As customers, the thought of calling support or trying to find the information we need on the “help” site, can fill us with dread. Sadly, many times the support experiences we have had throughout our lives have been exceptionally frustrating. Typically, when you need support, you are not in the best of moments anyway. Everything, from the irritating music that plays while you are informed that you have a 34-minute wait, to the voice recognition that doesn’t seem to recognize your voice, to the representative that runs through the script, feels that much more frustrating.
Chatbots, when designed well, offer a way to escape these experience pitfalls. They can quickly lead the customer to their desired outcome, in an engaging, natural way. However, to reap the rewards, the experience needs to be designed strategically. Sadly, many support chatbots fall short of the promise for both the customer and the business. Many chatbots today often get confused when a user deviates from simple linear flows or set an expectation far higher than what they deliver, leading to a discouraging experience. Operationally, many times the promise of the chatbot results in various decisions around capacity and demand for other support channels. When the chatbot does not meet the promise, this can cause issues across the other channels impacting both operations and the customer experience.
At EffectUX, we create experience success models. These research-based models define the attributes needed to deliver a desired experience. We use these models to evaluate experiences, the results of which reveal which actions need to be taken to attain the outcome or goal. From research and practical implementation of our chatbot experience model, here are some considerations for those creating support chatbot experiences. While some of them are easier than others, it is worth thinking about each!
1: Determine the value of your chatbot.
In excitement, with the best of intentions and to meet their timelines, teams speed forward creating their conversation flows. However, this rarely results in a great experience or gains the business benefits promised. Starting with questions like, what’s the goal of your bot? What is its purpose? What experience are you looking to deliver and to who? How are you measuring success? All help you gain clarity that serves as a common, shared direction across teams. It will also help you focus on the right problems to solve via chatbot, as not all problems will be a good fit for a bot conversational experience. Simply, you are better off doing a carefully selected core set of issues really well, then growing it out from there, rather than trying to include everything and doing it poorly. Here, you can use your data to help guide you. Considering questions such as, what are the top issues? Are they well known and defined in terms of triage and solutions? What value will the conversation experience bring each particular case?
2: Who is your bot?
The bot’s personality plays an important role in the experience. The key here is thinking, is the personality a fit for your audience? The personality impacts how the bot responds and how your customers feel. Using customer insights and testing can help you make better decisions, but at the end of the day, you want your bot talking in a way that works for your audience.
3: Create a conversational feel.
Even though customers know they are chatting with a bot, given the interaction is a conversational one, when it doesn’t feel natural it can cause the experience to feel “off”. Remember, we all have our own mental models of the world based on our experiences. Often, a customer’s point of reference will be chat applications that they have used in the past or use today. Things like timing and response eminence can help the interaction feel more natural. When a user types something, then nothing happens, then suddenly the bot’s response or responses comes up on the screen, it can feel abrupt and jarring.
4: Be wary of canned answers.
Part of a conversational feel is responding in context. Understanding the key language types that are commonly used can help you create constructs that best respond to different situations. Of course, there are going to be defined language models and sets, however, make sure you have reviewed how they materialize in flows and in context. It is extremely frustrating when you are telling a chatbot your issue and having it respond with something that makes no sense, or something out of context like “awesome”, when what you have told it is not so awesome.
This leads to another point, set expectations. If you cannot handle something yet, that’s ok, it is better to let the customer know as soon as possible. Especially when you are starting out with your chatbot and getting customers accustomed to it, even letting them know the scope at the start of the conversation, or the common requests that you do not handle yet, helps them cut out to the right support channel without wasting more time.
5: Don’t send your customer loopy.
Same as with that voice tree on the phone, going in loops repeating yourself to no avail, a looping chatbot inspires similar feelings of frustration. Typically, on the third repetition or loop, most people get annoyed and want to abandon the process. The first time, it’s, “oh the chatbot didn’t know, let me write it a different way.” Keeping in mind that the customer is already likely feeling frustrated by virtue of their situation needing support, the next time it’s, “this stupid f*#!ing chatbot”. Making sure you have loop limits that guide customers to a way to get the support they need in this case, having a path out such as an escape option, utilizing quick replies and option selection can all help.
6: Look at how the bot plays a role with the rest of the support ecosystem.
Often there are other support channels and flows. How does your bot interact with them? How does your experience connect? For example, if the bot asks a customer questions to gather information, and then passes them to a representative, does the customer have to repeat everything all over again? Repetition is one of the biggest experience detractors from the “passed call” or escalation experience and it is no different if the point of contact is a bot first. Thinking about how your customer moves between the channels, how data is used and transferred, and how each touchpoint works together, will help you be more purposeful in your design.
7: Optimize conversation flows.
Seems obvious right? However, often the customer is left thinking, “shouldn’t you know that already, why are you asking me!?” Looking at what information your bot has access to, can remove customer input and effort. The bot can then focus on guiding the conversation to solicit only the direct information needed to bring the customer to the right next step and gaining clarification when unsure.
8: Investigate user language norms.
Like many things, this will be a part of continuous learning, however, it is worth getting started on the best foot. Spend some time thinking about the language your audience uses. You can investigate this using existing data or conducting some insight gathering. For example, how are customers articulating problems on other channels? Which words, patterns and correlations in language exist? What are common request words or confirmation words? For example, in the context of a confirmation flow, “y”, “yeah”, “yea”, and “yep” can all be ways your audience may say yes when in chat mode. Speaking of confirmations, make sure you confirm actions, especially destructive ones, such as deletions or removals.
9: Distinction between problems and identifying change.
Perhaps a customer has two or more issues or pieces of information that they mention. How does the bot respond in this case? Or, perhaps the customer changes their mind mid flow. For example, the bot asks them for some information, they respond, then realize they made a mistake and say, “actually its [x].” What does the bot do? This also brings up the topic of track back, which can mean looking at conversation history. How does the bot track back to previous points in the conversation? While these interactions are more complex, even if you don’t handle them yet, it is worth thinking about how the bot responds.
10: Keep learning.
While it is wise to spend some time thinking strategically so that you get started in the best way, you also need to get the bot used by people and start learning. Here, consider questions such as, what are your learning mechanisms? How are you gathering feedback? What are you learning from the interactions and for what purpose? How are you implementing improvements to the experience and expanding on language, ruling, conversation flows and constructs? How are you creating more intelligent interactions? Don’t forget the customer side too. How are you gleaning deeper insights about the customer experience and the customer perspective and sentiment? How are you gathering user input to help guide decisions?
These are just some considerations. Like any experience, think about what your desired support chatbot experience is, your various audiences and how they interact with the experience, how you measure success, the customer flows, and the experience ecosystem. When chatbots are purposefully designed experiences they can provide positive support interactions which reduce frustration, optimize utilization of support channels, and get the customer back to doing whatever they need to be doing quickly.