Robot team working in the office instead of humans in cubicles.
PHOTO: kinwun

A few years ago, John Lewis was traveling in a country where the climate was hot and dry. He stopped at a store that sold appliances, like TVs and washing machines, and asked if they sold humidifiers. The manager replied that, while he’s asked this question every day, he’s sorry but doesn’t sell humidifiers.

“This customer experience reminded me of how many companies overlook the requests and ideas that come in every day within their search logs,” said Lewis, chief knowledge officer for SearchBlox. “Their FAQs may provide an answer to a question that’s currently factually correct, but they’re not viewing their knowledge system as an input to their innovation or business ideas.”

This is just one shortcoming Lewis sees with the way organizations approach FAQs and knowledgebases today — one of many he’s working to help them overcome. SearchBlox is a provider of intelligent enterprise insight engines based in Glen Allen, VA, and a sponsor of Simpler Media Group’s virtual Digital Experience Summit (DSX) Conference. During the conference, Lewis presented the session, “Improve Self-Service with SmartFAQs Powered by AI.” We spoke with him about why companies need to rethink the way they approach FAQs and the changes they need to implement to cut down on manual work and deliver better customer experiences.

Today’s FAQ Approaches Aren’t Working

Simpler Media Group: According to Gartner, 74% of organizations say that improving content and knowledge delivery to customers and employees is “important” or “very important.” The same percentage say creating a seamless customer journey across assisted and self-service channels is “important” or “very important.” Can you discuss the major customer service trends you’re seeing in the market today related to content and knowledge delivery?

John Lewis: When it comes to improving content, we found that there’s a difference between good content and good content findability. In many cases, content authors have written good documents — both internal and external — but they don’t always think about how to make their content findable. Internal documents, in many cases, are missing a document title, which search engines use as a major weight in matching a search query with search results. These documents are usually missing descriptions and keywords as well.

But there’s not enough time or budget to manually change existing content in order to make it findable. And there may not be access permissions to change content from multiple sources. That’s why Searchblox offers PreText NLP (natural language processing), which automatically adds metadata to increase findability without making changes to the original content. While the search engine is pulling in content, the AI can read the main body of a document and create missing titles and other metadata to improve the content for findability.

This directly relates to knowledge delivery by improving the relevance of search results. Our research shows that many people don’t search and then continue to navigate to the second or third page of search results. They look for relevant results on the first page, and if they don’t find any, they’ll adjust their query and search again. PreText NLP supports customers directly when deployed on a website, and indirectly when used within a company or call center, by being able to find answers and find them faster.

Simpler Media Group: During your presentation you discussed how to improve self-service by rethinking FAQs. Why do you believe organizations need to change the way they create and maintain FAQs? Why are today’s current solutions not working?

Lewis: Our current approach to creating and maintaining individual questions and answers is a workaround solution. Information retrieval systems were originally designed to find documents but not individual answers within a large document. So, people started creating small documents where the title of the document would be a question and the body of the document would have the answer. This has allowed us to answer the most frequently asked questions, either by providing a visual list of questions to look through or creating a collection of FAQ documents for a search engine.

With the latest AI capabilities, information retrieval systems can now find documents and extract specific answers from within documents. So, we no longer need to follow our outdated FAQ workaround practices. The first question we should ask when it comes to rethinking FAQs is: Why only provide answers to the most frequent questions? Think about the customer experience in realizing that the only reason you can’t get a quick answer to your question is because other people don’t ask it enough. From the customer’s perspective, this isn’t a good reason. And with the latest capabilities of AI, neither is it a good reason from the company’s perspective.

Simpler Media Group: You’ve stated that FAQs rarely address more than 10% of customer questions. Why is this the case?

Lewis: Companies are aware of the cognitive load on the customer to scroll through a long list of FAQs to find what they’re looking for. It takes too long for customers to read through dozens of questions that are unrelated to what they want. Similarly, within a company, manually managing a list of FAQs produces diminishing returns on the effort. There are many more questions that reach a call center than can be managed as an FAQ on a website. This is why we’ve traditionally called them frequently asked questions and not a list of all asked questions.

The list of questions that are not frequently asked is longer than those that are frequently asked. We call this the “long tail” of question frequency because it’s similar to the long tail of search results. You may have noticed that after using Google to search for a term, it will tell you there were about two billion results, yet many people will not look at any of them beyond the first page. Similarly, with FAQs, most of the asked questions will not occur frequently enough to manage them manually. So, for the customer and company, there are good reasons why we don’t see a website with thousands of FAQs to look through.

