Guest post by Erica Sunarjo, writer and blogger.
While still in their toddler years, chatbots are becoming commonplace. Why? Because one of the driving forces of marketing and CRM today is to have more conversations with customers. Conversation can create a closer “relationship” between a customer and a brand, even if it is only with a bot.
And customers are engaging with chatbots with much greater comfort. In the financial services industry alone, for example, according to a Capgemini report, only 28% of consumers were comfortable interacting with a bot in 2017. That number rose to 50% in 2019.
In addition to developing an enhanced conversation with customers, there are other clear benefits of this technology:
- Time-Saving: Chatbots can give a quick answer to most frequently asked questions. Customers do not have to hang on a phone line or wait for an email response.
- Money Savings: Chatbots are cheaper than hiring more staff as a business grows. While they may be a bit pricey, depending on the type of chatbot a business develops, there are no other “staff” costs of benefits, vacations, etc. The only ongoing expense will be modifying and improving, especially as the technology continues to evolve.
- 24-Hour Customer Access: Chatbots do not keep regular work hours. Customers can access them around the clock and from any time zone. And if a customer gets the answer he needs, he may just proceed with a purchase.
- Consistency: Human CRM agents can give conflicting messages – bots do not. And bots never get frustrated or angry.
- Adding Some Humor: This helps to promote relationships with customers if done well.
Now About Those Chatbot Scripts
Once the decision has been made to create one or more chatbots for your business, you will have to develop chatbot scripts. And this will involve several decisions and tasks.
So, let’s go through the step-by-step process of creating your chatbot scripts.
1. Set Your Goals:
What do you want your bot to do? This can be as simple as just answering the most frequently asked questions that have standard, programmable answers. Or you may want a bot that incorporates AI and NLP so that it learns as it goes and becomes adept in answering increasingly complex questions or providing more sophisticated responses and even humor. These give the user a more personalized experience.
A famous and almost viral example of an early chatbot was Poncho, the weather bot. Of course, he was programmed to provide weather forecasts for any place in the world. But his scripts went far beyond that. Horoscopes were added, for example. And because AI and NLP were imbedded, he did learn more from conversations. Humor became a compelling reason for consumers to access him. Poncho’s developers continued to add additional scripts as its audience grew.
2. Look at Other Chatbot Successes
By studying other chatbots either in or related to your business niche, you will get a clearer picture of the scripts that are in use and that engage customers. These can give you some general ideas about what your script should “look” like.
But one thing you will come to understand is this: If the business has the same audience as yours, you will come to see how that same audience can be engaged through style, tone, and vocabulary. Think about the difference between a chatbot that has been developed for Red Bull vs. one that has been developed for Cartier Jewelers.
3. List the Topic Fragments You Want to Include
Here, you will want to make a list of all of the topics that your bot should cover. This may include the FAQ’s that your “history” has shown to be the most asked. But what else might you want to include? Do you want your bot to make suggestions to a customer for additional products or upgrades based on what they have ordered or are looking at?
Topic fragments may also include greetings, sign-offs, humorous injections, etc. Again, you can take some lead from chatbots in yours and similar niches.
Once you have listed the topic fragments, add detail that you will want to be included in that topic.
4. Combine the Topic Fragments into Threads – Dialogue Trees
Once you begin to string your fragments into threads, you will see that there are multiple statements/responses that could come from those threads.
Tacobot as an Example
Let’s take an example from Tacobot, Taco Bell’s chatbot. If a customer places an order for a few tacos, then the bot has a couple of options. It can total up that order and give the customer his final cost. Or, as the bot has also been programmed to do, it can ask about any sauces the customer may want.
Going even further, it can suggest some additional items that the customer may want to consider. Then, based upon the customer’s response, the bot can confirm the original order, add the other items the customer has selected, and then provide a final price. As you can see, this will look like a tree with branches.
Greetings as an Example
In terms of greetings, there will also be different responses, based upon what the customer says.
For example, if your bot says, “Hey there. How are you doing today?”, a variety of responses may be given.
The customer may just say “fine,” in which case the bot will simply move on to ask how it can help today.
If the customer is having a bad day, though, he may say so, and the bot should be ready to give a different response – “Sorry to hear that. Maybe I can make your day a little better. What can I help you with?”
And, if you want to add some humor, your bot can say, “I get it. I have a terrible headache today and maybe my medicine will kick in soon. In the meantime, what can I help you with?”
You Don’t Have to Reinvent the Wheel
Creating dialogue trees sounds pretty complicated, and it certainly can be. Fortunately, there are a number of tools that have worked out the methods to create these trees, allowing you to pretty much fill in the blanks to craft those that fit your needs.
You can begin by checking out Twine, Chat Mapper, Xmind, or Inklewriter. If you and your team are feeling energized and creative, though, you can even develop your own trees and branches.
One thing to remember here: Your first finished “product” will not be perfect. And that’s okay. You will learn as you go and make the changes that you see as necessary. This will require continuous monitoring and evaluation of user satisfaction.
5. Use the History of Your Own Customers
As you actually develop dialogue threads, go back and let your customers/users drive those conversation threads. And this is not limited to just frequently asked questions.
What language style do they use when they ask questions? What slang terms do they use? And what synonyms might be used? A bot can be “taught” all of this and programmed to respond to all of that terminology.
And, as you continue to move further into this whole chatbot technology, deploying AI algorithms that can learn as they go will probably be in order. The bot can teach itself.
And don’t forget this: users will often use abbreviations and misspell words. The most common ones should be programmed in so that your bot recognizes them.
Want to learn more about chatbots? Download our free e-book!
6. Bot Personality
You know your brand, and you know your audience. Give your bot a personality that matches both when you write the actual dialogue of the script.
And be prepared for users to get “off-task,” all in the name of fun. They may ask your bot questions that are totally off-topic, and it is the bot’s job to bring them back to the topic in a nice, and maybe even humorous way. A user of Tacobot, for example, may try to order a medium pepperoni pizza just to see the response. Plan for this in advance, by having a few humorous responses that then redirect the customer. “You know what? Our shipment of pizza dough and pepperoni didn’t come in today, so how about a taco, a great nacho plate, or any of our amazing burritos?”
Consider giving your bot a name, a gender, an age, and some other human traits like hobbies or interests. All of these things contribute to a better and more engaging user experience.
7. Keep the Dialogue as Brief as Possible
Once you have your initial completed chatbot script, go through it and dump any unnecessary words. The best responses are probably between 60-90 characters.
Steve Jenkins, a chatbot monitor and analyst for Studicus, puts it this way: “When we first developed our chatbot scripts, we were way too wordy. And users were bouncing from the experience. Once we shortened the responses and added just a bit of humor, users stayed in the interactions, conversions picked up. Lesson learned.”
Lose the Need for Perfection
This is the final word. Learning how to write chatbot scripts is an ever-evolving process. You cannot expect it to be perfect in the beginning, or even ever. What you can do is launch them, monitor success, find out what is not working, make the changes that you think will work, and commit to the understanding that the improvement process will be never-ending.