Service delivery

The operational job: turning the messy reality of mentor session notes, voicenotes, paper forms and supervisor WhatsApps into a structured record - per beneficiary, per programme, every week - that the supervisor can actually review without losing a Friday afternoon.

The shape Loop takes here is a four-Loop chain: session extraction across all the inbound channels, per-beneficiary aggregation, outcomes measurement on a standard instrument, supervision email drafted with the questions worth asking. A hard safeguarding gate that won't let "continue as planned" ship if there's an open concern.

How we'd shape it

For service-delivery teams the discovery work is anchored in the supervisor's review surface. We sit in with one of your supervisors for two days early on, watching how they actually triage the week's cases. The Loop is shaped to compress that triage, not replace it. The discovery output is a definition of "what the supervisor sees on a Friday morning" - and everything else is built to feed it.

What gets built and stays running

We'd build:

• Session extraction across whatever channels your team actually uses (email, voicenotes, paper-form scans, WhatsApp, your case management system)
• Per-beneficiary aggregation that preserves provenance - every line in the aggregate can be clicked back to the session it came from
• Outcomes measurement against a standard instrument (sWEMWBS, Goals-Attainment, your own equivalent)
• Supervision email drafted weekly, with the questions worth asking, ready for the supervisor to send

Once it's running, we watch the reviewer's edit rate. If it drifts, we adjust the prompts. Models change; your team's caseload mix evolves; we keep the calibration current rather than letting it drift. Monthly report of how the system is performing in your team's hands - never abstract benchmark numbers.

Where this stands today

Built end-to-end against a synthetic UK youth-mentoring charity. 69% of measurements approved as-is by simulated coordinator review; 31% edited - those edits drove the next round of prompt improvement. The four-Loop chain plus the supervisor review surface are ready to land on a real charity's data.