Building your chatbot
What to know to design a conversational UI that works
For companies wanting to maintain a competitive edge in their web and mobile experiences, creating a chatbot should be at the top of their wish list. The coronavirus pandemic has only caused a greater need for brands to be able to quickly and effectively engage their users with empathy in an effort to maintain trust and mitigate uncertainty. According to a recent Smarter With Gartner article, “Gartner predicts that by 2022, 70% of customer interactions will involve emerging technologies such as machine learning (ML) applications, chatbots and mobile messaging, up from 15% in 2018.”[1] The appeal of chatbot technology is undeniable.
However, as with any new and developing technological trend, the quality of chatbots that are being created today varies widely. Some are robust and closely resemble AI; others are ineffective (or worse, inaccurate) — but if you spend any time interacting with bots across various sites, you’ll notice there isn’t a strong industry standard. Just a couple of years ago, a Spiceworks study reported that 59 percent [of respondents] said that [AI chatbots/intelligent assistants] have misunderstood the nuances of human dialogue.
So what does a team need to do to ensure that the chatbot it develops actually serves its purpose and justifies its creation? How can you ensure that your users’ needs are being met during this time of massive change? How do you guarantee that your chatbot is aligned with your broader brand experience? Here are some best practices for developing a chatbot to help you start off on the right foot.
Decide what purpose your chatbot is going to serve
As useful as chatbots are, you probably can’t solve all of your issues with one bot. When you first embark on the undertaking of creating a conversational user interface (UI), it’s important to narrow your scope and determine what specific function you want your chatbot to serve. To do this, it’s crucial that you have a measurable goal in mind, such as easing customer care call volume or reducing cart abandonment. By isolating the type of interaction you want your chatbot to engage in, you will be serving a specific user need, which will ultimately create a more personalized experience. Here are a few possible types of chatbots to consider:
Troubleshooting bot: Guides users through self-serve troubleshooting steps; can escalate to support handoff if the issue is more serious
E-commerce bot: Assists users with determining which specific product will serve their needs best, provides specifications and details, and guides users to make a purchase
Training bot: Helps employees with onboarding, certification, and basic internal software training
Order tracking bot: Provides users with order tracking information and shipment updates
After you determine the objective and scope of your chatbot, you can set to work laying out a rough framework for what the user’s interaction with your chatbot will look like as well as a strategy for how you’ll track performance for optimizations down the line.
Know your subject matter experts
You might have a super specialized team working on developing your chatbot content. Or you might be using your content team. Or you might be working with an agency partner (Hi!). Whichever the case, it’s more than likely that the individuals creating the content for the chatbot are not the actual subject matter experts (SMEs) in that area (like technical troubleshooting experts, product marketers, etc.). That’s why it’s important to establish early on in the process where you’ll get the information that you’ll use to create the chatbot script.
After you’ve determined where your fountain of knowledge for developing chatbot content lives, start setting up interviews with the appropriate individuals. You should be able to confidently explain exactly what type of interactions your chatbot will be simulating, including the specific types of user questions and desired outcomes. If your SMEs don’t have time to be interviewed, send them some hypothetical question-answer pairings to review for accuracy.
Another tactic for generating content is to examine customer support logs. Note how your users refer to products and processes, and try to predict user pain points and frustration. The more you can model your chatbot script around real, positive customer interactions, the better it will be.
If this seems like an obvious or skippable step, remember: It’s not. Make it a priority to check that the individuals you need to interview for your chat material will have bandwidth for your cause.
Give your chatbot a voice
This is the part where you get to play Dr. Frankenstein — sort of. To give your users the best interactive experience, you’ll want to develop a name and personality for your chatbot. Remember that your chatbot should represent your company as an additional touch point in the customer experience and that, when using chatbots, 64 percent [of consumers] want AI to be more human-like. That means also creating a voice and tone style guide to shape the content that your chatbot will be delivering to your users. You can draw up scales to illustrate your bot’s formality, empathy, humor, and more. What is your bot’s welcome message and salutation? What kind of fallback messages will your bot use when it can’t determine what the user is asking?
