— Problem Statement

Supermarket customers frequently faced delays and inconveniences when seeking support and personalised recommendations. The customer service team struggled with repetitive inquiries and inefficient order management, which led to longer wait times and decreased customer satisfaction. The goal was to address these issues by implementing an advanced chatbot solution.

— Kick Off

During the project’s initial requirements gathering, I facilitated Product Vision and North Star workshops to align the team and stakeholders on core user needs and business goals. These sessions were crucial in establishing clear success metrics and defining product direction before diving into development, ultimately setting roles and responsibilities.

— Discovery Phase

I delved deeply into the landscape throughout the discovery phase to identify valuable opportunities. By analysing website analytics, I pinpointed high-traffic areas and unveiled pain points, which allowed me to prioritise intents effectively. Leveraging Medalia data, I uncovered common customer pain points, inquiries, and utterance terminology, which formed the foundation for establishing key metrics for post-release success.

SEO tools proved instrumental in identifying common terminology and inquiries related to the brand, affirming intents, and refining utterances. Moreover, demographic analysis sheds light on behaviour-specific intents.

Furthermore, I meticulously examined vital features and opportunities from leading supermarkets and indirect customer support globally to ensure that our chatbot would truly stand out. This entailed scrutinising existing customer service workflows to pinpoint processes that could be streamlined and integrated with human agents.

— Ideation and Prioritisation

Once we understood the landscape clearly, it was time to regroup, brainstorm, and prioritise. I held workshops with the development and product management teams to create situational personas and conduct affiliate mapping to brainstorm an extensive list of core intents. Using value vs. effort analysis, we identified and prioritised user needs and assessed the viability of different topics.

We aligned with stakeholders on key priorities and the MVP release scope through MoSCoW method workshops. We also documented potential sub-goals and future state wants for Phase 2 post-initial release.

— Conversational Design

With a clear plan, the design phase brought the chatbot to life. I created detailed flowcharts to map out conversation flows for each scenario, visualising the user’s journey and key interaction points.

Working with a copywriter, we developed scripts based on user flow mapping and utterance data reports, ensuring natural and effective conversations.

Using Dialogflow, we created quick prototypes and tested scenarios with target demographics. Multiple testing phases, starting with small groups and scaling up, identified gaps in flow and utterances, allowing for continuous refinement.

— Testing

We conducted usability tests with our target demographics to find gaps and areas for improvement. Working with the copywriter, we tested and refined the conversation flows and interaction scripts, incorporating user feedback to improve the chatbot’s functionality and user experience. This iterative process was crucial for developing a solution that effectively met user needs.

Additionally, we added a thumbs up/down prompt at the end of conversations during the beta release, and it’s still in use in the live release for ongoing data analysis and continuous improvements to the data model.

— UX / UI Design

Keeping components simple and aligned with common e-commerce chatbot experiences ensured familiarity and ease of use. Suggestion chips guided users and improved intent recognition. Maintaining brand presence with a simple UI design was crucial for a cohesive user experience.

— Outcome

The Chatbot now handles thousands of self-service conversations weekly, addressing common queries like refunds, order tracking, and store hours. This significantly reduced agent handling and call wait times, resulting in a two-point increase in Net Promoter Score (NPS) from the first week of launch.

The project underscored the importance of continuous user testing and feedback in refining AI-driven products, balancing technical capabilities with user needs to deliver an effective solution.