In my daily conversations and reading, I’m struck by a paradox that enterprise leaders are facing: companies have made huge investments in creating information and documentation of all types to support employees, customers and other stakeholders. Yet the consumers of that information are having a harder time than ever getting the precise answers they need.
Nowhere is this more prevalent right now — and more costly — than in the areas of employee and customer support.
As modern consumers, we have come to rely on our ability to self-service answers to daily questions about everything from video game tips to configuring a major software installation that we’re responsible for on behalf of our employer. On the IT support front, organizations have more than 300 SaaS applications at any given time to provide support for to large employee populations: managing user access, updates, security and more. IT help desks are overloaded.
In both cases, automation has become a go-to strategy. Let’s face it: we prefer to interact with chatbots and expect instantaneous responses. And when we become part of the 40-plus percent who don’t get a satisfactory answer that way, we redirect to a human agent, from whom we expect a more accurate answer — at the same instantaneous speed.
The investments that organizations are making in employee and customer support are simply not keeping up with demands for speed and accuracy — and that doesn’t mean service agents are incompetent, contact centers aren’t well managed, or chatbots as a whole aren’t functional.
Like most organizational problem solving, it just means we need to ask different questions.
The field of AI requires both its innovators and its consumers to be limitless in their thinking — and that applies in spades to both internal and external customer support. Even companies who are already considered world-class in service delivery will not take their position for granted, and will always look for ways to improve — that’s what makes them leaders in their field.
The field of AI requires both its innovators and its consumers to be limitless in their thinking — and that applies in spades to both internal and external customer support. Even companies who are already considered world-class in service delivery will not take their position for granted, and will always look for ways to improve — that’s what makes them leaders in their field.
As humans, the more we learn, the easier it is for us to answer questions; but for machines, the more information they have, the harder it is for them to deliver the right answer.
So whether organizations are building, modernizing, or fine-tuning their contact centers or help desks, there are a few key questions that can lead them towards greater optimization using AI, NLP and knowledge management strategies:
1. How can our chatbot work even harder for us?
Zendesk research reported that, while companies considered high performers in customer service are almost three times more likely to use AI-powered chatbots, half of those companies agree that chatbot performance has been disappointing. Chatbots can automate routine queries and free up contact center agents to answer more complex questions, but they have an inherently limited capacity for the number, variety, and complexity of questions they can answer.
This is where the ability to capture, transform, and channel all of those internal knowledge base resources makes so much sense – both experientially and financially. At Pryon, we’re applying unique ingestion techniques to easily fuse together a company’s disparate content sources into a single body of knowledge that chatbots can use to answer more of your internal and external customers’ questions, more accurately. What’s more, those answers are delivered instantaneously from either written or voice queries.
The levels of accuracy that we’re seeing are astonishing, even for someone like me who has been in the AI field for decades. Pryon provides results that are accurate 90 percent or more of the time. As a result, you may be able to eliminate 20 to 30 percent of contact center calls – or even more.
2. Are our contact center and help desk agents consistently armed with the information they need to help customers?
There are two challenges to overcome here: turnover and effectiveness. Contact centers struggle with annual attrition rates of 30% or higher. B2B companies often train new agents for months on portfolios of solutions, so the loss of experienced staff is especially costly.
The challenge of effectiveness is not a reflection of the quality or skill of the agent, but more about the volume and recency of the information that they are provided, given constantly updated product features and the shifting sands of corporate references such as policies, documentation, and best practices.
Experienced agents can typically handle four to six chats at once but may have to answer up to 12 to 13 questions to help customers successfully. And according to a Zendesk study, 72% of agents say they are not effective at finding the information needed to respond to customers.
Clearly, we are not maximizing the investments in our own knowledge bases. By easily connecting to a company’s systems of record (such as Atlassian Confluence, Microsoft SharePoint, ServiceNow, Zendesk, and others), automating the updating of that information, giving subject matter experts the ability to verify answers and actually improve the quality of the information over time, and providing natural language interaction for end users, companies can accelerate onboarding, reduce training, and ensure that agents always have the most up-to-date information for customers. This also improves the agent experience and reduces attrition and hiring costs.
3. What technology will make the biggest impact on the next phase of my contact center modernization?
Neural network algorithms form the DNA of the next generation of intelligent virtual assistants and chatbots and their importance cannot be overstated. The way most natural language processing is done, data scientists have to provide large numbers of paraphrases for every question they anticipate, and map those to the correct answer. That’s a heavy technical burden.
Because deep neural networks work at the semantic level, they can recognize context and automatically interpret natural variations of words and use that semantic representation to find the best answer in your existing documents. This dramatically expands the volume of questions that the system can recognize and respond to.
Experienced agents can typically handle four to six chats at once but may have to answer up to 12 to 13 questions to help customers successfully. And according to a Zendesk study, 72% of agents say they are not effective at finding the information needed to respond to customers. Clearly, we are not maximizing the investments in our own knowledge bases.
4. How much time and effort are we spending manually programming our chatbots?
For many companies, chatbots have provided the way to scale customer support and keep human agents focused on the more complex and costly queries. However, when programming a chatbot, data scientists must do specific question-and-answer modeling within the chatbot framework (which, by the way, may end up duplicating content that already exists in your knowledge bases). Every combination of entities and intents must be anticipated and articulated manually; every question and answer— and every variation therein—must be fed into the chatbot, which inherently means the chatbot is restricted to the imagination and knowledge of its human administrator.
When we were building the Pryon platform, I was insistent on developing the ability to keep an organization’s content in its original system of record. It’s more efficient, maintains continuity of process internally, and eliminates the burden of any manual, repetitive data entry—so the system can get up and running quickly and focus on adding business value immediately.
We can connect directly to existing content sources and work with all of the most popular document types (CSV, dictionaries, HTML, PDF, PowerPoint, text, Word files, and more). These integrations, combined with our engines and models, means our system comes pre-trained and ready to help chatbots provide a wider variety of accurate answers, automatically.
5. Can we handle an AI implementation if we don’t have a cadre of data scientists at our disposal?
AI projects take months to implement or even longer if your firm builds its own solutions and even the largest and most sophisticated organizations out there are having trouble finding and keeping data scientist talent.
As part of our vision to change the way the world interacts with knowledge, I led the Pryon team to build a no-code solution to help any company solve knowledge management challenges at the first mile (connecting to, ingesting, and cleansing the content) and the last mile (providing natural language access and delivering exact results instead of exhausting lists of links with potential answers). Compare the potential of same day deployments with our platform versus legacy experiences that force months of effort to deliver a third of the accuracy.
Our platform can be configured briskly, without requiring deep data science talent, and inside of a week your end users can be asking questions of any content. That’s the way AI can, and should, work to give the greatest number of people the greatest level of advantage.
