Search engines: How machine learning sharpens a search

- Knowledge is the key asset of every customer service organization. We use tacit knowledge in our people's heads. And we use explicit knowledge, written down in procedures and knowledge-based articles. Together these let us accomplish our mission. So if knowledge is so important, why is it so hard to find? SearchYourCloud found that workers took up to eight searches to find the right document. McKinsey says workers spend almost 10 hours per week looking for information. And IDC puts the number even higher, saying we spend 30% of the workday trying to find what we need to know. When I visit service and support organizations, the thing I hear the most is, "Search stinks. "How come it can't just work like Google?" There's real business cost to bad search. Agent onboarding takes longer, contact deflection opportunities are missed, people get frustrated, productivity suffers. Search for customer service is in many ways especially hard. Your customer has a question, so they can't search for the answer, because they don't know it. Also, knowledge is often written using insider jargon, so it's hard for your customers to find. If you're searching to find out if you can take time off to take care of a relative's kid, it's going to be hard for you to find the right policy if it's described as the in loco parentis feature of the Family Medical Leave Act. Before machine learning, when your staff or customers looked for information on your site, they were probably using a keyword search engine. Keyword search is pretty dumb, actually. It just looks to see how many times the words or simple variations of them in your search string match the words in each document. The rarer the word is in your documents, the more the match counts. There are refinements and nuances I'm leaving out, but that's the basic idea. What if instead, the search engine paid attention, not only to word matches, but to user behavior after they type their words? If people who type: how do I return my widget, seem to love one particular knowledge-based article, even if other articles contain the words return and widget more frequently, then maybe that's the best one to put on top of the search results. That's exactly what machine learning can do. In fact, your customer-facing staff can be your secret weapons in making your ML search smarter. For every customer interaction record, a case, a ticket, an incident, or a chat, ask your agents to attach the article or articles that resolved the issue or answered the question to that record. Now you've given some serious training data to your machine learning search. You have the customer's question in their own words, plus whatever other information your staff has added. You also have the precise article or other document that your knowledgeable staff member said solved the issue. This is exactly the kind of data that machine learning systems want, and as much as they can get. A favorite client of ours has a whole training and coaching program for her staff about this. She calls it How to Train Your Dragon. With their machine learning search engine being their metaphorical dragon. She emphasizes the importance, not only of linking, but linking accurately. If people link to any old content in order to make some kind of required number or percentage, they'll be feeding the algorithm junk. And we all know the first principle of computers, garbage in, garbage out. And of course, the world is constantly changing. Technologies, product names, even language. When I went to school I learned about latency, but users today prefer the plain English word, lag. Trying to keep dictionaries updated to say, Hey, these different words are really talking about the same thing. It's a never ending task. You'll always be behind the curve. But machine learning learns this on its own just by watching how people use the system. So if your service organization has not yet upgraded its search engine to one of the new generation of ML search engines, consider doing so quickly. There is a financial investment. But of all the recommendations I'll provide in this course, this one is probably the quickest and easiest to implement, and can make an enormous difference in the internal efficiency and customer success.