AI Assistant

Flint AI

Project Overview
I designed Flint, an AI assistant that streamlines workflows for the crews at Forge, a tech-driven home services company. My work includes shaping the strategic vision for Forge's AI-enabled products and aligning them with key business objectives.
Platforms
iOS
Android
Role
Lead Designer

Goals

Forge is a tech-enabled home renovations company that explored a variety of technology to be used to support their team. There are a wide variety of pain points that AI can potentially help solve for. I was tasked with helping discover these opportunities and designing the solutions to be built.

1

2

Research and discover problems that AI can have a high impact in solving.

Work with the team to design an experience that helps address the crews' pain points.

Flint, an AI-powered assistant will be able to answer common questions related to heat pump installation, as well as schedule and job related information that exists in Forge's database.

Project Complexities

Ambuiguity
I was given very loose objectives to start. I planned out my daily tasks based on assessing the project goals and status. I had to be proactive and make quick decisions to move the project forward.
AI Best Practices
AI is still an evolving field and I had to figure out how to design for it. I leveraged both available online resources, collaborating with the development team, and using the leading AI products for a first hand experience.
Vision Alignment
This project had a small team and I played the role of both designer and product manager. I worked closely with the stakeholder to weigh the pros and cons of varies concepts to finally land on a direction.

In a series of field research sessions, I observed and learned how our crews install heat pumps and troubleshoot issues.

Research

To ensure the AI experience resonated with our diverse workforce, I conducted user research with crew members of varying ages, experience levels, and genders.

Field Observations
I shadowed on different job sites to understand the crew's daily workflows, and observe any inefficiencies and pain points.
Focus Groups
In both focus group and 1x1 settings, I collected information about potential features and user preferences for the AI's presentation, including its personality, tone and visual representation.

Main Takeaways

Manual Usage
There are a handful of common things junior crew members use the installation manual for, such as the proper clearance around the heat pumps and how much refrigerant to add.
Troubleshooting
For complex issues, they always prefer to directly ask the crew lead. In absence of a senior team member, they will call their field supervisor to get instructions.
Error Codes
Dealing with an error code from the heat pump can drastically increase a job's completion time. The codes are often listed in a separate manual that sometimes is not included with the equipment.
Professionalism
There was a strong preference towards more direct and professional tone for this product. The crew cited that they take their jobs seriously and do not need the bot to "make jokes" or being overly friendly.

Iterations

SMS Frontend

The tech lead recommended looking into using SMS as the frontend so that we can focus on refining the AI's accuracy and speed of information retrieval. I explored ways the AI can work in a chat-only UI with the absence of any custom interactions.

Custom Chat UI

In contrast to the SMS version, I wanted to explore what functionalities are feasible in a custom chat interface to weigh the pros and cons of the approaches. I focused on surfacing up common actions at opportune times to make task completion quick and easy.

Dashboard

I want to paint Flint as a holistic assistant, not just a chat interface. That means as the product gets built, there could be more AI supported functionalities such as a dynamic dashboard that helps the user prep for their day.

Shortcuts

While chat-based UI seems to be the most common way of interacting with an AI, it is not the most efficient way for surfacing info the crews need. I narrowed down on the top information our crew needs and created shortcuts to start the queries without having to type to the AI.

Dark vs. Light

I explored different themes for this assistant. The light theme is more consistent with Forge's existing products. The dark theme is easier on the eyes in different environments.

Feedback Summary

Surface More Context
The crew mentioned that they want to know which equipment's information the AI is returning. They have multiple jobs a day, and worried any mix up can hinder the installation process.
Mental Model
They like the shortcuts but felt a bit confused moving from the menu sheet to the chat UI. It was not clear that the shortcuts was a way of interacting with the AI in the chat.
Dark UI
They like both versions aesthetically, but prefer the dark one because they might be in crawl spaces for certain jobs. The brightness of the light version might be annoying when used in a dark space.

Solution Highlights

Dashboard

As this product evolves, I envision Flint as an end-to-end part of the crew's workflow. For MVP, I wanted to start introducing Flint as part of the job prep process. The crew has access to see their day's scheduled jobs, their assigned team, and transit information.

Contextual Queries

Flint shows the current job and equipment the crew is working on at the top so the user can feel confident they are getting information about the correct job. The user can swipe to change the context to easily retrieve information about their other scheduled jobs.

Common Actions

The most common inquires are consolidated into a shortcuts panel. The interaction is connected directly with the chat UI to help users quickly be able to interact with Flint without having to type.

Transition States

I used motion to help the users contextualize moving between different states and screens of the assistant.

Since Flint may take some time to return information, I've also design loading states to inform the users of a potential wait time.

The Results

Test in Field
Both business stakeholders and the end users are excited for this product and eager to test it in the field.
Roadmap Creation
I worked with the tech leads to create Year 1 roadmap, highlighting the features and evolutions of Flint.
Development
The MVP designs are in development.

Notes & Thoughts

Flint is still in development. To measure the product's success, I would plan a series of 1x1 feedback sessions and focus groups to get qualitative feedback. For quantitive metrics, I would measure the number of times each shortcut is used, and number of times a user directly interacted with Flint. We can capture whether a physical manual was used on the job and cross reference that with Flint usage.

Project Artifacts

Visual Explorations

I explored a variety of representation for the AI Bot and gathered user preferences through focus groups and surveys.

Kathy Chau
LINKEDIN