What we’ve been building, and why
CivicReach is a company built on the idea that social services should be easy. Asking for help should be easy. And providing help should be easy. But too often, it’s downright hard. Too hard for community members looking for support to find it. Too hard for case workers trying to provide that support. And too hard for agencies to meet community needs, while also meeting the burdens of reporting, regulation, and existing, clunky technology.
Our path to making things easy means two things: think big and start small. We want to think big about the systems that need to change: state office mandates, lack of connectedness to healthcare systems, and the overwhelming amount of dual data-entry slowing agencies down. And we want to start small with voice AI and case management technology that can meaningfully and measurably make things easier for case workers and community members.
We’ve spent the past year investing deeply in AI built for social services - that has meant lots of work and resources directed towards understanding your critical workflows, defining AI safety for social services, and exhaustive testing. This work is culminating in a major release of the CivicReach platform. This letter explains what that means for you and what’s coming to the agencies we serve.
What’s working now
- At CoAction in Indiana, when our AI began answering calls for energy assistance grants, application denials fell by more than 15% over six months. Callers now get clear answers on what energy assistance is, if they are eligible, and how to apply. Folks who are in need get critical information, folks who are ineligible don’t have to waste their time.
- At CVCAA in Texas, the annual client satisfaction survey moved from 81% favorable to 96% favorable after our system went live.
Both of these examples are proof that an AI system can be built for case worker support without compromising community members’ experience.
How our approach is different
Most voice AI products work the same way underneath the hood: one large prompt dictates everything the agent is supposed to do, including what to say, when to transfer and how to handle a sensitive call. Sometimes the AI model follows the instructions perfectly, and sometimes it doesn’t.
CivicReach’s voice AI product is different. We have combined generative AI with hard-coded instructions, and proprietary code that is running throughout a call. The result is an AI agent that is accurate, quick and responds naturally and conversationally.
Here’s how this makes our AI a better tool for social service agencies:
- Crisis situations. We have trained our models to recognize when a caller is in trouble, even when they don’t say so directly. Our models identify the crisis situation, such as an energy shutoff, a medical emergency, domestic violence or a risk of self-harm, and guide the agent to respond with care and appropriate direction.
- Doing what it says. When most AI agents say “I’ll transfer you now,” nothing checks that the transfer happens. Our system tracks what happens using hard-coded steps, and doesn’t rely on AI to complete the job. When our agent says it will transfer someone, the system follows through, rather than leaving it up to the AI.
- Staying on track. While most agents use key words to understand a caller’s needs, ours looks at the full context of the conversation. This makes for more accurate conversations and makes the AI harder to trick or misuse.
What’s new in this release
Safety
- Crisis recognition. Each thing a caller says is checked for signs of a crisis, even if they don’t state it directly.
- Exact safety language. Callers in an emergency, self-harm or domestic violence situation hear approved wording for 911, 988 or the domestic violence line.
Transfers
- Transfer offers backed by action. The agent only offers to transfer to a person during business hours, and when it offers a transfer, the system follows through on it.
- Clear consent before transfers. Our agent asks the caller to confirm whether they want to be transferred. If the answer is unclear, the agent asks again, so case workers get fewer accidental transfers.
- Better information delivered after-hours. A caller who gets in touch after hours can still get program information or leave a voicemail.
Call experience
- Answers from approved knowledge. Answers come only from information that you give us. Exact details like phone numbers and dollar amounts are spoken more accurately.
- More natural conversation. The agent replies faster and handles interruptions and pauses better. It says times, ranges and addresses the way a person would, and pauses between list items.
- More reliable calls. If a voice connection fails mid-call, the agent reconnects and repeats itself.
- Spanish support improvements. Callers can get help in Spanish, with crisis recognition robustly tested in Spanish too.
New capabilities
- Eligibility pre-screening. The agent can now use fixed program rules, including income, to tell callers whether they may qualify for a program. Staff make the final eligibility decision. This will enable new agent workflows for callers, for example:
- Callers can check whether they may qualify for a specific program by offering information about their income and household.
- Callers can give their address and hear which programs they may qualify for.
- Callers can be guided through intake questions.
For agency teams
- Detailed call reports. Each call gets a topic and an outcome, including whether it was resolved, transferred or told to call back during business hours.
What’s around the corner
- Self-service content updates. Your team will be able to propose changes to what the agent knows, review them, publish them and roll them back if needed, all without waiting on us.
- SMS follow-up. After a call, callers can get a text repeating what the agent told them, like a list of documents to bring, a link to apply or your office hours.
- CivicReach CRM. Many agencies still run intake and referrals on paper and spreadsheets. For them, we’re building a simple case management system with intake, referrals, follow-up and reporting in one place. For agencies already using a case management system, we’re making sure our AI agent connects directly to it.