Build an AI Social-Service Referral Matching Engine

1. 1

   Solve the resource-data problem first

   The reason directories fail is stale data: agencies fill up, move, change eligibility, or close, and a list nobody maintains sends people to dead ends. Your hardest and most valuable work is keeping an accurate, structured resource database (services, eligibility rules, location, hours, and current capacity). Decide how you will keep it fresh, through agency self-updates, verification staff, or partnerships, because the matching is only as good as the data under it.

2. 2

   Build matching on need, eligibility, proximity, and capacity

   Real matching goes beyond keyword search: it weighs what the person needs, whether they are eligible, how far the program is and whether they can reach it, and whether the program has room right now. Use AI to parse needs from plain-language intake and to rank the genuinely available options. The goal is a short list of programs that can actually help this person today, not a long list of possibilities.

3. 3

   Close the referral loop

   A referral that no one follows up on is just a suggestion. Build closed-loop referral: the receiving agency confirms whether the person connected and was served, and the referrer sees the outcome. Closed-loop tracking is what makes the system trustworthy and what agencies, funders, and health systems increasingly require, and it is the main thing that separates you from a static directory.

4. 4

   Handle consent and sensitive data properly

   You are routing people's needs, which can include health, housing, substance-use, and domestic-violence information. Build consent-based sharing, minimize what you pass to each agency, and apply appropriate protections (HIPAA where health data is involved, heightened care for sensitive categories). People will only use a system they trust with their situation, so privacy is adoption, not just compliance.

5. 5

   Land a network and pick who pays

   Value grows with the network: the more agencies participate, the better the matches and the closed loop. Start in one metro or one need domain, recruit a critical mass of providers, and choose a payer. Health systems and managed-care plans (who fund SDOH referrals), hospitals, health departments, community coalitions, and 211-type entities are the realistic buyers, since the people being served cannot pay.

Do not take our word for the money. Put your own numbers in and see what this idea has to earn before it works. Free, instant, no account needed.

Best for: Builders who can maintain messy real-world data and coordinate a network of agencies

Not for: Anyone who cannot keep resource data accurate and current, or anyone who would overstate AI accuracy for vulnerable users. It does not fit a founder who neglects privacy or human dignity.