Unless you lived underneath a rock for the last couple of years, you’ve probably heard about chatbots. A lot. In fact, there’s so much information circulating the web right now, that it’s becoming increasingly difficult to get a clear picture. Are they the future of online retail? A cheap gimmick bound to die a quick death? Something in between?
It’s about time we move past the hype and start coming up with a more definite guideline for these little digital helpers. To make sure you really get set off on the right foot, we did some research of our own. Combining our insights in eCommerce and behavioral psychology, we cooked up the perfect strategy for effective chatbot use, based on their strengths and weaknesses.
Who’s to believe when it comes to chatbots?
Just to give you a faint idea of what the ongoing chatbot discussion looks like, here’s an overview of several conflicting opinions and statistics out there.
And that’s just a fraction of the amount of articles that’s floating around the internet. With so many conflicting findings, it gets hard to determine just what to believe. When it gets down to it, most of what’s being said or written is just white noise.
Chatbots throughout the years
Let’s back it up a second. While chatbots may seem like a relatively new concept, they have been around for ages. The first real chatbot ever created ELIZA, a textual program designed to engage in simple text-based conversations with humans, famously passed the Turing Test more than half a century ago.
Since then, there have been plenty of other early chatbot examples, varying from amusing diversions to unwanted PR disasters. Remember Clippy, Word's annoying paperclip giving you advice that was never asked for?
Or Cleverbot, a snarky online conversationalist handing out witty comebacks to bored teenagers? And let’s not even get into that one notorious experiment where Microsoft learned the hard way that teaching an AI how to speak from Twitter’s use base is guaranteed to go horribly wrong.
Thankfully, companies have figured out there’s more to chatbots than internet shenanigans. There’s a real potential value in virtual assistants programmed to help online shoppers: they’re available 24 hours a day to answer questions and a cheaper alternative to human customer support in the long run.
Recently, many different brands and online stores have come to this conclusion and started using chatbots on platforms like Facebook Messenger, Wechat, Kik, or their own professional websites to provide chatbots with assistance.
To give you a few examples there’s a Pizza Hut ordering service, a KLM flight update, a virtual Starbucks barista, a bot that orders your Uber, a H&M personal shopping assistant, amongst many others.
But are they really able to help us just as well as a human assistant would? To put things to the test, we’ve decided to visit H&M’s shopping assistant for a quick chat.
"Thankfully, companies have figured out there's more to chatbots than internet shenanigans."
Example: H&M’s Shopping Assistant
When it comes to shopping, I tend to be very picky. I take my time browsing through several webshops for inspiration, read blog posts and reviews for comparisons, and usually go through loads of different options until I find the one item that’s perfect for me.
Because of this, behavioral psychology would likely place me under the category of maximizers. As opposed to my satisficer counterparts - who generally stop looking as soon as they find something they like - I tend to experience more stress under an overwhelming amount of choices and will be more likely to experience regret after my decision. Yikes!
Luckily, this is where a chatbot could step in to help me out. With the help of some nifty AI features, a virtual shopping assistant might be able to reduce choice overload by presenting me with a selection based on my personal preferences. I decided to have a go at H&M’s shopping assistant that’s currently available on a messenger app called Kik.
Creating my personal style
Immediately, things got off to a shaky start. My virtual assistant needed to figure out my “personal style” by asking me some questions. First, it asked me which demographic I felt most closely related to (I picked “20s”). It then had me choose between two different outfits several times.
Being the tough customer that I am, I went with “neither” three times, after which, my assistant gave up and just asked me to choose a word that best defined my style. With my thoughts already drifting to my summer vacation, I realized a could do with some new pieces for my summer wardrobe and boldly hit “surfer.”
After getting through the awkward “getting-to-know-you-phase,” the fun was about to begin. I could either create a new outfit by picking out clothing items, rate other pre-designed outfits, or look for new items. I clicked “create outfit,” and, immediately, my assistant asked me to choose between some seemingly random pairs of jogging pants. I never asked for jogging pants, nor was I looking to buy them. Next up, a tank top, straw hat, and shoes, all of which, I wasn’t particularly excited for.
"Ultimately, the chatbot was just operating on a very flat, one-dimensional view it had just acquired of me."
What went wrong?
As you can imagine, I quickly became frustrated with my assistant for several reasons. First, it kept making the wrong assumptions on what I was supposed to like and what I was looking for. This quickly changed to disappointment over the lack of choice I was given. The few items that were included were not what I was looking for, and my maximizer brain immediately got me panicking about the number of better options this bot was holding out on.
My personally handpicked outfit. Quite funky, isn’t it?
More importantly however, I was unable to engage in an actual conversation to voice any of these concerns to my assistant. There was no way to explain why I didn’t like the options it presented me with, nor could I make any suggestions on items I would prefer.
