Humanitas Labs: Responsible AI for the 211 Ecosystem

Fiscally Sponsored Project of Digital Harbor Foundation, a 501(c)3 nonprofit organization.

Background: 211 is one of America’s best kept secrets. Most people do not realize 211 is the primary gateway for 20 million people to access social care. There are 240 call centers nationwide, composed of 200+ independent legal entities. Most funding for 211s comes from philanthropy and government, and broadly speaking, the operational budgets are inadequate relative to the staffing requirements and needs of communities. With no centralized leadership or common tech infrastructure for 211s across the country, the system is only as strong as each individual 211. Most 211s are struggling as call volumes have doubled since the start of Covid. This has led to insufficient capacity, suboptimal referrals, and high burnout and turnover amongst agents. The 211’s lack of shared investment and economy of scale is limiting their effectiveness. 

Most 211 leaders recognize the need to stay ahead of the technology curve, yet struggle as independent, single-entity organizations due to the constraints of their size. Through discussions with 211 leaders, we identified how robust AI techniques can improve four outcomes. 

Project Outcomes:

1. Improve the ability of 211 agents to search for resources 

2. Improve the ability of 211 agents to triage and diagnose caller needs

3. Reduce wait times for specific call types

4. Create a measurement framework to assess unconscious bias

These system-level outcomes require an interoperable infrastructure that enables a scale currently missing across the nonprofit sector. Too much money is currently spent on inefficient processes, causing a sub-optimal allocation of resources relative to community needs. The data and analytics infrastructure we propose will ultimately enable the sector to more effectively and optimally address the Social Determinants of Health (SDoH) at the community level, especially as AI capabilities mature.

We are using modern data cloud architecture, Gen AI, and proprietary 211 data to build a technology environment capable of transforming how people search and access social care information and referrals. The project consists of three incremental use cases to pilot and test. These capabilities build upon each other from most manual to most automated, and the deployment, monitoring, and analysis of each are the major milestones.

Use Case 1: Improve Search for Programs & Resources. Help a 211 specialist conduct a natural language search independent of the social care taxonomy with over 9,000 categories. 

Use Case 2: Diagnose & Triage Social Care Needs. During the call, use AI to diagnose a caller’s life circumstance and the corresponding social care needs. Provide a comprehensive list of needs and the most relevant agencies based on the caller’s location and cultural needs. Provide the results to the 211 specialist who delivers them to the caller, keeping the human in the loop.

Use Case 3: Answer Calls: For call types with high wait times (15+ min) or very simple call types, introduce a bot into the phone tree or Interactive Voice Response (IVR) to address the needs of the caller and reduce the number of calls that require a “live answer.” 

For our search prototype, we uploaded each 211’s proprietary resource directory into an enterprise environment of ChatGPT. We provided several instructions on the kinds of data we wanted to highlight in our test cases, and then validated the results with our 211 partners. Our prototype proved we have a viable solution that just needs to be built to add immediate value for the 211 system. 

The product suite consists of several capabilities designed to support the 211 call centers and 211 agents. The primary application for the 211 call centers is an administrative platform to allow 211s to register, manage their agents, update their data, and monitor overall performance. This is similar to most software as a service platforms. The primary application for the 211 agents is a browser extension that will allow agents to search the resource directory using prompt engineering and Gen AI. We chose to provide a browser extension because 211s use a wide variety of software platforms, and an extension is the fastest way to add value.

Over the next few months, we are focusing on Use Case 1 & 2. Our budget estimate is $1 million, which after a grant of $900,000 from GitLab Foundation, we only need another $900,000 to complete development of our product suite.

Our solution serves the 211 call centers, 211 agents, and the people who call for help. Many of the callers battle mental health issues, addiction, poverty, are living paycheck to paycheck, and/or face difficult and dire life situations. These callers span many diverse and vulnerable populations, meaning our program can make a real difference to create equitable outcomes for millions. Our 211 pilot markets collectively support 500,000 callers and employ ~500 agents a year. We have thirty 211s on our expansion waitlist; they support 3 to 4 million callers and employ thousands.

