AI for Public Interest
AI for Public Interest
The mission: build a shared data layer for social impact
I was excited to lead design of X4Impact, an intellegence marketplace for nonprofits, tech companies, and foundations. It started with Dr Ying Li's (Chef Scientist at Giving Tech Labs) work on Domain Specific Knowledge Graphs (DSKG) [1], and it's now the largest directory of Tech for Good products in the US.
The mission: build a shared data layer for social impact
I was excited to lead design of X4Impact, an intellegence marketplace for nonprofits, tech companies, and foundations. It started with Dr Ying Li's (Chef Scientist at Giving Tech Labs) work on Domain Specific Knowledge Graphs (DSKG) [1], and it's now the largest directory of Tech for Good products in the US.
Promotion video of the final product.
The biggest barrier to social change is a lack of cross-sector visibility between Foundation, Tech Providers, Nonprofits and Academic. Each sector's success depends on deep insights into the other three. X4Impact's ambition is to create a shared intelligence platform - turning individual data points into a shared roadmap for impact.
The biggest barrier to social change is a lack of cross-sector visibility between Foundation, Tech Providers, Nonprofits and Academic. Each sector's success depends on deep insights into the other three. X4Impact's ambition is to create a shared intelligence platform - turning individual data points into a shared roadmap for impact.
The social impact ecosystem between Foundations, Nonprofits, Tech Providers and Academic.
The social impact ecosystem between Foundations, Nonprofits, Tech Providers and Academic.
The problem space: difficult to find deep insights
To understand the problem space better, we interviewed over 15 nonprofits and over 15 tech enterprenurers, and 10 foundations.
The problem space: difficult to find deep insights
To understand the problem space better, we interviewed over 15 nonprofits and over 15 tech enterprenurers, and 10 foundations.
User journey of nonprofits, tech companies, and foudnations.
User journey of nonprofits, tech companies, and foudnations.
A typical many things to solve situation. Foundations wants data on which tech solution are adopted by npos to decide which to fund. Nonprofits are confused by the many tech products on the market. Tech companies looks for info to deeply understand social issues they can help solving. We brainstormed a few potential solutions to common painpoints. Presented to users and engineers to gauge info about the impact and effort needed.
A typical many things to solve situation. Foundations wants data on which tech solution are adopted by npos to decide which to fund. Nonprofits are confused by the many tech products on the market. Tech companies looks for info to deeply understand social issues they can help solving. We brainstormed a few potential solutions to common painpoints. Presented to users and engineers to gauge info about the impact and effort needed.
Fighting over what to solve first.
Fighting over what to solve first.
For MVP, we decided to focus on Search first, then Create and Money Flow experience, as they cover overlapping needs, and fairly low effort to implement.
For MVP, we decided to focus on Search first, then Create and Money Flow experience, as they cover overlapping needs, and fairly low effort to implement.
Site structure
The website is going to be information dense. Without much insights into the data structure early on, I designed a simple search bar. It was meant to surface the key actions, limiting noises.
Site structure
The website is going to be information dense. Without much insights into the data structure early on, I designed a simple search bar. It was meant to surface the key actions, limiting noises.
One of the early map of site structure.
One of the early map of site structure.
After testing, my assumption that users would come hear with specific questions, looking for specific answers was wrong. Actually, most of them look to be inspired. I looked into what a directorial or scoped search experience could look like.
After testing, my assumption that users would come hear with specific questions, looking for specific answers was wrong. Actually, most of them look to be inspired. I looked into what a directorial or scoped search experience could look like.
Scoped search
Scoped search
Because of the amount of information (usually thousands of self-created content and millions of 3rd party resources), users need quick feedback of the relevance and credibility of each piece of info.
Because of the amount of information (usually thousands of self-created content and millions of 3rd party resources), users need quick feedback of the relevance and credibility of each piece of info.
Search
How can users quickly know how relevent a paper, piece of article, or report is what they are searching for? I was experiementing different patterns to help users narrow down search results faster.
Search
How can users quickly know how relevent a paper, piece of article, or report is what they are searching for? I was experiementing different patterns to help users narrow down search results faster.
One thing keeps coming up during conversations with users is United Nations Sustainable Development Goals UN SDG. It's common launge between social sectors.
One thing keeps coming up during conversations with users is United Nations Sustainable Development Goals UN SDG. It's common launge between social sectors.
What if we transform the long boring list to a lucky wheel, using color and icons to speed up the search?I did a quick prototype. Tested with users. It resonates greatly with them.
