Worried your Chatbot isn’t meeting your customer’s expectations?
CX experts agree that if your customer is not happy, your business has a slim chance of survival. A happy customer comes back for more. However, customer satisfaction is feeble. A single poor interaction is enough for your customer to go to a competitor.
An effective enterprise chatbot can boost customer satisfaction instantly. Chatbots offer the customer the opportunity to interact directly with your company.
What do customers want from a Chatbot?
“63% of customers will leave a company after just one poor experience, and almost two-thirds will no longer wait more than 2 minutes for assistance.” – Forrester Chatbot Report
To put it simply: Customers want a quick and easy interaction that helps them achieve their goal.
When implemented right, a chatbot improves the customer experience by:
- Giving instant responses without waiting for a live agent
- Is available 24/7 at a time that suits customers
- The conversation is consistent with your brand tone and message
How can you ensure your chatbot meets customer expectations? In this article, we will give you 3 essential tips to get you started.
How Chatbot Fails Affect Your Customer Experience?
If you interact with chatbots when shopping, looking for information or making transactions, chances are you’ve met with some mishaps.
Common Chatbot Fails
Bots that don’t answer what you are asking, that don’t answer at all or repeat the same answer repeatedly can annoy a user. This typical fail can upset an user at the point of not coming back to your company. Let’s look at some common chatbot pitfalls:
The Endless Loop
AI-powered chatbots are increasing in popularity. But most chatbots in the market now are rule-based. Rule-based chatbots answer according to a predetermined chart flow.
That means that if the user query is not in the decision tree, the chatbot cannot recognise it. The bot gives the answers it has available while trying to understand the query. If you don’t give the conversation a way out, it may cause an endless loop that frustrates the user.
Give the bot two attempts at understanding. If it fails, direct the conversation to a human support agent. That way, you avoid annoying the user.
Misunderstandings
Sometimes bots misunderstand queries. With rule-based bots is often because the query is not yet built into the conversation tree. AI-powered bots need constant training to improve their understanding of queries. Lack of training can cause some funny and annoying situations like below:
Personality Issues
Every chatbot needs a personality aligned with your brand. The bot is the face of your company, so it should have the same tone and voice. For instance, banking bots have a more professional tone than fashion assistants bots. But, there are general rules that apply to all types of bots:
1. Not too stiff or chatty.
A bot that is too dry can put off customers. Be sure your conversational design includes a bit of small talk. A little humour doesn’t hurt either. But be careful that your bot is not too chatty. Customers reach the bot for a reason. Distracting them with useless chat can annoy them. See the example below of Poncho, the weather bot.
Image source: Poncho the weather cat
2. Not offensive.
Gear your chatbot to your target market. Sometimes an innocent joke can cause offence to your target market. For instance, Tay, the Microsoft chatbot, was reportedly tweeting racist jokes and comments. The bot was mostly retweeting other users’ racist statements.
Image source: TechCrunch
But the nature of AI means the bot actually learns from interactions. Logically, it caused an uproar and Microsoft turned the account into private. Only approved followers can see now @TayandYou’s tweets to prevent inappropriate learning.
So, what are the consequences of a poor chatbot performance?
Customers that have a poor experience with a chatbot are less likely to use it again.
Fewer consumers using the chatbot, in turn, result in a lower or slower ROI.
In addition, since there are fewer interactions, the bot’s NLU has less data to learn from.
Is it really the fault of the Chatbot’s NLU?
There is a common belief that when there is a problem with a chatbot it is the fault of the NLU. In fact, it is the lack of alignment between the NLU capabilities of the chosen platform/engineers and the conversational design which causes most chatbot failures to meet expectations.
4 Reasons Your Chatbot NLU Is Underperforming
How does Natural Language Understanding (NLU) work? The NLU engine receives the user’s input in text format (sometimes converted from speech). An”intent” and any additional “parameters” associated with it are extracted. Examples of intent can be “opening times on [saturday]” for a venue chatbot, or “what’s [my] [savings account] balance“, for a banking bot.
The percentage of language your chatbot recognises determines the experience your user has. We believe a good accuracy percentage is from 95% and up. Even an 80% accuracy means that the bot will not identify the correct intent 1 in 5 messages your user writes. That’s not ok and will surely leave your customers annoyed.
So, what is behind your chatbot not understanding your customer? There are four “common culprits”:
1. Poor Triage of NLU errors.
Deciding which user’s the chatbot recognised correctly and which not can be a real challenge. Moreover, as people start to ask related questions, you have to decide if mapping a new, specific topic or not.
2. Intent clashes aren’t identified quickly.
When training utterances overlap your NLU can have “intent clashes”. For instance, when you have NLU coverage for “what are your opening times today”, and people may start asking for “typical delivery times” or “do i need to keep the original packaging for returns when opening my delivery”. Without the right analytics or tools, ‘decoys’ to wrong intents fly under the radar. While this is a massive problem across most NLU platforms, some are starting to incorporate some level of potential clash identification.
3. Low confidence in NLU fixes, and they take a long time/a lot of effort to get right.
As the chatbot’s capability broadens it becomes increasingly difficult to fix potential intent clashes. Teams feel like playing whack-a-mole, fixing one intent and another breaking another intent elsewhere.
4. Too busy fixing instead of adding new value to the chatbot.
The backlog of useful functions continues to grow as you observe your customers trying new things. But new functions don’t take weeks or months, not days to release leaving customers with the dreaded “Sorry, I don’t understand. Can you please rephrase”.
3 Tips to Ensure your Chatbot Meets Customer Expectations
You want your chatbot to be effective 100% of the time, or as close as possible. The bot needs to deliver a good experience to customers, so you can grow your business. How can you ensure your chatbot delivers? Here are three tips you may find useful:
#1 Match Conversational Design to the NLU Capabilities
When the customer is unaware of the chatbot’s purpose, capabilities or limitations, it can lead to frustration and abandonment of the interaction.
Adjust the conversational design to the chatbot NLU capabilities:
- Add simple changes in the conversation, for instance, in the greeting message. You can add a greeting message that states the limitations of the bot.
- Provide a feature discovery experience for ‘new’ or infrequent customers.
- If your chatbot is immature/still learning, or you don’t have strong NLU skills on board, avoid open ended questions like “what can i help you with”. Stick with safety bumpers and use a directed dialog.
#2 Use the Right NLU Tools Built on Experience.
The right NLU tool, tailored for your company, can make a vast difference in the way your chatbot understands customers. You should look for a tool that:
- Identifies intent clashes as quickly as possible
- Helps the team easily move training phrases from one Intent to another Intent as you grow the Chatbot’s capability
- Extracts all the relevant slots of information while giving you control on how the Chatbot handles any follow-up questions.
#3 Invest in Training your Team and Upskill your Resources
If you’ve hired people with 5+ years of production experience in conversational design and NLU engineering you’re probably in good hands.
But if your team is relatively new, invest in your team for the long term and success of the chatbot.
- Have your Conversational Designers learn from your NLU engineers so they can create better designs;
- Have your NLU engineers learn from your Conversational Designers so they can create more robust NLU models;
- Take online training courses.
- Seek mentorship from seasoned experts.
The Bottom Line
Chatbots can improve your customer’s satisfaction, increasing your revenue and ultimately, the ROI. But only when done right. Ensuring the chatbot understands and answers promptly and correctly users’ queries are critical. Hopefully, the tips mentioned in this article can help you improve your chatbot with NLU.