Chatbots can be effective and cost effective mechanism to communicate with potential or actual customers in real time. Customers are more and more expecting immediate responses and if you are unable to provide this they quickly move to competitors who are able to.
Additionally chatbots can provide valuable insight in to what visitors and customers are looking for. Unlike static web pages where perhaps less than 1% of visitors actually leave a comment by having the visitor actually clearly ask a question to a chatbot you are able to capture information on what your visitors are actually looking for giving you insights in to possible changes in your go-to-market strategy. By also capturing additional information from visitors you are also building out an improved list of leads reachable through the chatbots, web notifications or other mechanisms.
From a customer service perspective many of the common queries and questions can be automated resulting in convenient mechanisms for customers faster response times resulting in happy customers or more leads while leaving the staff more capacity to deal with real exceptions and unique problems. There are many factors which needs to be considered for making a chatbot successful. Here is a list of what we believe are the top 10 strategies creating a successful chatbot.
1. Understand and guide your customer
A successful chatbot is not about technology but about good conversational design. Put yourself in the shoes of the different types of visitors and design the flow to guide them as simply and quickly to the answer. Focus on the major scenarios and design the flow according to that and avoid considering all exceptions and one-off scenarios thereby making the options simpler and clear answers.
2. Pre-train your AI/NLP to extent possible and continue to train the AI engines
When you initially create the chatbot the AI will often have limited training data to leverage for NLP training. Spend the effort up front to train the AI engine with as many customer phrases as possible to point to the correct intent. When designing the answers it is also important to design the intents in your NLP to be as generic and simple as possible which should make sense to as many possible questions as possible so the NLP will not need to differentiate minor differences in incoming phrases.
If you have historical data from live chat, Facebook, email or other sources try to extract as many of real life examples and carefully evaluate those to point to the correct intent. Once the chatbot is launched the new phrases which are processed will need to be validated or adjusted to quickly improve the NLP accuracy.
3. Fallback procedures
It is inevitable that the chatbot will fail to understand some questions from customers. Either because the AI is not yet fully trained or questions for which the chatbot is not designed to answer. This may be a smaller percentage of customer queries but just as important to properly deal with. Most chatbot platforms will have a simple fallback intent but which often times is too simplistic to properly handle the sessions which are getting stuck. Many poor performing chatbots have not carefully planned this out and those failed sessions are those which then customers will loudly complain about giving chatbot process a bad name.
Escalated fallback – if the fallback has been triggered multiple times in a row or if there is an extended period of silence from the customer after fallback intent, it is better to trigger an escalated intent which then takes the session to handle the confusion.
Live agent – if you also have Live Chat enabled these sessions should then be initiated to transfer to a live agent better able to resolve the query
Offline follow-up via email or phone - capture email address or phone number and have a representative review the chat session with resulting offline follow-up using email or phone.
4. Immediately declare what the chatbot can do
By declaring up front in terms of what the chatbot can help with you avoid having visitors wasting time asking questions which the chatbot is unable to help resolve. This is a simple and effective mechanism to avoid fallback failures and irritating customers.
5. Capture relevant data
The chatbot can capture information from the customer dependent on the specific scenario. For example if the intent triggered is order status the chatbot should capture order number and contact information. If the chatbot is integrated or has an automated mechanism to retrieve the status it can be provided back to the visitor. If not the information are then captured for the customer service to be able to respond back to the customer with the information requested.
6. Let customers know that the chatbot exists
Customers need to be able to find the chatbot to interact with and there are many strategies to be evaluated and will depend on the mechanisms you currently use to interact with the target audience. If the potential volume is large it may be better to do a slow and controlled release so you have time to stabilize and validate the AI and refine the conversational flow before releasing to a large audience.
Refer the chatbot in signature of confirmation emails or any other email sent to customers Marketing campaigns through Facebook or other social media accounts Place web chat widget in all relevant pages of website Announce the existence of the Chatbot through newsletters or website news release
7. Analytics and continued improvement
Customers can ask all sorts of questions and you will never really know until you are in production mode. Once you start seeing the results quickly work on improving the AI and modifying or adding intents/answers based on what you are seeing. If you see irrelevant question being asked consider modifying the introduction messages to better clarify what you are able to do. By having good chatbot and AI analytics while transaction volume increases you will quickly start to spot improvement opportunities. Do not assume that the chatbot project is finished at the time of launch but rather continue to release improvements to the chatbot on a weekly basis.
8. Request feedback
In addition to the information you will get from analyzing sessions you should simply ask them for feedback and suggestions for improvements. By reviewing negative feedback sessions improvement opportunities can be identified. Keep it simple such as a simple thumbs up/down likely to result in higher response rates. If feedback is negative then ask the user on how we can improve. This can provide gems of suggestions and ideas which to continue improving the chatbot.
9. Good conversational design
Conversational design is more than copying your FAQ answers to the chatbot and needs to follow good principles for language, quick reply designs and overall design of the conversational flow. Key elements of good conversational designs include:
- Simply and clear natural language
- Personality representing the organization
- Be helpful and friendly
- Draw out the expected common flows
- Avoid dead-ends
10. Give the chatbot a name and a face
It should be clear to the user that the chatbot is a robot and should not be tricked in to believing it is a human. To make it more memorable the chatbot should have a name and a face. The name should be something short, simple and easy to remember and avoid names like “Order Status Bot” but instead use names such as Lisa, the SomeCompany Customer Service Bot. The face does not need to be a human face but still something memorable. When promoting the chatbot the face and name should be included.
What we do?
Symprio helps customers create engaging and cost effective chatbots to automate customer service, support and more
By understanding end user behaviors and needs we design a conversatonal flow to engage end users and answer their questions.
Bot Platforms & AI EnginesWe can design and implement your chatbot on a variety of platforms and AI/NLP solutions coupled with custom intergrations dependent on your needs.
Let us know your chatbot requirements here
MALAYSIA | SINGAPORE | USA