What if we transform the long boring list to a lucky wheel, using color and icons to speed up the search?I did a quick prototype. Tested with users. It resonates greatly with them.
Scope filter translated to a lucky wheel interaction.
The quality of search results is a mix of frequently cited research paper to self-generated content by sectors. It's critical to guide users through noise to build their trust in the system. In database, there are a few metadata we capture to indicate the quality of a written problem/tech solutions.
The quality of search results is a mix of frequently cited research paper to self-generated content by sectors. It's critical to guide users through noise to build their trust in the system. In database, there are a few metadata we capture to indicate the quality of a written problem/tech solutions.
Metadata structure
Metadata structure
Surfacing metadata in a scannable way to show credibility of the content and guide users become critical.
Surfacing metadata in a scannable way to show credibility of the content and guide users become critical.
Translating metadata to scannable info bits.
Translating metadata to scannable info bits.
On top of that, a refined search is activated when detecting a negative association from users. This helps users to narrow down search results, while training our data set. Overtime, it helps build users' trust into the system.
On top of that, a refined search is activated when detecting a negative association from users. This helps users to narrow down search results, while training our data set. Overtime, it helps build users' trust into the system.
Refine search experience
Refine search experience
Create: gradual filling out
Another big part of the experience, is creating. Nonprofits to create challenges. Tech companies to create solutions, and foundation to offer information about grant available. A weel-written challenge is the key to match with the right tech provider and foundation resource. However, drafting one is decent amount of efforts. After creating a content template, I broke down the information pieces to two tiers, and marketed them as "create" and "edit". The goal is to prioritize getting people start than making the content perfect.
Create: gradual filling out
Another big part of the experience, is creating. Nonprofits to create challenges. Tech companies to create solutions, and foundation to offer information about grant available. A weel-written challenge is the key to match with the right tech provider and foundation resource. However, drafting one is decent amount of efforts. After creating a content template, I broke down the information pieces to two tiers, and marketed them as "create" and "edit". The goal is to prioritize getting people start than making the content perfect.
"Create" handling light-weight info input; "Edit" handling in-depth data entry
"Create" handling light-weight info input; "Edit" handling in-depth data entry
Since we made "create" part much lighter-weight, I expect users would work on them on mobile and other smaller protable devices, not like the "edit" part, which mostly happens on desktop.
Since we made "create" part much lighter-weight, I expect users would work on them on mobile and other smaller protable devices, not like the "edit" part, which mostly happens on desktop.
Make the "create" bit of dta entry fliud across screen sizes.
Make the "create" bit of dta entry fliud across screen sizes.
Funding money flow dashboard
Biggest struggle facing nonprofits and foundations is to find each other. For tech startups, they also find it challenge to find where investments go into different social issues. Those are public data. In the database, we fetched recent 10 years of data, and surface on the platform in the easy to digest way.
Funding money flow dashboard
Biggest struggle facing nonprofits and foundations is to find each other. For tech startups, they also find it challenge to find where investments go into different social issues. Those are public data. In the database, we fetched recent 10 years of data, and surface on the platform in the easy to digest way.
Make the "create" bit of dta entry fliud across screen sizes.
Make the "create" bit of dta entry fliud across screen sizes.
Reflections
x4impact is a complex product to start with. We tried to solve multiple problems and serve a couple of user groups simultaneously upfront. It was chaos. I realized a good technique to bring the team back on track is to remind everyone what the use cases are and how our decisions affect them. For example, although any user is encouraged to create challenges and solutions, it is not the main purpose of the platform. This informs our decision to emphasize the explore and browsing experience. By minimizing and limiting secondary or edge cases, we were able to ensure that the primary use case is always optimized even in the face of other needs and features.
Reflections
x4impact is a complex product to start with. We tried to solve multiple problems and serve a couple of user groups simultaneously upfront. It was chaos. I realized a good technique to bring the team back on track is to remind everyone what the use cases are and how our decisions affect them. For example, although any user is encouraged to create challenges and solutions, it is not the main purpose of the platform. This informs our decision to emphasize the explore and browsing experience. By minimizing and limiting secondary or edge cases, we were able to ensure that the primary use case is always optimized even in the face of other needs and features.
Acknowledgments Most of the heavy-lifting work in backend is done by Dr. Ying Li, and researchers in AI4PI Fellowship.
Acknowledgments Most of the heavy-lifting work in backend is done by Dr. Ying Li, and researchers in AI4PI Fellowship.