It’s interesting to know that the semantics of an individual's language in an organization vastly differs according to his or her function.
Many stakeholders have access to central information systems in an organization. They are usually given customized views of data as per their roles and hierarchies. Many a times they end up quizzing the information system for the same information, but in a different lexicon. In legacy, business intelligence interfaces, there would be different dashboards and reports for different employee roles to ensure that the system makes sense to them.
However, imagine if there was one person (A certain "Mr.KnowItAll") who could have answers to the questions of all these stakeholders. How would they end up communicating with him?
If the Head of Sales wants to know more about his team's sales numbers for yesterday, he will probably quiz him by asking, "What are my sales for yesterday?".
If the CFO wants to know the answer to the same question, he might frame his question as, "What is the company's revenue for yesterday,”? Infact, if the truck driver wants to know what were the number of shipments he carried yesterday, he might ask, "What were my shipments for yesterday"?
Interestingly, as you would have observed, all these stakeholders are quizzing him for the same information. Since Mr.KnowItAll knows the role and hierarchy of the information seeker, he ends up sharing contextually relevant answers with each of them.
However, if Mr.KnowItAll were an artificially intelligent entity, his answers though correct, would typically go through an interesting learning curve to make them contextually relevant to information seekers.
Here's how their learning curve is supported.
Artificially intelligent BI assistants typically use different adaptive learning techniques like supervised learning and reinforcement learning, which help them perform many tasks.
Reinforcement learning is more of a trial and error approach. When the assistant is not able to understand the user’s question, it tries to provide options on the current context to help the dialogue progress faster and more effectively. This process helps the assistant to deliver context to the dialogue with the end user considering the user's need for information.
Supervised learning encompasses teaching AI assistants by feeding them with enough training data and also clarifying the options with SME knowledge so that the assistant can understand and answer the end user next time. By this way of training, more you use the assistant, gets smarter and end up understanding better and answer accurately.
For example, customer service departments of many large companies are using AI assistants to speak to their end customers and answer their queries. The AI assistants are trained by making them scan through thousands of hours of customer-agent call records, which ultimately helps them build context and humanize the conversation.
Conversational platfroms like ConverSight.ai from ThickStat use a combination of such learning methodologies to power up their intuitive interfaces and serve their customers’ information needs, as per their roles and hierarchies. Should you like to learn more, take a look at the recording of their recent webinar here - https://www.youtube.com/watch?v=l64f5vreKCc