The chain reaction that leads to homelessness
On a single night in January 2024, 771,480 people were counted homeless across the United States. This marks an 18% increase in just one year, the largest single-year increase on record. Every year, billions are spent on emergency responses that struggle to keep up. The issue is not a lack of effort or funding. The real problem is that we are trying to solve the wrong thing.
Homelessness is not just a collection of individual hardships. It is a failure of systems working together. Today, two fields give us a clearer way forward: network science and artificial intelligence.
771KPeople counted homeless in the U.S. on one night, Jan 2024 — the highest number ever recorded |
7.1MAffordable homes currently missing from the U.S. market for the lowest-income renters |
35Affordable units available for every 100 extremely low-income renter households nationwide |
What network science actually is and why it changes everything
A network is simply a set of things connected by relationships. Your city's housing system is a network. People, landlords, employers, shelters, hospitals, housing waitlists, social workers, and benefits offices are all nodes. The referrals, lease agreements, appointments, and income flows between them are the edges, the threads holding the system together.
What keeps a person housed is not just a home. It is the combined stability of every thread in their web: income, a workable lease, healthcare, family or friends nearby. Homelessness begins when too many of those threads snap at once. Network science, a branch of mathematics and computer science that studies interconnected systems, gives us the tools to map those threads, measure their strength, and predict which ones are about to break.
The same mathematical laws that explain how power grids fail in cascading blackouts, how epidemics spread through populations, and how the internet routes around broken infrastructure also govern how housing instability spreads through cities. This is not a metaphor. It is the same underlying graph theory applied to a different system.
Four network properties that explain why homelessness clusters
We don't need to be a mathematician to understand the core ideas. Four properties of networks explain most of what we observe about homelessness patterns in cities and point directly at what to fix.
Hubs
A small number of nodes carry the majority of connections. In homelessness systems, one large emergency shelter typically routes hundreds of people toward housing and services. When that hub is weakened through underfunding or overcrowding, every path depending on it begins to collapse at the same time. This is why budget cuts to central services cause disproportionate harm.
Cascades
Failure spreads. Losing a job makes missing rent more likely. Eviction collapses credit. Poor credit blocks re-renting. Each broken connection makes the next one more probable. Using network projections of medical records, researchers found that the health and economic signature of imminent homelessness becomes visible in data as early as 60 days before it occurs.
Bridges
Bridge nodes connect otherwise isolated communities to the rest of the network. Outreach workers are the most critical bridges in any city and they physically create the first link between unsheltered people and formal services. Remove one outreach worker and a pocket of 15-30 people becomes structurally unreachable. No AI system can substitute for this relational role.
Anchor nodes
Stable housing is the anchor node of a person's entire network. Without a home address you cannot apply for jobs, receive benefits, or maintain medical care. Providing housing first, before addressing other needs, works precisely because it restores this anchor, letting every other connection rebuild around it. This is the network science basis of the Housing First model.
Why homelessness concentrates where it does: the bottleneck
If homelessness were purely about personal misfortune it would be distributed relatively evenly. It isn't. Around 64% of all homeless people in the U.S. live in just 7 states. California alone accounts for nearly one-third of the national total despite holding 12% of the U.S. population.
Network science explains this with the concept of a bottleneck. A bottleneck is a point where a large flow must pass through a limited opening. In cities with severe housing shortages, the housing inventory is the bottleneck. The NLIHC's 2025 gap reportfound that nationally only 35 affordable and available homes exist for every 100 extremely low-income renter households. In the worst metro markets — Las Vegas, Dallas, Austin , the ratio drops below 20 per 100. No matter how well a support network functions, it cannot route people to housing destinations that do not exist.
How Artificial Intelligence operates on the housing network
Network science tells us where the system breaks. AI tells us when and for whom and increasingly, how to fix it before the break happens. There are three distinct roles AI plays in the housing network, each targeting a different part of the system.
Role 1: Prediction - catching cascades before they start
The most powerful AI application is predictive prevention, using machine learning to identify households at imminent risk of losing housing before the cascade begins. In Los Angeles County, UCLA's California Policy Lab trained a model on anonymised records from nearly 100,000 people across 8 county agencies covering physical health, mental health, substance use, arrests, and benefits sign-ups going back to 2010. The model generates a risk score representing each person's probability of losing housing within the next 18 months.
The results are striking, people on the high-risk list experience homelessness at a rate 3.5x higher than the general eligible population. Armed with these lists, LA County's Homelessness Prevention Unit proactively contacts people before they lose housing, offering cash assistance and case management at the moment it can actually prevent a fall rather than just respond to one.
