67% of global consumers have used a chatbot for customer support in the last year. It's a staggering number that continues to rise. For many people their first impression of your business may come from a chatbot. In fact, it may be the primary or only way they interact with your business. However, many consumers are feeling frustrated with your chatbots.
Did you know half of consumers indicated they would become repeat buyers after a personalized experience? This is why getting personalization right is so important.
Yet, a recent Cyara Botium Customer Survey found that 71% of consumers feel that the conversations they are having with chatbots are impersonal and this has left them feeling frustrated with the service provided by chatbots.
Eclipse expects this problem to grow as many of the solutions being deployed today depend on the same or similar Large Language Models (LLM) that will produce a similar tone and grammatical structure across the brands that deploy them out of the box.
This means that customers aren't seeing differentiation in the way your chatbot answers their questions between you and your competitors even if the actual response is different. All hope is not lost though as prompt engineering is allowing brands to create highly personalized and brand conscious responses to consumers interacting with their chatbots.
What is prompt engineering?
Simply put prompt engineering is the practice of refining the results provided by the LLMs powering many chatbots. It allows you to augment the information provided by the customer to provide more personal and relevant answers. There are a myriad of ways and techniques to do this. We aren't going to cover them all here and this blog only provides high level guidance.
It's important to understand prompt engineering can be done with automation. Every sample prompt provided below would be accomplished by having your chatbot intercept the customers prompt and improve upon it prior to sending it to the LLM powering your chatbot.
We also aren't suggesting CX leaders need to become prompt engineers or hire prompt engineers, but we believe they should understand the concepts to help them evaluate and improve the tools they are using.
What are some ways I can use prompt engineering to create more personalized chatbots?
Teach your chatbot your brand voice and give it a role to play. Did you know you can provide instructions to LLMs to write in a specific tone or voice? This is one of the most effective ways to ensure your chatbot stays on brand. Through prompt engineering you can tell it a specific tone to use such as conversational or formal, style (e.g. playful, friendly, etc.), and even adopt a particular persona. Here is an example of a prompt that will generate an on-brand response to a customer issue: "Act as a customer service representative for a technology brand known for its innovative culture and high-touch service. Compose a response to a customer who is expressing disappointment with their newly purchased laptop being slow. The language should be conversational and empathetic. Offer a tangible resolution, such as an appointment with a technician at one of our stores." You can even take it a step further by providing desired responses that the LLM can learn from to create a more on-brand experience.
Provide your chatbot with demographic data about the customer. You can further enhance your responses by providing your chatbot with information about your customer to have it further alter the way it answers inquiries. By including information about their age, generation (e.g. boomers, Gen Z, millennials, etc.), and gender LLMs can craft responses in line with the expectations of each customer as customer expectations in a "human like" interaction are largely informed by the culture they have consumed throughout their lives.
Set the length of the response It's very easy to set the length of the response your LLM and chatbot provides. Do you want something short and sweet or a more verbose and detailed answer? You can achieve both through prompt engineering. Better yet vary the length of your response based on what the customer is trying to accomplish. By understanding the nature of their problem, you can decide if a short answer will do the trick or if a more thorough explanation is needed.
Turn up or down the creativity You can directly influence how creative your chatbot gets in the answers it provides in a very straightforward way. Are their certain interactions that require the chatbot to provide near verbatim answers to customer's inquiries? No worries, you can ensure your chatbot doesn't get creative and only provide responses it has seen before. Are their times where you chatbot needs to be a bit more "creative" to properly provide an answer by synthesizing information across a variety of responses and data it is has been trained on? You can easily allow your chatbot a bit more latitude in its responses as well.
Provide your chatbot with data about past interactions. Want to take your prompt engineering game up a notch? This is how you can take personalization to a whole other level. Let's assume you are clothing brand that offers exclusive collections that often sellout and you have a customer inquiring about the availability of your latest collection. The customer starts their chat with a simple request: "When are you releasing your spring line?". Without prompt engineering our bot would respond with a very factual and impersonal response: "Our spring line will be released in early March." When we use past customer interaction data, we can start to provide a highly personalized interaction. Here is what we could get by combining some of the techniques above with some additional prompt engineering techniques. First, we have to gather important data about our customer. Here is what we know about our customer from past interactions. Julie is a frequent buyer of our product who always buys items from our newest collections anytime she can get them before they sell out (which means she is considered a VIP who we give priority access to our newest collections). We also know Julie is 22 years old and is from Gen Z. Historically, Julie has preferred dresses and embraces a more feminine aesthetic. She prefers this to a more casual streetwear style. Now we have to consider our brand voice. We are a fast fashion brand that strives to be seen as chic while embracing modern trends. Our brand voice is conversational and sometimes playful. We need to ensure our response to Julie reflects our brand voice as well. Here is our improved response to the question: "When are you releasing your spring line?". "Julie, we know a fashionista like you can't wait to get your hands on our latest collection. Since you are a VIP, we are going to put you at the front of the line where you belong. Sis, because we love you, we can give early access to the collection in late April. Do you want me to put you on the early access list?" Julie responds with "For sure, put me on the list" Our chatbot can then continue the conversation using even more past interaction data. "I just added your name to the list. You will get an email with an early access code soon and your exclusive VIP link. Julie, we know you are the farthest thing from a normie and love to stand out from the crowd. Here is a link with some things from our spring collection that fit your #cottagecore style. Is there anything else I can help you up with?" Of course, our response would be radically different if our buyer was a 40-year-old mother who is shopping for clothes for her kid's birthday but this is just one example of a highly personalized chatbot that knows our customer and reflects her expectations of a conversation with a coveted fast fashion brand that largely sells to Gen Z.
In today's highly competitive marketplace personalization is leading to differentiation for brands. It's important to stay ahead of competitors by offering the kinds of experiences your most discerning and frequent customers expect.
If you aren't providing highly personalized interactions across all channels, you risk losing those customers to those companies that strive to create experiences that delight in their highly personalized nature.
Ready to talk to Eclipse about finding the tools that can help create highly personalized customer experiences or how to better leverage the tools you have?
Contact us here to start a conversation about personalization.