~10 minute read

Most of us have had an awkward chatbot experience and received the classic response “I’m sorry, I don’t understand what you’re saying.” Regardless of the type of technology that’s behind conversational systems like chatbots and voice assistants, the input for any kind of digital conversational experience is always human language.

And as you’re probably aware, human communication can be extremely complicated. Even when you’re having an everyday conversation with someone you might know very well, you’re likely to experience miscommunications, missed social cues, and general misunderstandings.

We’ve all been there:

“I thought you said the remote was on the coffee table!?”

“No, I said the remote is sitting beside the black cable!”

Sometimes these exchanges can lead to feelings of frustration, but usually you can navigate this friction through further conversation without even really thinking about it.

When you’re interacting with another human, you benefit from the context surrounding each conversation you have: your relationship to the person, where the conversation takes place, why you’re having it, whether you’re talking about the past, the present, or the future, whether you’re trying to get something, explain a concept, or express your feelings.

Technology is advancing, fast, and here at Chata we’ve developed next-generation conversational AI solutions for database access. Through our experiences building this technology, we’ve learned that the key to providing exceptional conversational AI experiences is to adapt the systems themselves to the intricacies and nuances of human communication. The system needs to evolve to accommodate humans, humans shouldn’t have to adjust their natural behaviours to the system.

Humans have all kinds of ways to pick up the slack when meaning is missed or context is lacking. Conversational AI designed for database access? Not quite as much. The only context the AI system has is the data that’s in the database it is built for, and the patterns of human communication that it has been trained to recognize.

As businesses move into the future, more and more people – regardless of what department or industry they are in – will engage with conversational AI systems to get their work done. In fact, Gartner reports that by 2022, 70% of white-collar workers will interact with conversational platforms on a daily basis.

Read More: 7 Reasons Data-Driven Businesses Need Conversational AI

 

Improvements in technologies of today represent the next step in human-computer interactions where conversation itself will become the primary mode of interfacing in business and beyond.

Conversational AI experiences are designed to be just as intuitive and natural as having a chat with another human. Therefore, the rules that apply to having a conversation with a computer are (mostly) the same as those that apply to talking to other people.

 

Here are our top tips for having a great experience with a conversational AI solution built for database access:

 

#1 Know “who” you’re talking to

You wouldn’t ask a barista for details about last week’s sales meeting, right? And that same barista wouldn’t ask you, their customer, to add an additional flat of almond milk to next week’s distribution order.

Conversational AI is built for particular outputs; it’s designed to provide answers about a specific topic or domain.

For example, at Chata, we build conversational AI solutions for database access, which means that the questions our technology can help answer always pertain to the data that’s contained in a given database.

Just like the sales numbers of your internal company are outside your barista’s sphere of knowledge, there are business details that may be common knowledge within a company but are not actually present as data in the database, which therefore excludes those details from the conversational AI’s sphere of knowledge.

Having a clear understanding of what the system is designed to do, or “knowing your audience”, means you’ll have a much better idea of the types of answers a conversational AI system can provide to you.

 

#2 Have a clear goal and expect a specific outcome

When you order from a server at a restaurant, your end goal is to eat a meal. There’s a lot that goes into the whole experience, but ultimately, your expectation is to be served a plate of food.

When beginning an interaction with a purpose-built conversational system, it helps to start with a clear goal in mind. This ensures a higher level of precision in your initial question or request, which leads to better understanding on behalf of the AI system, and, in turn, an accurate and efficient answer.

But a great conversational experience should also include wiggle room to allow for recommendations and clarification (if you know you feel like “pasta” the server should be able to point to “spaghetti, ravioli, or linguine” as potential options).

Knowing what outcome you’re expecting from the interaction will also help you recognize instances where the conversational AI system returns an output you weren’t anticipating or looking for. In situations where you don’t get the output you were expecting, you can easily check your own work to see if you communicated your question clearly and effectively.

Read more: Demystifying Conversational Data Exploration: How to Ask Effective Questions

 

#3 Know your unique data

When you’re interacting conversationally with your database, it’s helpful to have some knowledge of the unique words and terms contained in your database.

It’s kind of like knowing your friend is allergic to peanuts and not bringing PB & J sandwiches to the picnic. “Allergic to peanuts” is a piece of unique data, specific to that friend.

In your database, you might have a vendor called “John’s Supplies”, but you might also have a customer named “John Doe” and a warehouse employee named “John Smith”. In this case, vendors, customers, and employees are all unique identifiers (column or table labels) contained in your database. So if you ask the system for: “All invoices from John this month” you might need to clarify or move through an extra step to verify which John you’re actually talking about.

However, truly powerful conversational AI systems are equipped with robust language models that can handle this kind of ambiguity by recognizing that, based on the context of your question, you most likely mean John’s Supplies and not John Smith in this example.

Find out more about how AutoQL leverages conversational AI to make it easier than ever to interact with your data →

 

#4 KISS is better

Complex answers don’t always require complex questions. “Keeping it stupid simple” is a great way to ensure that the questions you ask while exploring your data with conversational AI systems are straightforward and to-the-point.

Even if you have a lot of different dimensions or filters you’d like to apply to the data, make sure that each question you ask is about a single topic, or that the relationship between the topics you’re querying is crystal clear.

A good rule of thumb is to think about it this way: if the statement would be difficult for your coworker or friend to understand, it’s going to be difficult for the AI to understand too. For example, if you approached your coworker and just said “Summary this month” they would probably respond with “Summary about what this month?” You’d have to clarify that you want to know details about all revenue gained and expenses paid in the past month.

When it comes to keeping it simple, remember that ambiguity is the enemy. Clearly ask the system for what you’re looking for and you’ll be more likely to get the results you need, faster.

Read more: Demystifying Conversational Data Exploration: How to Create Queries 

 

#5 Embrace an exploratory approach

Just like a conversation with another human, it can sometimes be difficult to get all the information you need by asking a single question. But when it comes to communicating about data with conversational AI, the clue really is in the name: these systems are fundamentally designed to support conversation, a back and forth exchange between the agents involved.

You’d never ask your significant other “How was your day at work? How was your dinner with that friend you told me you don’t like very much? How is your mom’s new boyfriend?” all in the same breath and expect one succinct answer. Instead, you would have a conversation where you would ask about each of these things separately or perhaps even start with one of those questions and see the conversation evolve in a different direction entirely.

Likewise, exploring data with conversational AI works best when you ask one specific question at a time, take in the information provided, and ask follow-up questions to dig deeper into the details and gain more information. Let the insights you derive from each Q&A interaction lead you to ask another question so you can get a more in-depth understanding of the whole picture and find the clarity you need to make better business decisions and take strategic action.