Designing conversational AI that delivers legitimate business results while providing a functional, easy-to-use customer experience, will be critical to success for those who are investing in voice app development using voice driven AI chatbot solutions.
Gartner predicts that by 2020, the average person will have more conversations with their voice activated bots, than with their spouse.
With the rise of Artificial Intelligence (AI) and conversational user interfaces, we are increasingly likely to interact with a bot (known, or unknown) than ever before.
These predictions indicate that conversational AI is here to stay and when it’s designed really well it can make life a lot easier for us as consumers and it can also save businesses a lot of money, as much as 70 cents per transaction, according to Juniper Research.
This could result in billions of dollars in savings for businesses that can develop a successful voice driven AI chatbot, but, designed incorrectly, you will be left with a frustrating user interface that will drive customers away.
If you have used an AI chatbot, it’s likely that you have had a frustrating experience at some point. Often, the reason for the frustration comes because the expectation is there, that conversational AI is really fast and easy to use. It’s almost like going to a fast-food restaurant and having to wait thirty minutes for your meal.
When that experience doesn’t deliver on expectations it is going to cause frustration while also reducing user activity and overall retention. Research from VoiceLab suggests that when a voice app or a skill acquires a user, that there’s only a 6% chance that that user will be active in the second week.
Fortunately, there are things you can do to win over the hearts and loyalty of customers.
Here are four tactical tips for designing conversational AI chatbots:
Users should have a clear path to complete their goals.
Designing an AI chatbot for clarity means that the user should spend less time trying to figure out what you mean or how to do things and more time actually doing things they need.
There are two key areas to consider to improve clarity.
We know words are powerful and the right words can be an invitation to create engagement, trust and enthusiasm with your audience.
Be mindful of the language you are using.
Ensure that the language you use is understandable for the customer and enhances their experience. Avoid internal jargon or terms that only employees would understand.
It is also useful to consider, as you create your conversational flow, that you want the conversation to be what sounds natural, as if two people were actually speaking to each other, which is different than what may sound good in writing.
You can avoid robot speak by using contractions in your language. In normal spoken language, you might say: “I can’t find that”. Whereas in robot speak, you might hear: “I am unable to find that”. Try to keep your language really clear, conversational and friendly.
The way you present choices to your users will strongly impact the clarity and usability of your voice interface.
With a visual interface, there is more room to put in a lot of different choices.
When presenting choices in a voice interface, the path, options and methodology, need to be handled differently.
As an example, IVR phone systems (where you “Press 1 for English etc), provide a perfect example of voice interface design mistakes you want to avoid.
The early IVR systems relied on associating a number on the phone keypad with an action on the back-end, most often to either transfer the call or ask another question.
Even as IVR voice recognition systems evolved to allow you to speak the word for the option you wanted to choose, the voice layer was mostly a speech recognition element sitting on top of the current numerical choice / association framework.
IVR was an add-on to an archaic phone system infrastructure that had limited functionality for enhancement.
Conversational AI voice interfaces are an add-on to the internet infrastructure, which offers unlimited opportunity for enhancement.
Voice interfaces still need to solicit input, however you can often do this conversationally by presenting the options as questions. For example, “Hey, what type of directions would you like, walking or driving?” This is a lot more friendly and natural.
If you want to give the user different options you want to try to limit them to three.
Where you have lots of different options, let’s say you are a stock investment company and it’s not feasible to only have three options, consider whether you could use an open prompt like “Hey please name a stock you want me to look up?”
What this is doing is providing clarity to the user so they know what their next step should be and what is expected to keep the conversation moving. It helps them set expectations and not have to wonder like what to do next.
A good way to start is by writing out the ideal conversation how would it happen from beginning to reaching their goal and try to say it out loud so you can see how it works.
Once you have an initial framework, you can build upon it and develop the conversations in a unique way for your application.
As you are writing your conversations, consider how your chatbot interface will respond in certain situations.
Developing a character or persona profile can be a great guide for writing conversation and help you decide how to construct consistent responses.
If you have ever had that conversation with a customer service rep on the phone and you just know that they’re reading from the same monotone script to every single customer, and it is obvious to you that they are not even really listening to what you’re saying.
These types of conversations are not satisfying…for anyone!
