Final Product
AI Generated Results
Tools Integration
One-click data logging
Flow AI is a company that is dedicated to develop tools that maximize sales-reps daily activity. Turbo is their lead source tool that use AI to present qualified accounts and help speeds up the customer acquisition process.
As a UI/UX Design Intern at Flow AI, I led the design of the AI feature and the tool integration feature for Turbo. In the course of three months, I work alongside a full-stack team, an AI team and 2 other designers.
Before launching, we sent out surveys and invited users to test out Turbo, receiving a SUS score of 72.5.
In today's hyper-competitive business environment, sales representatives face an increasingly daunting task of identifying and converting high-quality leads into customers.
This was quite a challenge for me since I have no prior knowledge to the sales industry.
I decided to do a competitor analysis to understand the sale stack tools and the customer acquisition process.
Then I interviewed our stakeholder, who is also a sales-representative. He provided me with a more holistic view of what a sale-reps day-to-day look like.
I map out a simple user journey to help myself visualize the pain points during the customer acquisition process.
These ideas are valuable as they were derived from user pain points and could help is design for success.
My original thought was to design this like as a conversational AI. I researched on different ways of interacting with a language model, including using a quality meter to suggest improvements and a checklist that updates in real time while inputting.
Show prompt checklist, update the checklist in real time during inputting
Input a prompt for AI, we provide examples to guide users to add in keywords
Prompt quality meter and provide suggestions for improvement
After talking to the AI prompt engineer we agreed that we need to do a testing to see if the conversation of AI can generate results that meet users' expectation.
In order to validate this idea during an early stage, and also ensure this would work on the engineer side, we conducted a testing where we ask 2 users to ask AI the same thing.
Then I quickly realized that people’s conversational style are extremely different. When I ask them to input a sentence for a same type of target, they constructed the sentence very differently, which eventually led to inconsistent results.
I'm trying to find companies in New York that have around 1000 employees. Specifically, I wanna get in touch with folks who work in both the IT department and high-up positions.
I am searching for businesses situated in New York that have approximately 1000 employees. I am interested in contacting individuals who hold positions in both the IT department and upper management.
Users are less likely to provide accurate descriptors when they see this layout.
And in order for our model to work, user would need to input parameters and descriptors that our model recognizes. I collaborated with our AI engineer to consolidate a list. Here are some examples of the parameters:
After clarifying how our AI model works, and also considering the complex nature of account prospecting, I sequenced the long form and ask users to complete only 1 step on each page.this iteration is extremely crucial to the product development because we were able to identify the constraints that our AI model and look for solutions that could meet users' needs and feasible on the technical side.
Based on feedbacks we got from the survey and the product matrix I created before, I have several improvements in mind including improve onboarding experience, expand our tool selections, set goals and reminders for users...etc.
I tried so hard in the beginning to understand problems and the challenges, and also a lot of ambiguity along the way. I learned how to effectively communicate in these circumstances and never stop asking questions!
