Satigi Case Study

Summary

The following case study is about the initial stages of a startup I co-founded. Our primary aim was to use AI to help UX designers speed up their UX research synthesis process.

Timeline discussed in this case study: 4-6 weeks
Startup timeline: About a year

Background

UX Designers have to do many jobs at once: research, usability testing, UI design, content design and more
One of the jobs that consumes a lot of time, provides a lot of value, but is not immediately justifiable to the client is the research work. Specifically the synthesis of the research.

Problem Statement

How might we… help UX Designers speed up the process of research and synthesis?

End Results/Findings

Determining what to include in the MVP for an application can be challenging given the time constraints
Determining a price point can be difficult
The difference between the buyer and the user is crucial to discern
AI is and will be a productivity booster

Key questions to ask before starting out

Is your solution at least 10 times better than the current way your ICPs (ideal customer profile) are solving it? In time, money saved or other.
Why is now the right time to solve this problem? Technological, regulatory, behavioural shift?
Is your user the same as your customer?
If they are different, do you have access to the people that would actually pay? (Can you name 10 potential customers?)
Why or how would your first 10 customers tell other potential customers?

Research

Key points of the reseach phase

Target Audience

UX Designers, User Researchers, Design Agencies, product managers

Objective Key Results

• To understand how designers conduct user interviews and usability studies
• To understand their current frustrations in their job
• To understand how designers synthesise data from user research

Tools

Google Meet
Otter (for transcribing audio)

Recruiting

• What kind of criteria do people need to meet?
UX Designers, researchers, product managers

• Where and how do we find participants?
Friends, LinkedIn, ADPlist

Questions To Ask Participants

As a UX Designer, what would you say are the top 3 frustrations or issues you have with your job?
Could you walk me through the process of how you conduct user interviews and synthesise the data?
Could you walk me through how you create affinity diagrams and what tools you use?
What about user personas?
When you have synthesised the data, how do you come up with solutions for the problems that arise?
Do you ever turn to psychology to come up with solutions or to explain to stakeholders?
How do you often do this?
Have you ever used the software Dovetail?(heymarvin is also a competitor)
What were some of the manual work that you did that took a lot of time?

Research Results

Direct Quotes from UX Designers include:

“Synthesizing the data is really stressful”
“Going through the transcript is quite draining”
“Sourcing quality participants is quite tricky”
“Recruiting users is very manual and sourcing the best candidates can be quite tedious”
“Userzoom and Intellizoom are not very good. Participants game the system”

General observations from the research include:

Synthesis is tedious
Recruitment is tedious
Overwhelming amount of stakeholder management
Information discovery can be challenging especially against deadlines
For UI design, most designers use Figma. For UX Design, the tool used ranges across the board, from Miro, to usertesting to intellizoom to dovetail and many more

Prototypes

The following is a rough sketch of prototypes i made in order to guide my thinking for the designs to show the stakeholders

Homepage

The initial designed aimed at showing the features the product provides as well as the copy

Projects Page

Showing the number of projects a designer in an agency is working on

Dashboard View

Dashboard shows a view of ongoing research or completed research, it can be viewed by multiple parties including senior stakeholders

Affinity Diagram

Automatically generated affinity diagram based on the transcripts provided.

Startup Direction

How might we… redesign a website that allows hiring managers to unambiguously view, monitor or take action on all activities regarding their upcoming hires?

Creating a website that is simple but caters to a lot of variables takes time, lots of stakeholder sessions and lots of iterations.
The website was upgraded to the new version using the updated branding and was received as a much better version than the previous versions.

High Fidelity Designs- 1st Iteration

Figma- Workflow Page

We decided to use the colour blue as blue invokes:

Intelligence and Expertise
Dependability
Professionalism

The workflow page shows 3 main menu items- Projects, Team, Client as this is a typical segmentation for an agency’s workflow
In this flow we decided to start from the projects page as it helps to showcase the flow of the primary use case

Figma- Project Page

The projects page was designed to show a typical agency’s view. It showcases the different projects they may be working on.
Not too much emphasis was put on this page (e.g. should we segment it by clients or projects) as this was not our primary focus at the time

Figma- Dashboard

This page was our main selling point
The idea was to show all of the insights from the interviews conducted on that project in one view
We also wanted to make the page interactive, so you can click on some comments next to each theme and find out more
You can also click each theme and it will take you to the affinity map
Additionally you can ask questions about the data on the dashboard

Figma- Affinity Map

The affinity mapping was created to help users automatically sort themes so they don’t have to
The page also gives the user the ability to determine how many groupings they want. At the top, it defaults to “5” but a user can change this to their preference

Figma- Transcript example

The transcript page gives a transcript that is automatically annotated at the right lines for the user.
The transcript page gives a transcript that is automatically annotated at the right lines for the user.

Testing Results

Unmoderated Testing + Letter of Intent + Endorsement

General Introduction

In a race for time, we decided to test some of the designs before proceeding with development
As mentioned previously, the design direction was to target design agencies
We did some prototype testing on potential customers to get some general thoughts on the application and any feedback they had
We also used the testing as a way to guide our Proof of Concept app which would ultimately guide our MVP

Results

The interviewees were generally pleased with the initial prototypes we presented to them. A few tweaks to the journey was recommended but generally they were pleased
We were able to get 2 letter of intents from 2 agencies in 4 days which is important for investors as it signals that there is demand for the product
We were also able to get 3 reputable advisors that could help us on our journey

Updated Design

In the updated version, as seen in the video, we improved the flow of the initial mockups. Most notably:

The Interaction of how a user uploads a transcript/video/audio
The form fields on the details to fill in
The updated dashboard view, affinity map view and transcript view
How a user can access all the transcripts uploaded. Clicking on one transcript would take you into the transcript view

Proof of Concept

OpenAI API

GPT 3.5 Turbo

With the introduction of GPT 3.5 Turbo, I was able to connect the API from OpenAI to the application for our proof of concept
On the front end as soon as the user uploads the documents, I created a series of actions to be taken on the backend to facilitate the synthesis
Prompt engineering proved extremely useful in order to get the desired results.

Limitations

Token size was a big limitation when the API was first introduced.
In order to solve for this, i used the pinecone API to help split the document into different tokens
As time went on, OpenAI generally started to increase the token limit and the dependence on pinecone reduced
LLMs are probabilistic not deterministic, getting the same results every time is not certain but most of the time it was good

Bubble Low Code App Builder

Web Application Builder

Bubble is a popular web application builder that is perfect for building web applications for Proof of Concepts or MVPs
I had some familiarity with the tool which is why I chose it
It is quite powerful and highly customisable

Limitations

Cannot extract the code
Self learning and resolution. The online forum can help but sometimes you have to figure out the solution yourself
It is quite powerful and highly customisable

Results

Got it to work for the proof of concept

Conclusion

Summary

Satigi aimed to help sole UX Designers speed up their User Research process, from recruiting participants to insights gained
Satigi’s MVP aimed to reduce the amount of time a sole UX Designer on a team spent synthesising transcripts and data collected from user testing

Lessons

This project was an exercise in how to assess good ideas
It was also an exercise on how to guide potential customers to use an application they have not experienced before
Given it was a startup, it was an exercise on how to work in an ultra fast paced environment
An exercise in figuring things out yourself as what you are building is entirely new