The four pilot market 211s are:

• United Way of Connecticut (entire State)

• 211 San Diego

• 211 Tampa Bay Cares

• United Way of Greater Toledo

The primary stakeholder for our solution is the 211 agent. Staff turnover is high across the 211 network, with many agents having less than 12-18 months of tenure. These agents face a high learning curve as they must learn a social care taxonomy of more than 9,000 categories. At the same time, they must navigate clunky software platforms, on top of having to provide relevant information and referral services for individuals facing challenging life circumstances.  Our solution will decrease the learning curve and provide better agent and caller experiences. To be clear, the AI use cases are to augment and assist existing staff, not replace them. 

Our human-in-the-loop AI solutions will help 211 agents quickly search resource directories for applicable services, diagnose a caller’s social care options, and present the agents with the most relevant agencies and services to refer to the caller. Current searches can take minutes, if not 10s of minutes for agents to find the right resources. This drives up call center costs and decreases the caller experience. Our Gen AI prototype has proven we can find relevant resources in seconds. This will reduce wait times, call handle times, and drop rates. Our aim is to provide better and more consistent experiences for both callers and 211 agents, while driving significant capacity and efficiency gains throughout the entire ecosystem. Ultimately, our aim is to dramatically transform how people access social care, and be available to the entire country.

The secondary stakeholder for our solution is the 211 caller. Per the GitLab Foundation grant assessment of the economic value of our solution, they created a year-1 model for the “aggregate present value income increases” of the 500,000 callers in just the four pilot markets. The present value of aggregate income gain is more than $113 million. If we extrapolate that to the 20 million callers for the entire system, and 10s of millions more uncounted on digital search channels, we have a society-changing solution for the social care sector.

Humanitas Labs is taking a non-traditional approach to modernize the social sector by addressing the primary reasons why systemic change has not occurred to date. We have the technical and AI expertise, a strong product roadmap with validation from our partners, a robust administrative infrastructure designed for rapid expansion, an established and well-respected fiscal sponsor, and trusted relationships with the majority of 211 leaders nationwide. One of the Humanitas Labs leaders is a former network insider, and through this relationship, we partnered directly with several of the most advanced 211s in the country to define and validate our use cases and suite of products. We are willing to work within the comfort levels of the 211s, and are building our products in close collaboration with leaders who represent a diverse set of states, communities, and capabilities. Another one of our leaders is an established tech founder from Silicon Valley, who is up to date on the latest advancements in the tech sector. In short, our partners told us our combination makes for the most promising approach to deliver true value for the system.

What makes this one of the most exciting and important applications for Gen AI in the social care ecosystem are the operationally oriented use cases. The solutions we highlighted are focused on solving the biggest problems of the 211 system. And through a 14 page legal agreement, we built a partnership and data sharing framework that allows us to work with all types of 211 data. This includes the resource directories, call center performance data, agent data, caller information, and needs data. There are numerous safeguards to protect the data security and privacy of the 211callers, and conversely, the reputations of our 211 partners. Very few, if any organizations have this level of trust with 211s, and this gives us an edge over others.

We have a working prototype that uses an enterprise version of ChatGPT to search the resource directories of our 211 partners. We cannot move to the pilot phase until we receive our next round of funding to build the actual solutions for use in the 211 call centers.

We validated the prototype functionality with our pilot partners. They shared positive feedback and excitement. Once we have the AI products developed, we will test within the call centers.

We desire to be part of a learning cohort to advance the use of technology and AI to address the biggest social challenges in the country. We are experiencing the benefits of being part of the GitLab cohort for Economic Opportunity, and want to expand our sphere by being a Solver team.

