top of page

CEMEX Go Chatbot

2019 - 2020

CEMEX Go is a digital platform designed for order placement, real-time order tracking, and efficient management of invoices and payments for CEMEX's primary products. Recognizing the need to enhance user experience, boost customer engagement, and reduce customer service costs, the concept of integrating a chatbot was introduced.

I played a vital role in guiding the design and research aspects of the project, which influenced the direction of the CEMEX chatbot's development. Throughout this journey, I worked on overcoming challenges related to how effectively the chatbot communicates, how it looks visually, and how users interact with it.

chatbot-header.png

User Research

A third party initially developed the first version of the chatbot, which provided an opportunity for me to conduct user interviews and beta testing to further enhance it.

The study aimed to assess several aspects:

  1. The chatbot's ability to answer common questions about CEMEX Go services.

  2. The chatbot's capability to understand users' everyday language and provide suitable responses.

  3. Whether interacting with the chatbot helps users efficiently resolve their queries.

  4. How users engage with the chatbot and how to improve its search functionality.

  5. Identification of general pain points.

For this study, we recruited six service agents who have regular interactions with end users and possess knowledge of the CEMEX Go platform. We employed the following methods:

  • Voice and screen recording with participants using the "thinking aloud" method.

  • Semi-informal interviews conducted both before and after participants interacted with the chatbot.

  • A Likert scale to quantitatively measure user satisfaction.

During interview with agent

Findings

  • The system's responses were often laden with technical jargon, making it challenging for most users to understand.

  • Users frequently had to rephrase their questions multiple times because the chatbot struggled to provide precise responses. This was largely due to the chatbot having a limited list of specific keywords and phrases for processing input questions.

  • Many users posed relevant questions, but the chatbot couldn't provide accurate answers. For instance, users inquired about contact information or the option to speak with a real person.

  • The chatbot tended to offer the same response to different types of questions. Consequently, users frequently had to rephrase their queries to obtain the correct information.

  • While some users asked direct questions like "get balance," others used longer, more conversational sentences, such as "good afternoon, can you tell me my balance please?"

Early sketches of the Chatbot

Recommendations

 

Here are some key improvements the team should consider:

  1. Understanding User Expression: The team should analyze how users express themselves when interacting with the bot. This can help improve the chatbot's ability to comprehend a wider range of user inputs.

  2. Enhanced Language and Responses: Enhance the chatbot's responses by expanding its vocabulary and offering varied responses to make interactions more engaging and informative for users.

  3. Incorporate Links for Navigation: Consider adding links that can redirect users to relevant sections or resources, helping them find what they're looking for more efficiently.

  4. Reduce Text, Increase Button Options: To streamline interactions, reduce the amount of text presented to users and incorporate more buttons that allow users to select options easily.

  5. Diverse Responses for the Same Question: Ensure that the chatbot provides diverse responses for the same question, offering a more dynamic and informative user experience.

  6. Clear Understanding Indication: The chatbot should clearly communicate what it understood from the user's input, particularly in cases where the user's question is complex or challenging to process.

  7. Suggest Related Questions: Implement a feature that suggests related questions or possible next steps to users, reducing the likelihood of misunderstandings and facilitating smoother interactions.

Screen flow sketches

Desk Research

I conducted desk research to explore best practices and benchmarking for chatbots, aiming to inform the final deliverable. The findings were centered around key features that contribute to an exceptional conversational experience for users. For instance, it is considered a best practice in chatbot design to infuse the bot with a distinct personality.

Some noteworthy examples of these practices include:

  1. Welcome Message: Starting interactions with a friendly and informative welcome message sets a positive tone for the conversation.

  2. Use of Buttons: Incorporating buttons within the chat interface can simplify user interactions and guide them through options.

  3. Suggestions: Providing users with helpful suggestions or related questions can assist them in finding the information they need more efficiently.

  4. Talk to a Person Option: Offering users the option to escalate to a human agent when needed adds a layer of personalization and support.

  5. Avoid Repetitiveness: Preventing the bot from repeating the same responses enhances user engagement and minimizes frustration.

  6. Persistent Menu: Including a persistent menu that offers easy access to core features or actions maintains user orientation within the conversation.

These practices are essential in creating an effective and user-friendly chatbot experience.

"I can see this isn't going so well.

Would you like to talk to a real person?" - IBM

Styleguide

 

In my role, I focused on enhancing the chatbot's user interface and expanding the CEMEX design language. This involved incorporating new elements into the design library, including features like the loader, chat window, and dialogue boxes. By doing so, I contributed to the overall visual appeal and functionality of the chatbot interface.

Conclusion

The results obtained from the beta test played a significant role in shaping the first release of the chatbot. Users expressed satisfaction with the introduction of the chatbot, acknowledging its potential to streamline their workload.

However, it's important to acknowledge that perfecting a chatbot is an ongoing process that demands careful planning and continual improvement. While a chatbot can serve as a valuable customer engagement tool and self-service resource, there remains room for enhancement. The chatbot should aim to better understand the diverse ways users may interact with it, recognize various terms, and expand its general knowledge to offer improved assistance. To achieve this, it's crucial to adhere to established best practices for chatbot design, which can help prevent common pitfalls and ensure a positive user experience.

Moving forward, it's essential to continue measuring the adoption of the chatbot and collecting user feedback to inform the next design iteration. This iterative approach will contribute to the continued enhancement of the chatbot's performance and user experience.

  • By conducting UI/UX design in-house, we achieved substantial cost savings, amounting to approximately 1 million MXN pesos.

  • Through testing, we identified the necessity for a UX writer, underlining the importance of well-crafted user experiences.

  • The findings gleaned from testing significantly informed the successful development of the chatbot's initial release.

  • The Olivia Chatbot engages in approximately 1,250 conversations each month, a noteworthy achievement considering the 17,000 visits to the chatbot window (As of 2020)

bottom of page