Although some of this content will be net-new, it’s also important that you ladder back to your primary content style guide to maintain brand integrity and keep the brand experience consistent throughout. If your audience is IT professionals, “Hey fam,” is probably not on brand. Again, your chatbot is simply another touch point in your brand’s ecosystem; its content should be a reflection of that alignment.
Get Ogilvy’s stories in your inbox
Join Medium for free to get updates from this writer.
Subscribe
(NOTE: It’s also fair to say that the majority of consumers are experiencing a great deal of anxiety right now. It might be a good time to really take a close look at the empathy scale you will be applying to your chatbot. This may extend to your making some adjustments more broadly to your entire content style guide. Ask yourself what type of language a user might prefer to interact with if they were undergoing significant stress, and always design with compassion.
Testing, testing, testing
It’s not an exaggeration to say that testing your chatbot with real users (or guerilla testing with colleagues) is just as important as the actual content development. If you’ve ever interacted with a chatbot that responds with messages that are utterly useless or laughably bad, chances are it wasn’t properly tested.
I recommend developing a rigorous regimen of testing (guerilla and user) to determine preferences for every variable you can come up with. Use surveys and moderated tests to narrow it down to the most successful, effective interactions. A few of the variables you might consider testing for are:
Response length: While best practices say to keep messages short and succinct, you’ll want to get more granular to determine what the message length threshold is based on your business, bot type and user preference.
Tone: It’s not a bad idea to begin testing when you’re first developing your chatbot style guide. You can gauge the best-received voice and tone initially and eventually test for tonal shifts for longer interactions (this is especially important for escalation instances, where a user’s stress or frustration levels may be gradually increasing).
Images and media: As you begin to better understand the preferences of your test participants, you might want to experiment with adding reference images or illustrations to your chat interface (used correctly, this can add a good amount of clarity to complicated instructions).
As you isolate and test specific variables, you can continue to update your chat content and document a running list of your test findings for future reference.
Document everything and optimize
It’s worth repeating from the testing step: Document everything. Consider creating a formalized chatbot playbook that includes all of your learnings. Some possible chapters might be:
Findings from your user tests and user preferences
Voice and tone style guide
Directory of SMEs and their areas of expertise
Chatbot testing notes (what worked and what didn’t)
Tokenized messages and fallback intents (welcome message, unresolved query message, escalate to customer support team message, etc.)
By creating a playbook, you’re making it easier for future chatbot project teams at your company to employ tried-and-true messaging parameters and communication methods. After all, if you keep these recommendations in mind, your first chatbot is going to be so successful that leadership is going to want to deploy another.
Final thoughts for future bots
There are plenty of resources you can explore on how to write for chatbots, including Chatbots Magazine and Erika Hall’s wonderful book Conversational Design. And there are seemingly endless considerations for how to make bots communicate more effectively and ways to get the most out of these user interactions. You may want to include a customer satisfaction (CSAT) survey at the end of all of your interactions. Or you might consider putting a concerted effort toward developing more training phrases for your bot (the pilot we developed relied mainly on buttons for user responses, but ramping up bot training phrases can make interactions more lifelike for users). If you work with seasoned user experience (UX) writers and UX testing specialists and remember to keep your chatbot communication strategy aligned with your broader brand experience, the world is your conversational oyster.
[1] Smarter With Gartner, “Top CX Trends for CIOs to Watch,” February 27, 2020. https://www.gartner.com/smarterwithgartner/top-cx-trends-for-cios-to-watch/
[2] Spiceworks, “Data snapshot: AI Chatbots and Intelligent Assistants in the Workplace,” Spiceworks blog, April 2, 2018, https://community.spiceworks.com/blog/2964-data-snapshot-ai-chatbots-and-intelligent-assistants-in-the-workplace
[3] Capgemini Research Institute, The Secret to Winning Customers’ Hearts With Artificial Intelligence: Add Human Intelligence, July 2018, https://www.capgemini.com/resources/ai-in-cx/#
Author: Scott Beck, Sr. UX Writer
As a Sr. UX Writer at Ogilvy, Scott is passionate about taking a user-centered approach to content and design. He enjoys partnering with clients to help them develop and improve their content design ecosystems. Scott’s experience spans UX copywriting, content design, product content strategy, and business-to-business (B2B) content marketing. He currently lives in Denver, Colorado, where he plays music and writes creatively in his free time.
