Baseline aggregates are the APSC's published figures (APS Employment Data, 30 June 2025: 198,529 employees, separation rate 6.4%, published job family shares). Open-source data fast-tracks the platform's efficacy: agencies get something valuable from day one, then internal sources deepen it over time. Record-level detail, histories and projections here are modelled from that published open-source data — real baselines, real open sources, modelled forward.
FY2025 endpoints are published APSC family shares; earlier years modelled. Hover for values.
Share of task-hours matching automatable or augmentable patterns. Modelled estimates.
Modelled on published distributions, scaled to the 30 June 2025 headcount.
All 102 APS agencies, selectable below, with their published 30 June 2025 headcounts (APSC). Everything beyond headcount is modelled per agency archetype for this demonstrator. Tier 1 data (APSED, annual reports, budget staffing estimates) makes this view possible before any agency system is connected.
Modelled from the agency baseline; bands as elsewhere.
Modelled composition against the APS family framework.
Scaled from the APS-wide register; the closure levers on the Gaps & risk page apply per agency.
APSC, 30 June 2025: 198,529 employees, 102 agencies, 586 locations.
What the AI transition is worth per agency once transition costs are paid, framed as capacity redeployable to priority work, consistent with the evidence that Gen AI augments more than it replaces. Published headcounts (APSC); impact model coefficients are calibrated to cited evidence and shown below.
Share of task-hours exposed to Gen AI per job family (modelled, 18–46%). Anchor: with LLM tooling roughly half of work tasks could be completed significantly faster (Eloundou et al. 2023); exposure varies strongly by task mix.
Exposed hours split 65% augmentation / 35% automation. Anchor: JSA's Gen AI Capacity Study (2025) finds Gen AI more likely to augment Australian work than replace it.
Only 60% of the theoretical effect is credited by FY2030, ramping over 3 years. Anchor: the whole-of-government Copilot trial: a third of participants used it daily; time savings up to an hour; 40% reallocated time to higher-value work; training was the strongest lever (DTA 2024).
New oversight and data roles (AI Plan mandates), a reskilling drag (~20% of affected staff at ~0.4 FTE-years each), plus the delivery machinery the capacity numbers depend on: cowork-grade AI tooling and embedded enablement squads at team level, security and change-lane uplift, and program management. All costed in the transition budget.
Net = automation capacity released + augmentation headroom − new roles − annualised reskilling drag. All figures are FTE-equivalent capacity, not positions: the evidence supports reshaping via redeployment and mobility, not headcount cuts. Coefficients are editable assumptions in the platform; this page shows one defensible calibration. Read alongside the Forecasts page: the demand-supply gap there (~7,400 FTE short by FY2030 at current settings) is what this capacity gets spent on first; the remainder is the savings-or-growth choice space.
Do the transition: about $1.7 billion once, spread over three years (roughly 0.7% of annual payroll each year), with operational budgets held flat in real terms. That figure deliberately funds the whole job: not just planning and oversight, but cowork-grade AI tooling and embedded enablement squads for every team, security and change-lane uplift, and program management. Do nothing: fund roughly 7,400 additional staff from FY2030, about $1.0 billion every year, recurring and growing. The swing between the two futures is about 14,200 FTE-e, near $1.9 billion a year by FY2030 at a modelled $135k fully loaded average staff cost: the full program pays back within a year of the swing, and under two years against avoided hiring alone. All modelled; unit costs stated and editable; per-agency budgets on each scorecard below.
Click a row for its summary. Filter by name. Sorted by headcount; every headcount is published (APSC, 30 June 2025); all impact figures are modelled as above.
A curated baseline mapped to the APS Job Family Framework; extraction from job descriptions, role profiles and learning records lands with the pilot and build phases.
Coverage index 0–100: share of roles in the family with the cluster evidenced. Hover any cell.
A worked example: skills mapped from a position description.
Confidence scores from the extraction model. Below-threshold matches queue for human review.
Change in evidenced headcount, APS-wide.
Each pressure cluster against the 2025 Occupation Shortage List and pipeline evidence. Sources and the full supply picture live on the Labour market page.
Demand and supply projected to FY2030 from the published 30 June 2025 baseline, with honest uncertainty bands. Models are validated point-in-time: scored only on data they could not have seen.
History to the published FY2025 baseline, then projected supply (with bands) and projected demand. Y-axis zoomed to the data range so the divergence reads; the gap chart below shows it directly.
The number the fan chart hides at headcount scale. Above zero = shortfall at current settings.
This gap is not a valuation and not a savings forecast. It says: at current settings (today's hiring, the 6.4% separation rate, no AI-enabled intervention) the work demanded of the APS outgrows the workforce supplying it. Left untreated, the options are hiring to fill it or under-delivering. The Net impact page models the treatment: the AI transition releases more FTE-equivalent capacity than this gap consumes, and the difference is the genuine choice space, taken either as fiscal savings (headcount drifting down through natural separations, no redundancies) or as new demand absorbed without growth. Why believe demand grows at all: the APS added 13,671 people in the year to 30 June 2025 because demand required it (APSC, published), and the DDC Workforce Plan expects specialist demand to keep growing; this projection assumes demand growth at a small fraction of last year's realised rate, a conservative floor. No hiring wave and no redundancy program sits behind either number: the gap is avoided hiring, and the residual is realised through natural attrition (about 12,700 separations a year at 6.4%) or absorbed growth. The gap is the problem at current settings; net impact is what disciplined execution buys back.