Ultimately, the chatbot was just operating on a very flat, one-dimensional view it had just acquired of me. To be fair, I couldn’t really blame it for that. We had only just met. What did it really know about me? That I’m in my twenties, dislike a lot of outfits, and enjoy surfing.
Sure, that might describe me to some extent, but it doesn’t mean I’ll start wearing flip-flops and puka shell bracelets to work everyday. And definitely no jogging pants or straw hats.
Psychological strains of chatbots
You’ve probably gathered by now that chatbots are far from perfect. But they do have the potential to greatly improve the user's experience by reducing their psychological barriers to online shopping.
So, before we get into defining the best chat bot strategy, let's learn from our predecessors' mistakes and take look at few of the most common flaws.
First off, one of main contributors to negative user experiences with chatbots is setting the wrong expectations. With all the hype that is currently circling around chatbots, people tend to get unrealistic ideas on what they are able to do for them.
Moreover, designing your chatbot as an actual person, with a name, appearance, and personality, creates the expectation that a chatbot will be able to assist you exactly the way a human assistant would.
Obviously, it won’t. Consumer psychology teaches us that customer (dis)satisfaction is based on discrepancy between expectations and perceived product performance. In other words, without any form of expectation management, many chatbot users will inevitably end up disappointed.
Chatbots aren’t that smart
It’s a hard truth to face but one that needs to be addressed: most chatbots aren’t that smart. While some operate on true AI, the large majority of them use a simple decision tree model. This means they require specific phrases or keywords from their users, to match with the right corresponding responses.
Now let’s consider the way we approach solving problems. Here too, we can make a distinction between matters that are simple or complex. Simple problems have distinct goals and a clearly defined path to follow towards reaching the desired solution. This is ideal for chatbots, who can quickly match an inquiry with the right answer. Rather than being actual problem solvers on their own, they help by directing us to possible solutions.
But what happens when you don’t know the exact input to give to get the answers you need? In the case of complex problems, there’s no clear definition of the state of the intended goal, what factors will come into play, and what barriers need to be overcome. They result in open-ended search questions, where users aren’t even sure what it is they’re looking for.
Finding a solution for such problems requires more creativity and out-of-the-box thinking from the problem solver and, unfortunately, that’s not what chatbots are designed for. Even using words outside of their vocabulary can result in failure to resolve simple requests. Due to this lack of flexibility, most people still prefer human assistance over chatbots for more pressing matters.
"Rather than being actual problem solvers on their own, chatbots help by directing us to possible solutions."
If that’s not enough to drive you crazy, I don’t know what is.
Missing the human touch
But there are more ways in which chatbots fall short of their human counterparts. A good analogy to explain this is John Searle’s Chinese room thought experiment.
Imagine yourself locked inside room. Every once in a while, a message written in Chinese characters is being shoved underneath the door. You take this message to a computer that provides you with the matching response, also written in Chinese.
The person on the other side of the door would be convinced he/she is having a conversation with someone who speaks the language fluently, but in reality (unless you actually speak Chinese but for the sake of the argument let’s assume, you don’t) you don’t have a clue what you’re talking about.
This very much applies to chatbots as well. They might be able to simulate a conversation, but they lack the ability to actually reason with or understand is being discussed.
While machine learning enables AI to better predict your preferences and behavior, this requires a lot of data. As we know in today’s digital climate, with the ongoing Facebook scandal and GDPR looking to shake up eCommerce landscape entirely, people are naturally reluctant to share their information online. The alternative, to have a chatbot learn your preferences through continuous conversation, requires a lot of time, which most people won’t be willing to invest.
And then there’s the issue of empathy. Unlike human assistants, chatbots aren’t able to place themselves in the situation of the customer or relate to their emotions. And it just so happens to be that these qualities play a major role in customer satisfaction and brand loyalty. Granted, there may be some exceptions where AI comes pretty close to simulating empathy - virtual therapist Woebot actually booked some promising results - but these come closer to self-help tools than establishing deep connections.
Disruptive to UX
All shortcomings combined, chatbots are guaranteed to be a bigger disruption to the user’s experience than they are helpful - when not designed the right way. As you may have learned from one of our previous blog posts, fluency is crucial for creating a positive user experience. Chatbots are no exception. It takes careful design to guide users through the conversation smoothly and has plenty of potential pitfalls that need to be avoided.
The main problem with many conversational search tools out there is that they either have you following too many steps or too few. For instance, when you need to make a quick trip to the supermarket and ask a weatherbot if you should bring a jacket, you probably won’t be very excited to spend a few minutes specifying your location and the exact time you’re leaving your house before getting an answer.
Having users walk through too many questions before getting an answer is a sure-fire way to overcomplicate a simple request, and might even induce them with cognitive overload.
