Outcomes

Outcomes register · live

Every AI feature in the system has a baseline (without it), a target (with it), and an observed value. If a feature can't pass this test, it's cut. Per ADR-019 + outcomes-register.md.

Live outcomes hitting target

12 / 12

· ≥95% of target

Time saved this week

~62 hr

↑+11%· across all crews + ops

Knowledge artefacts captured

↑+4· from 3 sessions

Hard-stop incidents avoided

· calibration scheduler caught them in time

Quote draft from RFP (C2)

Time-to-quote, p95

Baseline: 120 min
Target: ≤15 min
Observed: 12 min

End-of-shift wrap (C8)

Time-to-file shift, p95

Baseline: 900 sec
Target: ≤30 sec
Observed: 28 sec

Daily diary → report (H7)

Time-to-report, p95

Baseline: 90 min
Target: ≤5 min
Observed: 5 min

Knowledge Capture (C1)

SOP drafts per offboarding

Baseline: 0
Target: ≥6
Observed: 9 (avg)

Flight risk predictor (H1)

Lead time, flag → notice (median days)

Baseline: 0
Target: ≥14
Observed: 21

Pulse synthesis (H2)

Time-to-themes per cycle

Baseline: 180 min (or skipped)
Target: <30 min
Observed: 22 min

JSA prefill (C5)

Time-to-ack JSA, p95

Baseline: 1200 sec
Target: ≤90 sec
Observed: 75 sec

AS 5488 QL classifier (C4)

Senior rejection rate (target <10%)

Baseline: —
Target: <10%
Observed: 7.2%

1:1 prep agent (H15)

Manager prep time (median min)

Baseline: 30 min
Target: ≤5 min
Observed: 4 min

Onboarding/offboarding playbook (H4)

Time-from-event to playbook live (min)

Baseline: 60 min manual
Target: <5 min
Observed: 3 min

Career path nudges (H5)

% milestones recognised in 24h

Baseline: ≤10%
Target: ≥90%
Observed: 94%

NDB anomaly detector (H14)

Median time pattern → flag (min)

Baseline: Days to weeks
Target: <60 min
Observed: 12 min

Cut for failing the outcome test

Outcomes filter→ kill features that don't earn their placeHaiku 4.5 · Sydneysuggestion

Generic "AI dashboard" widgets — no specific outcome. Replaced with role-specific KPI tiles tied to outcomes per role.
Real-time staff chat — Slack already does this. Adding it duplicates without outcome benefit.
In-app point-cloud viewer (Phase 1) — outcome ("preview point clouds quickly") doesn't justify the build cost. Deferred to Phase 2; clients use desktop tools.
Generic "ask AI anywhere" widget — superseded by outcome-bound features + Cmd-K command palette.
Multi-currency in v1 — no Phase 1 customer outside AUD; built when first international job lands.

Quarterly review walks this list with the user. New cuts go below; resurrections require a fresh outcome statement and a metric proxy.