A conversation designer creates the flow and the structure of the interactions between users and AI systems, like chatbots, to ensure smooth, natural communication. They focus on making the way the bot answers, manage the entire flow of dialog to provide the best experience to the user in a way that reflects the brand’s tone and personality.
I feel this a role is still largely misunderstood, but I have seen first hand how valuable it is for any conversational AI project. A proper carved conversation has the ability to transform the entire user experience.
There are a number of challenges that conversation designers face and need to learn how to handle. Here are five of the most common ones.
1. Too much information
When users provide excessive information, it can make it difficult to isolate the key issue. A well-trained conversational designer ensures that the bot is able to break down these complex inputs and guide users toward relevant responses without losing track of the core problem. This requires a balance of natural language understanding (NLU) and strategic question framing, which can clarify the user’s intent. Designers are essential here as they create a structure that simplifies this process for the bot, preventing it from being derailed by too much information.
It’s pretty natural in human conversation to provide a long description of a situation without initially being clear about what the primary problem is. A bot needs to do its best to parse through this information to come up with an appropriate response. But trying to pick out one thing might not be the best approach.
In the below example, the bot tries to deal with the same response by providing some options to help move the conversation along in the right direction.
2. Ambiguity in user message
Human language is ambiguous, and users often provide imprecise statements. A conversational designer should anticipate these ambiguities and create mechanisms to deal with them, such as offering multiple choices, asking clarifying questions, or rephrasing prompts. These strategies are crucial to avoid frustrating the user and ensure smooth interaction. The designer’s expertise allows the bot to handle vague or unclear input while gently guiding the conversation in the right direction.
Above is an example where it’s not clear if the expense is for two or three people. The bot sees the number two and responds to that and is probably incorrect.
Below is an improvement on the previous response, which guides the user to properly account for everybody covered in the expense.
3. The user expects the meaning to be inferred
This refers to a common expectation in human-computer interactions where users assume that chatbots will understand implicit information or context without explicitly stating every detail. This is natural in human conversations—people often rely on shared context, implied meanings, or shortcuts, expecting the other person to fill in the gaps.
In this example, the user has implied that they indeed are ready to confirm that order by saying that they’ll pay with Mastercard. It is easy for a human to infer that as an acknowledgment of the order confirmation, so the conversation designer should be aware of this type of possibilities.
4. Contextual awareness
In a conversation, a user may provide information that is relevant to other parts of the dialogue. However, the bot may fail to make the connection between earlier and later inputs, missing the link between related details and responding in a way that overlooks important context.
Below, the bot doesn’t make the connection that the first question relates to the second question, that is, that they want to pick up their meds in Daly City.
In the below example, the bot is successful in associating the context of the first user message with the second.
5. Ensuring the persona shines through
Ensuring the persona shines through is crucial, as it helps to establish the tone and voice of the chatbot, providing consistency and familiarity for the user. Nurturing and encouraging the persona throughout theconversation can help to build rapport and trust.
In this example, the bot responds in an inflexible way that demands that the user respond in a precise manner to a question that they might not clearly understand. And here’s a way to handle the same situation in a more forthcoming manner that takes into account that, in real life, it might not be so clear where to find the order number.
Below the same interaction, but now re-worked to include all of these best practices.
In summary, there are a number of common challenges in conversation design that you should learn to handle. These challenges include long, ambiguous, and imprecise messages, handling contextual awareness, and designing the conversations so that they reflect the persona you envision for it.
Conversation Design in the world of Large Language Models
Even though LLMs function as a “black box,” conversational designers can still shape and guide interactions through strategic design techniques:
Prompt Engineering: Conversational designers can craft specific prompts or input structures that guide the LLM to generate desired responses, controlling tone, context, and persona. By iteratively testing and refining prompts, designers influence the model’s output despite the complexity.
Fine-Tuning and Guardrails: Designers collaborate with developers to fine-tune LLMs on domain-specific data, ensuring the model better understands industry nuances. They also set up rules or fallback mechanisms to handle inappropriate or incorrect responses, guiding the model’s behavior in special cases.
Conversation Flow and User Experience: By designing structured conversation paths and fallback scenarios, designers manage the flow, ensuring the LLM stays on track even when dealing with ambiguity or complex user inputs.
The control is less strict in these cases, but it can be (iteratively) improved over time.
References
Conversation Design Fundamentals
Leave a Reply