Building great dialogue for a virtual agent requires finesse and a clear understanding of the needs of your customers.
Interacting with conversational AI should be as easy and natural as talking with a human. Achieving this, however, is not nearly as simple as it seems.
A virtual agent might be the first (and, in many cases, only) point of contact someone has when interacting with your business. It should be a shining example of how you want people to perceive your brand — the ultimate ‘model employee’ — not a messy chatbot that is constantly leaving customers with a bad impression, potentially turning them away for good.
To that end, we have established some guidelines that should help steer you towards building better conversations.
Most communication is characterized by the context it occurs in. For example, when asked if you’d like an energy drink, responding with “That will keep me awake”, has a drastically different meaning first thing in the morning as compared to later that same night.
When writing dialogue for a virtual agent, being mindful of how a query can be interpreted is the key to understanding it.
Want to truly help your customers? Then you should consider starting by walking a mile in their shoes. Empathy is a crucial part of writing good chat dialogue and it’s important to have your virtual agent come across as such.
If a customer comes to you with something positive, your virtual agent’s response should reflect that. Similarly, if they’re approaching you with a delicate matter, it’s advisable to hold back on overusing the winking-face emoji. There’s a time and a place for jovial replies, but it’s more important to try and show your customers that you recognize their situation. It will go a long way to helping build loyalty towards your brand.
Before you can accurately manage the expectations of someone using your virtual agent, it helps to know (as much as possible) who they are. Don’t skip customer insight and mapping and get them to talk about their needs and expectations.
It’s also important to be transparent - customers should understand that they are chatting with a machine. Your virtual agent should make this clear while also outlining what it can and can’t do, minimizing any potential confusion for customers.
It’s far easier for humans to express ourselves when interacting face-to-face, rather than in writing. The “mmm’s” and “aha’s” we casually throw into conversations work in concert with facial expressions and body language to convey a sense of empathy which is more difficult to achieve over chat.
A virtual agent powered by conversational AI can, however, still inject personality into its interactions by using emojis (with restraint, depending on the situation!) and humor. The skill comes in establishing the right level of personality — too little and you might alienate your customers, too much and you run the risk of coming off as overly familiar.
An interesting trait in human conversations is that we tend to mimic the tone of voice, choice of words and phrasing of the person we are talking to. We should be mindful of this when building chat conversations. If your virtual agent responds to customers with long, wordy answers it sets the expectation that it is able to understand long and wordy questions. It can be frustrating if, instead, it turns out that your virtual agent isn’t quite as intelligent as its inflated vocabulary makes it out to be. Tailor your conversational AI to fit the situation and you’ll leave a much better lasting impression.
You can try to manage expectations as much as possible, but the truth is: mistakes will happen. The important takeaway is to not try and build flawless conversations (you can’t!) but instead to prepare for the inevitability of mistakes (whether human or machine-made) and find ways to get the conversation back on track.
One of the advantages of our conversational AI is that, thanks to Automated Semantic Understanding (ASU), we are able to have a better overall understanding of a customer’s query than a standard chatbot. This significantly reduces the rate of bad false positives by up to 90 percent.
While ASU is something unique to our solution, there are still some best practices you can follow to minimize mistakes in general, including: