Back to cases

TeamCorp

AI call analysis that home‑care supervisors can verify in one click.

Sole product designer, working with developers and business analysts.

AI care‑coordination platform · US healthcare · 2023–2024

  • 100% of calls reviewed, up from spot‑checks

  • Critical cases acted on in minutes, down from
     about a day

  • Built for agencies of 20 to 100+ agents

The problem

Home‑care agencies with 20–100+ agents coordinate nurses and caregivers for patients and elderly people. Calls between agents and patients were reviewed by trained call centres, and only a sample of them. From a problem call to resolving a critical conflict took a day or more, and agencies were losing clients. Follow‑up tasks were created by hand. The goal: catch every critical call fast, without growing the QA team.

Key decisions

1. A call‑result view built for verification, not blind trust

Audio, a transcript with critical moments flagged in both the text and the audio timeline, communication scores for every participant, and tasks generated from the call. A supervisor can check any AI flag at its source in one click.

2. A supervisor dashboard that ranks by urgency

The team’s current state, the agents who need support and the most critical calls come first – because time to act matters more than seeing everything.

3. Coaching, not surveillance

Agents see their own scores, a comparison with the previous period and friendly tips. The aim was to make them want to come back to the platform rather than feel watched.

HIPAA shaped the design throughout. The AI agent’s analysis scope was set by the client team; my part was where its results live in the interface, their hierarchy, transparency and in‑place verification.

Where AI helped. Where I overruled it

Claude helped me learn a US government elder‑care program from its documentation in 1 day and synthesise stakeholder interviews into JTBD and pain points.

Where I overruled it. ChatGPT’s interview synthesis ranked shouting as the top problem to flag. Live conversations showed that loud doesn’t mean conflict – someone might just be telling a funny story. So I pushed for the assessment to weigh the context and mood of the whole conversation, not volume alone – and the team built it in.

Result

  • Call QA: manual spot‑checks → 100% of calls
  • Time to act on critical cases: about 1 day → minutes
  • Agents started using the platform more regularly
  • Less reliance on external QA call centres

What I’d do next

Let supervisors mark an AI flag as wrong, and use that signal to measure and reduce false alarms.