To avoid this, you want to make sure your bot has a very distinct personality and style of talking which will allow you, with a bit of thought, to understand how your chatbot would respond to the situation you are considering.
For any virtual agent to be successful, the aim is allowing users to perceive personality and talk to them just as they would talk to a human friend.
The goal when you’re designing is not to make the user think that they’re talking to a human. That’s not the case you actually should be really clear that they are talking to a bot.
Research from both Microsoft and Facebook has shown that people are more comfortable and more forgiving of any errors if they know they’re talking to a bot and talking to a machine. They understand that the machine has limitations you just want to make the conversation flow in a way that feels just natural.
Personalities can also differ so your persona can differ to answer the same question in different ways.
Let’s say if the user said “Thank you”…
- A professional bot would respond with “You’re very welcome”,
- The friendly bot would be like “Oh you bet”,
- Maybe my more informal, casual, irreverent bot would be like “No problem”.
What if it was a jokey question like “Will you marry me?”
- A professional bot will be like “Listen I think it’s best if we just stick to your professional relationship”.
- The friendly bot will be like “Oh you’re three-dimensional I am non-dimensional…our love can never be”.
- An irreverent bot would be like “Sure take me to alter and let’s see what happens!”.
It’s all the same answer essentially in both of these scenarios but you could see how personality made the response different. You felt like each was a distinct different person almost, and what personality your bot should have really depends on what the use case is.
The personality your bot takes on should reflect the nature of the task.
- A high end financial institution would likely have a professional bot to build credibility and trust.
- A restaurant or entertainment venue has far more leeway to be friendly and engaging.
There doesn’t have to be just one brand voice you can have different personalities based on the end use of the customer.
3. Compassion and Empathy
Regardless of the personality type that you’re trying to build in, if you design your bot well, you can really resonate with your users by trying to really research and understand their core need that is delivered in a way that encompasses compassion and empathy.
Virtual agents can build one-on-one relationships with the audience because you can talk to them on a very individualistic basis.
If you were designing a banking app where the user wanted to come and transfer some money. The bot should be designed to really understand what your goal is.
If transferring that money would put you at risk of overdraft, even if you didn’t ask for a balance check, the bot could warn you that “Hey, you know if you transfer this you’ll be at risk of overdraft and that there will be overdraft fees. Do you still want to go ahead and do this?”
If you try to tap into the need, if you want to step into your customers shoes you can build much more loyalty and understanding.
Doing research before you start designing allows you to see what it takes and understand what the users goals are and what their pain points could be, then you really can build a voice based solution that meets their needs.
One less obvious area where empathy can be integrated into a chatbot app is tied to human nature, and the need for some casual small talk.
This is becoming more common and is also an area where a lot of bots could improve by adding in contingencies for off-script chit-chat.
If someone is to go off-script for a little fun to ask “Are you real?”…nobody likes to hear the response “Sorry I do not understand!”…so as you’re designing your bot, try to build in different small-talk scenarios to help you avoid that. This also goes back to thinking about your character or persona profile to decide how to react.
Try to build different types of scenarios such as the greetings or questions a normal conversation contains and plan for some potential chitchat if you can.
While, you can never completely eliminate the potential that mistakes will happen.
One main thing to avoid, however, is that you don’t want to say “Sorry” too frequently, as it can become very repetitive and annoying.
- I’m sorry I don’t understand that.
- I’m sorry I should understand that.
- Sorry I do not understand that.
You can find different ways to convey the message but clearly that can be too much of a good thing when it comes to saying sorry.
There are lots of ways to correct an error without having to say sorry and one of the clever ways to keep the conversation going is to offer alternatives.
If the user went to a flower bot and said “Hey please order me a dozen red tulips” and there were no red tulips in stock, rather than saying “Oh I’m sorry we don’t have any red tulips”, it would be so much better if it said “We’re out of red tulips, would you like yellow or white tulips instead?”.
Designing a conversational flow in this way makes it so much easier to keep the customer happy and simpify the process in a way that also helps you increase the chance of making the sale.
When designing conversational AI chatbots, if you can deliver a solution focused on clarity and helping the users achieve their goals, can add in suitable personality and character traits, while embedding empathy and compassion with correct guardrails in place for correction then you can build a solid foundation for providing exceptional customer experiences for happy, loyal customers.