Most people today are fairly comfortable interacting with chatbots. We see them pop up in a window in the lower right-hand corner of the web pages we visit, inviting us to a digital conversation so they can help us find what we’re looking for (or sell us something). Often, the popup in the corner is an actual person offering to chat live, but sometimes it’s a robot programmed to talk like a person and answer basic questions.
Over the years the artificial intelligence (AI) software that was able to choose answers to questions based on input data has become increasingly sophisticated. Machine learning—the ability to incorporate the data from each interaction to improve future interactions—has also made exponential leaps forward. While the first “chatterbot” known as Eliza could pick up on a keyword like mother and use it to ask about someone’s family, evolutions in chatbot technology allow them to use natural language processing to understand context and intent as well as keywords. Chatbots have become so good in their ability to mimic humans that new legislation in places like California requires companies that use chatbots to be transparent about when customers are dealing with a bot and not a person.
This proliferation of competent chatbots is good news for companies. Research shows that chatbots save companies roughly four minutes per inquiry interaction, and are expected to save businesses up to $8 billion annually worldwide by 2022. So, while chatbots are adeptly handling the boring, routine questions with speed, human customer service experts can tackle the more complex ones, optimally leading to a better customer experience for everyone.
Help bots are chatbots that do exactly what their name implies — help companies across various business departments. For example, you may have a bot that answers questions for IT or HR departments. In IT, these would typically be common help desk issues, also known as Tier I problems. One IT support company listed the most common IT help desk issues including: I have the blue screen of death, my computer shut down automatically, I can’t print anything, I’ve lost connection to the internet….
In HR, similar problems might be: How do I fill out my timesheet? How do I add someone to my benefit plan? How do I put in for vacation? What do I do about an error on my paycheck?
Spoke’s recent report published in TechRepublic showed the top 10 questions for help desks in 2019 include:
Answering these questions several times a day would be highly redundant for a human, but with a support bot, solutions to these problems can be programmed (with all their language variations). The chatbot can then be trained to choose answers from the company knowledge base, walk employees through a troubleshooting process, or direct them to an article that might help. With this kind of support, the company doesn’t have to pay an expensive employee to answer simple questions that come up all the time.
Most chatbots are trained to ask whether their answer solved the problem. If the customer (employee) replies “no” either by texting, by clicking on a button, or whatever method is provided by the system, the chatbot can escalate the problem to a person.
Without AI, training a chatbot to properly answer questions relevant to your business required collaboration between the people who generally field inquiries and the IT department. For example, the IT team will know how to train and build the chatbot, but may not know the answers to the common HR problems. They’d need someone who is an expert in HR to provide those answers.
One questioner might ask “How can I add someone to my healthcare plan?” while another asks “Can I add a family member on my benefits plan?” They both mean the same thing, and without machine learning IT would need to program all these different versions of the question and enlist the HR expert to recommend answers, knowledge base articles and so forth.
But with machine learning, finding an answer to the question isn’t just about matching programmed questions and answers. After initial training of the chatbot, machine learning collects data from every interaction and uses it to learn how to better answer questions in the future. And the more interactions the machine has with a particular circumstance, question, or set of facts, the more accurately it will be able to identify keywords in questions and provide correct responses.
This means that over time, bots get better and better at customer support. They can learn which responses solve customer problems and which don’t, and start to route customers only to the more successful ones. They can do rapid A/B testing of responses. Some customer support software is able to recognize that the customer is getting frustrated from the words the customer is using in the chat, and automatically escalate to a human agent who specializes in that issue to salvage the relationship before it is damaged.
Machine learning also allows chatbots to collect data that eventually reveals patterns about customers and can help create better customer products and experiences in the future. For example, machine learning might be able to identify that a lot of customers are calling in with a problem about a particular product. It might even be able to use the data it has on those customers to identify commonalities among the customers who are having the problem. Perhaps there’s a compatibility problem with a particular operating system or browser that no one had identified before. Machine learning can collect and aggregate the data that reveals this, so the engineers can solve the problem.
Some research shows bots can handle 80 percent of ticket items. Bots process data and recover answers much faster than a human who might have to put a customer on hold while they locate the answer, or locate the expert who can provide the answer.
Moreover, bots work 24/7, holidays, weekends, even during disasters. They don’t take sick days, or holidays, or ghost their employers. If a customer or employee has a question any time night or day, the bot is always available to provide an answer. In the business of the day, employees might not remember to ask their HR issue. Or they might be working remotely when the IT glitch occurs. With automated support, they can find an answer, no matter what time it is.
atSpoke delivers the chatbot experience through Slack, an internal messaging tool used by thousands of companies across the world. A user opens a Slack direct message thread with the atSpoke Slackbot (at atSpoke, our bot is called Koko) and asks a question they need answered – say, “can I add my wife to my insurance?”
Chances are, someone has asked this or a similar questions before. atSpoke’s AI will attempt to answer the question based on the company’s knowledge base resources. In this case, atSpoke may send a link to the company’s HRIS system, such as Gusto, ADP, or Workday. Responses are curated by machine learning and natural language processing algorithms which have managed to identify keywords (“wife” and/or “insurance”) and match them with the relevant knowledge base resource.
Say this question has not been asked before, or does not have the relevant resource. atSpoke will still recognize that this is an HR request (the word “wife” is probably not included in an IT request) and create an HR ticket. This ticket will be automatically assigned to an HR team member who can manually resolve the request. This resolution will then train the machine learning algorithms so that next time, the problem will be resolved by the AI. And all of this happens within the atSpoke Slackbot!