We are introducing Gen AI into a historically conservative call center environment with significant mistrust and skepticism from front-line staff. One of the reasons for the mistrust is a lack of understanding in how AI can be used as “assistive technology.” Everyone reads how AI will replace humans, but that is not our intent. When 211 leaders hear about our solutions, they get excited. The reality is that there are many innovative 211 leaders across the system. We have done our due diligence in partnering with several of the best and most willing. They know AI is here, and want to explore the most advanced applications for the network, or risk becoming obsolete. These 211 leaders care about making a difference and helping people in need in their states and communities, and believe new tech can help. Beyond that, one of our 211s is a blended call center with 988 and if the use cases prove successful, would like to test the capabilities for mental health and suicide prevention services. We foresee expansion of these capabilities into the 988 system, the 311 system, No Wrong Door system, and possibly even 911.

We co-designed our solution in partnership with several of the best 211s in the system. We know it will work because they told us it is exactly what they need. We conducted desktop reviews with representatives from each of our partner call center agents. We listened to calls, reviewed their processes, their systems, and their data. We listened to the challenges of the leadership team and drafted a strong concept paper. We then built our use cases, evaluation framework, and initial prototype. What’s more, we then shared our results with over 30 other 211s across the country. There is near unanimous excitement for our product suite, and a sense of urgency to move from prototype to production to meet the demand and needs of the system.

In collaboration with our 211 partners, we drafted an extensive measurement and evaluation framework that spans five pages. We include the traditional measures used by 211s, and a new set of aspirational measures we will develop. The framework covers the following areas:

1. Referral & Outcome Success

2. 211 Agent Performance

3. Impact for 211 Callers

4. 211 System Efficiency

5. Digital Channel Performance

Some of the measurement & evaluation highlights include:

Most 211s count as the basis for understanding the success of their contact center operations and referrals. They count the number of calls, number of referrals, number of need types, etc. There is a recognition 211s need to move beyond counting to a measurement system that also values the quality of their performance. We need to know what makes a good referral, a successful call outcome, and even an optimal referral. Answers to these questions are vital as we train our models. It will allow our AI to provide agents with recommendations at a speed, accuracy, consistency, and efficiency that transforms social care navigation for 10s of millions annually.

Despite annual budgets totaling over $300 million collectively as a network, the 211s lack shared investment and economies of scale, limiting their effectiveness. 211s are struggling as call volumes have doubled since the start of Covid. This has led to insufficient capacity, suboptimal referrals, and high burnout and turnover amongst agents. A conservative estimate of our AI solutions can provide 10% efficiency gains for our 211 partners, meaning $30 million annually when fully adopted by the network. We foresee greater capacity gains coming via: more available call handle time, more individuals served, more referrals per individual, and lower agent turnover. 

We know Social Determinants of Health (SDoH) influence life outcomes. The GitLab grant impact model from the AI for Economic Opportunity Fund showed the 500,000 callers in just the four pilot markets would see an “aggregate present value income increase” of more than $113 million in year 1. We want to continue to advance this methodology, and evaluate how 211s and our AI solutions make lives better. How does our work lead to more economic prosperity and a healthier society?

211 callers are people from all walks of life, but are predominantly those battling poverty, living paycheck to paycheck, facing mental health issues, or struggling with addiction. 211s curate resource information for “Target Populations” which represents diverse and vulnerable populations. We need to understand the influence of unconscious bias that 211 agents may introduce on the calls. 211 leaders across the country identified unconscious bias as a concern, but it has never been measured.

We have a secure AWS cloud data environment to manage the various sources of 211 data, and an enterprise license for ChatGPT as the starting point for our AI. For our use cases, we will implement various AI/ML techniques like naturalized search, audio transcription, machine translation, summarization, and a recommendation engine. We are building the browser extension in Chrome. We also plan to build integrations with the existing telephony and information and referral vendor platforms to improve the speed and transmission of data. 

While we believe we know what we need, we are open to collaboration from the MIT Solve team to build the optimal solutions.

Full-time staff: 2

Contractors: 5

We have a small leadership team composed of Phil Chow, Kevin Claybon, the fiscal sponsorship of the Digital Harbor Foundation, and technology contractors to support product development. It is worth noting that our 211 partners are contributing their thought leadership to the program.

