Project Context

Castorama asked us to design a new conversational assistant fully integrated with a ChatGPT module that we trained with a tool & products database to advise users on the best product that suits their needs. This project presented multiple challenges :

• How do we train the GPT module ?
• How to make sure that user don't ask things that aren't in the scope ?
• How do we push the GPT feedback to the user ?
• How do we design the interface ?

Project Objectives:

• Test & learn with ChatGPT prompt engineering
• Build a seamless e-commerce experience with ChatGPT integration
• Train the ChatGPT
• Test MVP prototype with users
• Design a nice front end interface

Starting points : Benchmark & playing with ChatGPT

Benchmarking and playing around with ChatGPT helped us understand the limits of the projects and define our area of intervention.

My Tasks

• Experiment tests with ChatGPT & benchmark industry example & practices
• Design userflows & identify touchpoints & pain points
• Design Wireflows & chat experience
• Test MVP prototype with a pool of users
• Implement insight and learnings from tests
• Train ChatGPT & prompt engineering with the tech team.
• Pair design with UI Designer for the final interface.
• Final AB testing.

Team Process :

• Research, testing and ideation with the tech team.
• Presentation of a MOOC & proof of concept to client.
• After client green light, we started developing an MVP for testing.
• After successful MVP, project was green lit for next stage : Full end to end design experience.
• Constant testing and iterations for GPT training.
• Published

🔍🔍🔍🔍🔍🔍🔍🔍 Close ups 🔍🔍🔍🔍🔍🔍🔍🔍🔍🔍 

🔍 Focus on : Products push

Customer needs at the heart of our reflection.

• We rôle played multiple context where customers would look for a specific product, in this case a drilling machine.
• This exercice allowed us to pin point the conversations starters and design shortcuts suggestions to help the user quick start his conversation.
• This also allowed us to orient the AI questions to refine user needs.
• In the end, the AI asks a minimum of 3 questions to pin point the best products that suit the users project.

🔍 Focus on : Interactive Prototype

Interactive prototype was part of our MOOC & proof of concept presentation. This allowed us to demonstrate to our client how we envision the product and how it will function. This facilitated the approval for the project and green light it for the next stages.

🔍 Focus on MVP user testing :

A pool of 8 users accepted to test our MVP product. Our objectives where to :

• Pin point the most frequent conversation starters
• Measure if users understand what is expected of them on the starter screen
• Measure if users understands that he's speaking to an AI assistant
• Measure if users understand the context & that he has to search for tools.
• Collect feedback on the overall experience.

These user tests allowed us to acquire interesting insights that allowed us to refine our product's copy & performance.

Screenshots of every tester's AI conversation for documentation.

Key learnings:

Test it & discover the project !