A first AI agent for 211.

The challenge

Many organizations have more information to keep current than they have time for. United Way British Columbia runs 211 for the entire province, a database of over 11,000 community resources, and to stay accredited through InformUSA it has to validate at least 85% of those records every year.

They wanted an efficient way to keep the database current, and earlier software brought in for the job had run for months without delivering. This was their first time building with AI, so the project had to stay small enough to ship while still showing the wider organization what that could look like.

What changed

Over six weeks, we helped United Way build a new curation app and put their first AI agent into production. The organization came away with better data, a hands-on delivery experience, and a foundation for what comes next.

• We eliminated the need for manual spreadsheets. The app allowed Helpline Services to move away from Excel spreadsheets for managing the validation process and assignment of work.
• United Way's first AI agent is live. For each record in the 211 database, our AI agent reads the organization's website and pulls the relevant details into the app for a curator to validate.
• The data going into the database is more complete. Curators once summarized each entry by hand, which meant lost details. The AI agent captures a fuller record from the source, so each record better reflects what each organization actually offers, directly improving Helpline Services.
• The team can change the tool safely and has been through AI delivery once. Updates are tested in a separate environment before reaching production, and the curators and IT have seen firsthand the process of developing an AI agent.

What we delivered

Why it worked

• Shaped around the curators' real work. We watched where validation actually slowed down and built the tool to fit it. The AI handles the gathering while the curators maintain judgment, as accreditation requires.
• Proof before the build. An early prototype showed the approach would work before the full build, and a working v1 four weeks after kickoff meant we built from the team's real world feedback.
• Built with the team, not just for them. The curators and IT were involved throughout the build, making decisions alongside us rather than receiving a finished tool.