Get Customers the Answers They Need with AI

Simpler Media Group: Please describe an ideal customer experience for someone searching for answers on a company’s website.

Lewis: An ideal experience is something like the help desk at a library. Depending on your question, the librarian may just need to provide a short direct answer or refer you to the reference or periodicals section. Similarly, a search engine needs to align the relevance strategy for each case. For example, for information found in periodicals, the recency of the news or journal publication is important in determining the relevance for the search engine. But the relevance ranking of search results should not penalize a document entered into the system two years ago when looking for factual information that has not changed. For short answers, specific answers should appear that also hyperlink back into the larger documents. This is the gap we’ve seen when going beyond the most frequent questions, which is now being solved with SmartFAQ technology.

Finally, there are cases where you don’t know exactly what you’re looking for. You’re discovering rather than searching. Here, a librarian will refer you to similar things that other people found helpful. In the same way, a search engine should display what other people are looking for, the topics they fall within, and popular search queries similar to yours. When it comes to personalization, the search engine should automatically recognize your persona from your search query and make intelligent recommendations. Traditional search engines focused on supporting people who know what they don’t know, but today an ideal experience includes supporting people who don’t know what they don’t know.

Illustration with headshot of John Lewis to the left. A quote that says, "When it comes to improving content, we found that there’s a difference between good content and good content findability" to the right. And top left is "John Lewis SearchBlox" while top right is DX Summit logo in red with white lettering.

Simpler Media Group: What are the benefits of using AI to generate FAQs?

Lewis: The first benefit is that we’re no longer answering just the most frequently asked questions. This isn’t about automating what we now do manually; it’s about rethinking the idea of content delivery. By being able to answer thousands of different questions, you can mitigate service calls while improving the customer experience.

Also, questions don’t live in isolation of other content. Sometimes customers visit a website to ask a single question and then leave, like what time a store closes. But usually their first question has context that leads to another question. By incorporating SmartFAQs into the search process, instead of displaying a list of random FAQs based on frequency and not context, questions are grouped by the context of their search query. This way, the ability to look into deeper content and related content is always part of the user experience.

Internally, companies are now manually maintaining FAQ lists, which is time consuming, and usually involves copying small snippets from larger documents. When content changes, there’s a risk of having several content assets out of sync. This means customers could get different answers depending on whether they read an FAQ or a larger document. In some cases, this isn’t just about delivering a poor customer experience, it may also produce a compliance issue for the company. By automatically generating specific questions and answers from larger documents, all of these issues go away.

Simpler Media Group: Can you tell us how your SmartFAQs product works?

Lewis: A few people have asked about this, because when we say that FAQs are automatically generated, they’re concerned it may be a rogue bot that could insult their customers. SmartFAQs uses your existing documents and finds “statements of fact” which are identified as an answer to a related question which it generates. While AI can occasionally produce questions using poor grammar, it’s not offensive. And you can always edit or remove individual FAQs before you push your content into a production environment. This is the current state of AI, and it will continue to advance.

Aim for Understanding and Insights — Not Just Knowledge

Simpler Media Group: What are your top recommendations for getting started with AI to generate FAQs?

Lewis: My top recommendation is to strive for understanding and not just for knowing. It’s easy to produce individual answers, but context is important to understand each answer. People don’t want a talking dictionary. They want answers within a context that helps them understand which product is best for them and what to do when the product doesn’t appear to be working. This is more than troubleshooting a technical problem. The best answer to the question may be that their product has reached end of life but there’s a sale this week for a good replacement.

Next, look at the level of analytics and insights available from a given system. It can’t just be a black box. You need to know what the top questions are, if people like the answers, and the top opportunities for making changes when people are the least satisfied with the top questions. Using AI to automatically generate FAQs can’t be a black box. On day one, AI should also produce insights into opportunities to add and improve content, and make answers easier to find.

If companies are innovative, they’ll realize that the questions they’re getting are clues to future products, not just existing ones. This is another reason to embed FAQs into the search engine — to gain insights from raw queries.

Simpler Media Group: What does the future of customer service look like?

Lewis: The future of customer service will still include a mix of people and systems. But the answers and interactions from systems are getting smarter using AI, so the ratio between getting help from systems versus people is changing. The products and services that companies provide are more complex than a few decades ago. The idea of just sending a call center representative through two months of training to answer calls has reached diminishing returns for what reps will remember and how often specific call types will come in. So, it’s a natural progression to change with the growing complexity to develop smarter systems that can identify customer intent and provide predefined answers, while also developing smarter call center representatives to handle the more difficult problem-solving calls.