On the other hand, keeping things too vague or open-ended doesn’t work either. Opening a conversation by only asking, “What can I do for you?” without directing the user towards any follow-up steps, will likely result in confusion.
"All shortcomings combined, chatbots are guaranteed to be a bigger disruption to the user's experience than they are helpful - when not designed the right way."
Solutions: How to effectively use chatbots
Set the right expectations
Disappointment is something you’ll certainly want to avoid, which makes it important to lay out the rules very clearly from the start. The first thing you’ll want to get out of the way: let your users know they’re talking to a bot!
This may seem like an obvious note but something that’s still overlooked by many programmers. Especially in the case of chatbots that appear on your webshops homepage via a pop-up dialog, people like to know whether there’s an AI or an actual person on the other end. This already sets users up with a certain idea of whether a casual tone or precise phrasing is appropriate and whether or not simple pleasantries like “Hello,” “How are you,” and “Thank you,” are desirable.
You’ll also want to clearly indicate at the start of the conversation the purpose of your chatbot. What can it do for you, and what not? What kind of input does it need from you to give you what you want? Giving users a quick heads-up on what’s about to follow allows them to manage their expectations.
CNN gets it right.
Steer the conversation
Next, try to design your chatbot in such a way that the conversation keeps running smoothly and works towards reaching a certain goal. This means guiding the user through the process by pointing them in the right direction for each next logical step. It can also help to give the user a few choices to choose from per question to narrow down their options.
Throughout the conversation, be sure to regularly provide feedback. Display which information the chatbot has gathered from its user so far, so they can check if everything is in order.
For example, when using a chatbot to book a trip, have the chatbot repeat the important details back to the user, such as the destination, travel date, number of travellers, after each answered question, to let the user know he/she’s on the right track.
Additionally, when the user needs to go through a longer process to reach his goal, provide a visual on their progress. Something like: “6 out of 9 questions answered, you’re almost there!” When the conversation is completed, let the users know they’re done, and be sure to ask for feedback. “Did you find this conversation helpful? Yes/No.”
Ease the buyer’s journey
In our blog, we’ve spoken before on how to guide customers through the buyer’s journey seamlessly by considering the psychology behind the online sales funnel. A similar approach can be taken for chatbots. But to do this, thorough research should be done on your target audience to understand who your bot is talking to.
Using those insights to optimize the user’s experience, as well as chatbots’ functionality, chatbots should be designed in such a way that they fit smartly into the buyer’s journey. For each step, consider where the customer is in the buyer’s journey, and what kind of assistance is required at that particular point. Are they still looking for information?
Provide them with a FAQ list. Are they suffering from choice overload? Include options like: “show today’s deals” or “show most popular items” to narrow down the choice. Do they want to look at the items in their shopping basket? Include options like: “display shopping basket” to have them change, remove, or add different items. And so on.
"Once we start figuring out how to use chatbots in a more strategic way, they can ultimately be the oil that greases the gears of the buyer's process, with increased conversion rates as the ultimate pay-off."
Keep it simple!
Keep in mind, chatbots are unreliable for too large and complicated requests - people prefer chatbots that keep it simple! According to a recent report, chatbots are valued most for being able to provide 24-hour service and being able to provide them with a quick response to things they’d rather not look for themselves or bother a human assistant with.
That makes them especially useful for helping out with small, tedious tasks, such as looking for a checkout, placing an order, answering commonly asked questions, or providing information about their shipment. Spare the user the effort of having to look up this information by themselves and make each step from awareness to checkout a little bit easier.
Once we start figuring how to use chatbots in a more strategic way, they can be the oil that greases the gears of the buyer’s process, with increased conversion rates as the ultimate pay-off.
So, after a thorough dissection, what have we learned? As of now, chatbots are basically still children. While some might assume they have a bright future, they can’t be expected to solve all our problems just yet.
They certainly aren’t ready to take over the role of human assistants, and for some tasks, they never will. And that’s okay. Chatbots have different areas in which they can shine.
To summarize your key takeaways:
- Chatbots are limited in their capabilities. They lack the flexibility to solve complex issues, and can’t provide us with empathy like humans can.
- However, they do have a few specific advantages: 24-hour availability, and being able to match small questions with quick answers.
- Nevertheless, it’s important to manage your users’ expectations. Steer the conversation, give them regular feedback, and make things easier for them.
- Fit chatbots smartly into the buyer’s journey, to optimize your user’s experience.
- And most importantly, don’t overcomplicate. Keep chatbots busy with easy tasks until they’re ready for the big leagues.
And that’s that!
Found this article helpful? At Crobox, we specialize at all things related to consumer psychology and eCommerce. Curious to see how we can help to improve your website? Download our white paper to find out!