An evidence based practical guide to designing strong conversational experiences for SMBs
Designing for conversational interactions presents new design challenges and opportunities that designers must be prepared to tackle. In this constantly evolving field, it can be hard to reach the scale necessary to unearth significant trends and best practices.
In 2022 XAPP AI launched Optimal Conversations™ for SMBs to bring enterprise-level Conversational AI to a broader audience. The product was designed for small and medium businesses and offered intelligent chat solutions targeting specific industry verticals such as roofing, home remodeling, legal, plumbing, etc. To give you a sense of our scale, at one point this year we launched 500 conversational AI experiences for clients in just 60 days.
When new clients sign up, the technology crawls all of their websites to build a Unified Knowledge Model™ that returns highly relevant and accurate answers to specific questions. Our low-code and no-code options make it easy for users to easily add custom welcome messages and rich message options to help qualify and route drive users to preferred actions such as free estimate, free quote, free consultation, or contact us.
After months of collecting data from the thousands of conversations had between small and medium-sized businesses and their prospects and customers, we’ve identified six best practices about human-computer interaction (HCI) design for chatbots and intelligent virtual assistants:
Provide guidance upfront
Everyone has experienced attempting to create a password for an account and then getting an error because it didn’t fit the required format. Knowing the required format before the first attempt avoids errors and frustration. This same idea translates to Conversational AI experiences.
Things like addresses are notoriously tricky to capture outside of a structured form. Users can include street no., street name with or without street type, city, state, sometimes with or without zip code.
Guidance also includes educating the user on what and how they can ask questions. Our auto-complete functionality helps the user understand they can use natural language to make inquiries and not focus on single keywords.
Whether providing clarity around how to engage with the bot or the expected format for an address, both use cases should be addressed upfront before the user “makes a mistake” gets frustrated, and abandons the flow.
Chips are immensely useful
Chips, a type of rich message, offer up a list of responses that the user can tap on rather than type from scratch. It serves as a fallback when a user typed some utterances that the bot doesn’t understand and helps users feel confident that the bot should understand the chips 100% of the time.
Parse critical information from complex utterances
It’s easy to expect users to just say simple words like, “I want a new roof” but in reality, users are real people in the real world with real problems. Their requests are a direct reflection of their complex problems. We often receive complex, multi-intent utterances that need to be parsed in order to provide a proper response. If you’re allowing free-form text entry you need to be able to parse complex utterances. For example, a message to a roofing company might look something like, “You recently provided me with two quotes. My name is Jerry Smith, 2115 Center Highway in Norfolk, Virginia. My insurance company was out today and did an inspection and I should have their answer by the middle of the coming week. Either way, I am interested in replacing the whole roof. So I provided them with your quotes and will get back with you once I get their reply. My cell is 410-555-8675.” There are three pieces of critical information that need to be collected and passed onto the client company, including the customer’s name, address, and phone number.
Don’t rush the process
Our experience does not ask for users’ phone number, name, and address right off the bat. Most digital-first consumers are looking to learn first before committing to giving up a phone number or other personal information. Understanding intent and only asking for contact info when the consumer is ready is key to building trust. Instead, we answered any questions that the customer has about the business and the product they offer before we start asking lead generation questions.
Give users agency around data collection
Some users might not be comfortable giving their address. This information can be obtained after the initial phone contact, so if a user declines to provide it, rather than pushing it as a requirement a response such as, “That’s ok. We got everything we need to contact you. Someone will be in touch soon.” is recommended. Any contact method should be subject to this methodology, including phone number, email, zip code, and address. For most use cases, as long as one of these is obtained so there’s a way to get in contact with the customer, other information is simply a backup. Remember that this experience does not exist in a vacuum and use other touchpoints as needed.
Acknowledgment builds trust
It is important to build trust with the users not only by answering initial questions before the lead gen flow starts, but also answering any follow-up questions after. An eager user who has an urgent request might want to know when they will get a call back and ask follow-up questions such as, “can someone come today?” Not being able to address this and reassure users is a big missed opportunity. Even a broad response to acknowledge the additional question is helpful and lets the prospect know that they’ve been heard.
At the end of the day, it’s important to remember that the user’s experience is the most important one. The conversation should be focused on the user and their needs, in a way that supports the objectives of the business. While this may seem obvious, it’s important to remember that it’s not just about what the product can do for them – it’s also about how you will help them through each step of the process.
We’ll continue to collect learnings as we grow. Look for more insights and tips in the future!
Yina Smith-Danenhower is Head of Product Design at XAPP AI. Yina has led conversation design on solutions for multiple Fortune 100 customers and previously worked as a Choreographer and Dancer, Architectural Designer and UX Designer. Her diverse experiences have brought rich perspectives to her approach in her work.