Role 2: Routing - AI as GPS for the shelter network
Once someone is inside the system, AI dramatically reduces the time it takes to connect them to the right resource — a bed, a housing unit, or a specialist. This is classic routing optimization, given a graph with limited capacity at certain nodes, find the shortest path from any person to a destination.
In San Diego, a countywide bed-matching app now matches individuals to available shelter beds in minutes rather than the days the previous manual phone-tree process required. In New York City, the StreetSmart system equips outreach workers with field tablets that log client updates in real time while ML models surface individuals with elevated overdose or hypothermia risk, allowing finite outreach hours to reach the people most likely to be in immediate danger. Portland has gone further, using algorithms to track re-entry risk after housing placement, enabling case workers to adjust support before someone loses housing a second time.
In the academic literature, researchers at State University of New York demonstrated that ML-based facility recommendation for 38,800 people seeking homelessness assistance in New York's Capital Region could significantly improve matching between individual needs and available services — reducing inefficiencies that leave people in the wrong part of the network entirely.
Role 3: Supply - building the missing destination nodes
No routing optimization solves a network with structurally insufficient destination nodes. With only 35 affordable homes per 100 extremely low-income renters, the pipeline itself is the problem. In 2025, Community Solutions and the Brookings Institution launched the Home Genome Project — an initiative to use AI for mapping underutilised land, modelling zoning reform scenarios, and identifying viable sites for new affordable housing at a scale and speed no human planning team can match.
More than 150 U.S. communities now participate in the Built for Zero initiative, using real-time person-level data integration and AI-assisted case management to work toward functional zero homelessness. Rockford, Illinois became the first U.S. community to actually reach this milestone, ending veteran and chronic homelessness through this exact model.
Homelessness is not a mystery. It is a network problem with network solutions. The threads that hold people in their homes are visible, measurable, and repairable, if cities choose to treat this as the system's failure it actually is rather than the personal failing they have long assumed it to be. Network science shows exactly where the web tears. AI shows when and in whom.
Why this matters locally
This is not only a national issue. It is becoming increasingly visible at the local level as well.
In Rochester, Minnesota, rising housing costs and limited affordable options are putting growing pressure on vulnerable communities. Recent reports show that homelessness has been increasing in the city, with more people relying on temporary shelter and some turning to encampments across the area.
Local service systems are also feeling the strain. Shelters are seeing a surge in demand, with more families seeking assistance and facing longer wait times than before.At the same time, broader housing data shows that affordability gaps continue to widen, making it harder for low income residents to secure stable housing.
In 2024 alone, more than 1,000 individuals were counted as experiencing homelessness in Rochester, including both sheltered and unsheltered populations.What we are seeing in Rochester reflects the same system level patterns described earlier. The network is under strain. Key connections are weakening, and the housing shortage continues to act as a bottleneck.
The OPAL Project is built on the same principles of network science and artificial intelligence outlined in this piece. Instead of treating homelessness as isolated cases, OPAL looks at the full network and works to strengthen it.
Through a network science approach, OPAL maps the relationships between individuals, housing options, service providers, and outreach workers. This helps identify where connections are missing, where delays occur, and where the system is most likely to fail.
Artificial intelligence adds another layer of support. It helps identify individuals who may be at risk of falling through the cracks, improves how people are matched to available housing and services, and enables faster, more informed decision making across the system.
In simple terms, OPAL is designed to do three things:
Detect risk earlier
Connect people to the right resources faster
Strengthen the overall system so fewer people fall into homelessness in the first place
We are proud to begin this work in Rochester alongside partners such as Mayo Clinic and ZVMS. Their local knowledge and commitment are essential in making this effort meaningful and effective.
Help us act sooner
If you believe in building systems that prevent homelessness instead of only responding to it, we invite you to support this work.
Your contribution directly supports the people, partnerships, and technology behind the OPAL Project. It helps us move faster, reach more communities, and strengthen the systems that keep people housed.
Without support, progress takes time. With your help, we can act sooner.
And acting sooner matters. Because for thousands of people, every single day includes the same question: Where will I sleep tonight?
If you would like to support this work, you can contribute here: https://www.pathcheck.org/donate
And if you are interested in going a step further, whether as a financial partner or collaborator, we would be glad to connect and explore how we can work together. You can reach us at: barsha.devi@pathcheck.org