Ten (10) Months

Digital Harbor Foundation provides opportunities of a welcoming and inclusive environment that include:

1. Diverse Leadership Team: Having a diverse leadership team is crucial for setting the tone and direction of the organization's diversity initiatives. Digital Harbor Foundation can ensure diversity at the leadership level by actively seeking out and promoting individuals from underrepresented backgrounds.

2. Inclusive Hiring Practices: Implementing inclusive hiring practices, such as using diverse candidate pools, ensuring unbiased job descriptions, and providing training on unconscious bias to hiring managers, can help attract and retain a diverse workforce.

3. Equitable Policies and Procedures: Reviewing and revising organizational policies and procedures to ensure they are equitable and inclusive is essential. This includes addressing pay equity, providing accommodations for diverse needs, and promoting flexible work arrangements.

4. Training and Education: Providing ongoing DEI training and education for all staff members can help raise awareness of unconscious biases, promote understanding of different perspectives, and cultivate a culture of respect and inclusion.

5. Community Engagement: Engaging with external communities and organizations that focus on diversity and inclusion can provide valuable insights, partnerships, and opportunities for collaboration.

6. Regular Feedback and Assessment: Conducting regular surveys and feedback sessions to assess the organization's progress on DEI goals and identify areas for improvement is important. Actively listening to employees' experiences and concerns and taking concrete actions based on feedback demonstrates a commitment to continuous improvement.

By implementing these strategies and fostering a culture of diversity, equity, and inclusion, Digital Harbor Foundation can create an environment where all team members feel welcomed, respected, supported, and valued, ultimately leading to greater innovation, creativity, and success.

In terms of revenue, there are over 200 independent 211 organizations across the country that make up our target customer base. We will provide our product suite as a SaaS offering to enable them to use AI to solve their most pressing capacity issues. The 211s need these solutions to increase their call center handle time capacity, decrease the learning curve for new agents, and provide more consistent referrals and caller experiences. With operating budgets over $300 million annually, our solutions are a bargain for what they will deliver. There is high demand.

In terms of impact, our primary customers are the 20 million callers to 211 each year, the untold 10s of millions of visitors to 211 websites trying to search for resources, and any individual in this country that is attempting to access help or navigate the social care ecosystem. We plan to conduct focus groups with end-users as we launch the product suite - especially products that would interface directly with a person in need of help. As AI becomes more ubiquitous in our everyday life, we foresee strong demand and acceptance for our solutions as it goes from novelty to expectation. While we are not “selling” these services, we aspire that our offerings make for a better user experience, and increase access to the most relevant services. If successful, the data at our fingertips can be used to optimize the social sector, and the untold billions that flow through it annually.

We have a nonprofit structure, and are not trying to make a significant profit with our solutions. Rather we truly want to add value to the system and reinvent how people access and navigate social care. Over the next year, we expect to cover our budget through grant funding. We have a phased approach to build our initial prototypes and expand across the 211 network. We successfully developed our prototype with $100k of funding from the Gitlab Foundation grant. 

Through our discussions with several major funders, we are optimistic we will receive greater levels of funding as we hit key milestones. We estimate we need $3 million over the next 2 years to build the solutions and provide the expansion 211s a trial year as they will want to test them first. The funding will also cover data environment and processing costs, which are not trivial. Our short-term funding needs are more modest between $200k and $500k depending on the scope of deliverables a partner wants to fund. We also are in line to secure funding from some of our 211 partners - we received a soft-commit of $200k - as they see the value of these solutions.

Beyond that, the longer term plan is to license the product to the 211s at cost plus roughly 20%. This will ensure a small staff to maintain the solutions and look to expand the capabilities over time. The value of these solutions will add to the 211 system will make it a no-brainer. And as we potentially license these capabilities to the 988 and other navigation providers, we can continue to lower the cost structure for our 211 partners.