Not-for-Profit /Drafting live field guide · 8 min

Case-Note Impact Synthesiser

Reads a de-identified case note against your own outcomes framework and drafts the structured entry — Star scores, goal progress, risk flags — with every value traced to its source, so a program manager reviews in an hour instead of retyping notes into report fields for days.

Open the live demo

theater/demos/nfp_case-note-impact-synthesiser.html · sandbox · read-only

Open

FIG. 1

The live demo, running on fabricated data. Open it to step through the full flow — every output is shown for a person to approve before anything happens.

How it would work

Reads a de-identified case note against the support plan and Outcomes Star baseline, drafts the structured outcomes entry with a source cited for every value, and holds it for a program manager to approve before anything reaches a funder.

Input 01

The note + the framework

A de-identified session note plus its attached support plan and Outcomes Star baseline — goals, goal numbers, and the current per-domain scores on the 1–10 scale.

Agent 02

Extracts, scores, attributes

Pulls goal progress and per-domain direction from the note, proposes updated Star scores against the baseline, flags a sharp wellbeing decline, and cites the source — case note, support plan, or Star baseline — for every value.

Output 03

A draft, in the client record

A complete draft outcomes entry sits in the participant's client record for a program manager to review, edit, approve, or reject. A flagged decline is held for supervisor sign-off before it can feed a funder return.

Where it works well

It keeps the nuance of the note alive all the way into the structured entry, with its source shown.

Best for a program manager or outcomes lead assembling quarterly reporting across a caseload — the person reading dozens of unstructured notes and retyping figures over days.
At twenty or fifty notes a quarter, a draft-per-note that arrives pre-mapped to your Outcomes Star domains turns a fortnight of dread into an hour of review.
The recaptured time goes back into reviewing more cases and the harder interpretive calls — whether a goal genuinely moved — not the data entry.

The slow, invisible loss is the gap between the work a team actually does and the evidence that survives into a funder report — a rich note about an independent bus trip and a money worry gets flattened to a couple of typed figures days later, with the detail gone.

Where it works badly

It is confidently unhelpful with no framework to read against, and confidently wrong on stale inputs.

It scores against the support plan and baseline as they stood when it ran — if a plan was reviewed or a goal closed since, it flags a "win" on a goal you have already retired.
It cannot read safeguarding tone the way a worker can — a note that says "more settled" logs as an improvement even where the worker meant it warily.

The honest test

Pull five of your own recent notes: could a colleague who never met the participant tell, from the note alone, which goal moved and in which direction? If they can't, neither can this.

Point it at free-text notes with no goal references, no domain language, and no Star baseline to compare to, and it has nothing to map progress onto — it surfaces a few facts and marks most of the entry as a gap. You have bought a slower way to do what you already do by hand.

What it doesn't do — and shouldn't

It drafts. A program manager approves. That boundary is deliberate.

WHAT IT DOES

Surfaces extracted goal progress and proposed Star scores, each with its source

Flags a sharp domain decline for supervisor sign-off

Marks anything the note doesn't state as a gap to verify

WHAT IT WON’T

Finalise or lodge an outcomes entry to a funder

Decide whether a participant is genuinely doing better

Invent a statistic to make a report look complete

A wellbeing decline feeding a funder acquittal, or a goal marked achieved on a thin note, has real consequences for the participant and for the organisation's funding and its ACNC Governance Standards obligation to be run accountably. The accountable person stays on the decision because the consequence lands on them — not the tool.

What your data has to look like

Notes that name a goal and a domain, against a current support plan and a recent Star baseline.

Typical readiness

across orgs we see, before the first job

Notes that distinguish a goal from an observation

Needs shaping

A current support plan per participant

Usual weak point

A recent Outcomes Star baseline

Usual weak point

Identifiers stripped before synthesis

Needs shaping

A named outcomes framework in use

Usually ready

The real first job

Most organisations have some of this in good shape and some not — a solid Star process but inconsistent notes, or careful notes but baselines living in three systems. Fixing how information gets captured at the point of the session is usually the real first job — larger and more valuable than the AI layer sitting on top of it.

Right fit if…

You run a recognised outcomes framework — the Outcomes Star, or NDIS goals and baselines

Your notes reference goals and domains, and you keep a current support plan per participant

You assemble quarterly outcomes reporting across a caseload of dozens of notes

You can de-identify notes consistently before synthesis

Walk away if…

Your team writes free-text notes with no goal references and no domain language

You have no Star or baseline to compare progress against

Your support plans and baselines are stale or scattered across systems

You want a tool that finalises and lodges the funder return for you

Open questions

The worried-buyer questions, answered straight

It can mis-read a note — which is exactly why nothing it drafts is final. Every Star score and goal-progress line carries the source it came from — the case note, the support plan, or the Star baseline — and the draft sits in the participant’s client record until a program manager approves, edits, or rejects it. A sharp domain drop, like a Money score falling against the baseline alongside a missed appointment, is held for supervisor sign-off rather than sliding into an acquittal unseen. The number a funder sees is one a human approved.

Partly, and honestly that is the real test. If a note distinguishes a goal from an observation and names a domain or a measurable change, it extracts cleanly. If notes are a wall of prose with no goal reference and no baseline to compare against, it surfaces less and marks more as a gap to verify. Most community-services organisations have a mix, and tidying how notes get captured is usually the first piece of work — often more valuable than the AI layer itself.

No. It drafts the entry; the program manager still owns the judgment — whether a goal genuinely progressed, whether a flagged decline needs a different response, what story the quarter tells a funder. The time it gives back is the retyping and cross-checking, redirected to reviewing more cases and the harder interpretive calls. We accelerate the thinking; the thinker stays in charge.

It reflects the session note, the support plan, and the Star baseline as they stood when it ran. If a participant’s plan was updated, a goal closed, or a fresh Star completed after those documents, the draft reads against stale targets and may flag progress on a goal that no longer applies. It works best run close to when the note is written, against the current support plan and the most recent Star.

The pattern is built to run on de-identified notes — names, addresses and other identifiers removed before synthesis — and it outputs no client names or locations. Case notes are sensitive information under the Privacy Act, and a charity’s APP 11 obligation is to take active steps to secure it; where the pattern runs and what it retains is scoped to your privacy and data-handling rules as part of the build. Nothing is shared externally; the draft stays in your client record until a person approves it.

No — that is a deliberate boundary. It only records what the note and attached documents actually state, and marks anything not stated as a gap to verify rather than estimating it. A thin note produces a thin draft with visible gaps, which is the honest signal a program manager needs, not a fabricated full page that quietly corrupts the evidence base.

What it takes to build

3–4 weeks · 4 phases

Reused from template~70%

Bespoke to this skin~30%

stack · Claude · framework mapping · review UI

What it would cost

Fixed scope, fixed price, fixed dates.

Bite-sized first piece

One contained change, low risk

Pilot build

Most builds land here

Embedded support

Scale on proof

Considering this for your org?

The honest place to start is a bite-sized first piece — one contained change, low risk. Tell us where the reporting hurts; we'll play it back, scope it, and show you what's possible.

Book a call How we work