call event data, which is checked by a human-in-the-loop process.
Lastly, the AI model can provide customer sentiment analytics and help contact center line managers with coaching, through representative performance analytics.
Automate 100% of the chat process
Chat gives customers another option to contact customer service. As discussed at the OCC summit, customers tend to call customer service and use online chat at the same time, ultimately using the channel which gets them a response first (admit it, you’ve done that too.)
Why? Because customers want results now. It’s the need for instant gratification that gets customers to hop on online chat if there’s a queue on the phone – I don’t have to explain how this creates inefficiencies across the contact center.
To reduce the workload on customer service agents and provide instant gratification, chat and email inquiries should be handled by intelligent bots first.
Through our work with contact centers, we’ve found that 100% of the chat & email process can be automated, with 60% inquiries via chat and email handled by a bot. The remainder of chats are intelligently routed to the best person to handle the inquiry the first time.
Chat & Email bot
These bot-handled inquiries tend to have simple answers, like understanding a mobile phone bill or ordering a new debit card – the questions and answers are repetitive processes.
The intelligent bot also understands to whom a message should be forwarded to if it decides it cannot satisfy the customer’s question, like an IVR, and asks follow-up questions to better understand the customer’s situation. Thus, complex questions that must be magnified or sales situations get to human agents.
The AI models for customer service are dependent on the data inputs, so the customer service-related models are amplified and get a lot better as they get more inputs. Calls transcribed to text can help chatbots and vice versa. The outputs and outcomes from a next-best-offer model help all other models better understand if initial customer calls or chats need to be sent to sales to help with the problem. While a single machine learning model improves the customer experience, combining multiple models across customer service and bringing in data from outside the contact center brings true efficiencies while vastly improving the customer experience.
A transplant from Portland, Ore., USA, Yuriy brings over 8 years of experience working with advanced technology businesses. Through this experience, he has worked with CIOs and CTOs of mid-market businesses and corporations with 5,000+ employees in the United States, helping them develop and deploy strategic technology solutions. An entrepreneur at heart, Yuriy also spent time advising small businesses in Oregon, while also successfully launching two of his own startups.
Yuriy acquired an MBA from Oregon State University in 2015, where he is honored as “Best in Business” from the Beta Gamma Sigma honor society. He also holds a Bachelor’s degree in Finance from the university.