Internal supply is only half the equation: the same skills are contested by every other sector, and shocks elsewhere change what the APS can recruit. This page keys the national feeds — JSA's Occupation Shortage List, Internet Vacancy Index and Employment Projections, higher education and VET completions, skilled migration, and the ACS Digital Pulse — against the demand signal from the Forecasts page. Published figures are tagged; everything derived is modelled.
How contested each family's talent pool is nationally. Score = 50·shortage + 30·projected growth + 20·vacancy trend, inputs normalised 0–1 from the sources below; the weighting is a stated modelling choice. Hover any bar for its evidence basis.
The folk wisdom is "everyone is short of AI skills". The data is sharper. Generalist digital is cooling: Developer Programmer, Software Engineer, Data Scientist and Data Analyst are all rated No Shortage nationally on the 2025 Occupation Shortage List, and ICT Professionals job ads fell 10.8% in the year to May 2026. Cyber is structurally short: 4 of 6 cyber occupations are rated Shortage in every state, and the ACS Digital Pulse puts the national need at 54,000 more cyber-skilled workers by 2030. The squeeze returns: JSA projects ICT professional employment up 25.4% over the decade to May 2035 (+106,700 people), against a tech workforce of 1,012,207 today and a national target of 1.2 million by 2030. The strategic window is now: recruit and build digital capability while the private market is soft, before projected demand growth reprices it.
Annual national pipeline against projected demand; APS position within it.
Labour-market shocks the APS does not control, applied to the Digital & Data family forecast. The mechanism is the demonstrable claim; magnitudes are labelled illustrative. Pick a shock.
This page runs on keyed published aggregates, refreshed by hand — honest and current, but a snapshot. The funded platform automates the feeds on their own cycles (IVI monthly, OSL and completions annually, migration quarterly), builds the ABS occupation-by-industry matrix to model cross-sector flows properly, and closes the loop with universities: aggregated skills-to-train demand becomes a live commissioning signal for micro-credentials and graduate streams, so the pipeline responds to forecast gaps rather than last year's org chart.
The register refreshes with every data cycle and ranks by time-to-impact, so interventions are proactive rather than post-mortem.
Full-time-equivalent shortfall at current settings.
Critical capabilities held by fewer than 5 people in an agency.
Pick a shortage for drafted commentary across the four levers: buy (recruit), build (reskill), borrow (mobility and surge), bot (automate task content). Drafted by the strategy layer from pathway and forecast data; a planner owns the final call.
Auto-drafted from forecasts and the skills inventory; owners and treatments assigned by planners.
Verified from primary sources in our research knowledge base. The pattern: policy architecture is converging everywhere; adoption tooling and workforce instrumentation are where jurisdictions separate.
Flagship moves, evidence, and the lesson each one carries for the APS.
| Jurisdiction | Flagship move | Evidence | Lesson for the APS |
|---|---|---|---|
| Singapore | Platform-first universal access: Pair suite for every officer, AIBots for self-serve automation, GovTech central stewardship (Responsible AI Playbook, LaunchPad) | 80% of 150,000 officers on Pair Chat; 20,000+ bots built by officers; 20+ agencies surfaced 40 use cases in 4 months | Give every officer a safe tool and let the workforce build on it. Adoption follows utility, not mandate. |
| United Kingdom | Restructure-first: departmental cuts (Cabinet Office 1,200, DBT 1,500 roles) ahead of the capability map; strategic workforce plan slipped to H1 2026 | Civil service 520,440 (Q3 2025), grown every year since 2016 and 35% above its 2016 low; entrants down over 30% in the year to March 2025 | Cutting before you can see capability leaves nothing to steer with. Instrument first, then reshape. |
| United States | Use-case registers and agency Chief AI Officers driving visible momentum (reported: 1,100+ federal AI use cases, ninefold GenAI growth in a year) | Reported figures from federal inventories; not independently verified in our register | Public use-case inventories create both momentum and accountability. Cheap to adopt. |
| OECD guidance | Building an AI-ready public workforce (Jan 2026): internal capability over outsourcing, training across all staff tiers, hiring mechanisms for digital professionals | Comparative brief across member administrations with the European Commission | Internal AI capability is what preserves accountability and compliance. Buy tools, grow judgment. |
| Australia | Policy architecture now among the most complete: AI Plan (GovAI Chat, CAIOs, mandated literacy), Policy v2.0, DDC Workforce Plan 2025-30 | Copilot trial: a third of participants daily, tailored training the strongest lever; APS 198,529 across 102 agencies | The frameworks are in place; the gap is instrumentation. No jurisdiction has solved AI-era workforce planning yet, which makes it an open first-mover play. |
Australia's policy scaffolding is now ahead of most peers: mandated AI literacy, Chief AI Officers and a universal assistant are commitments Singapore took years to reach. What Singapore proves is the adoption ceiling once tooling is universal and trusted; what the UK proves is the cost of reshaping without workforce instrumentation. The APS sits between them: frameworks ready, instruments missing. That is precisely the gap a shared, AI-enabled workforce planning capability closes, and no incumbent vendor or peer government has closed it yet.
Adjacent-skills matching connects people in transforming roles to future-critical roles they are already most of the way to.
Reskilling programs succeed when the destination is close. Pathways above 60% overlap complete at roughly three times the rate of aspirational moves, and the platform only recommends pathways where the skills-to-train list is short enough to schedule. Every recommendation is reviewed by a human before it reaches an employee conversation.
Every "to train" list above is a course specification. The platform turns pathway demand into a commissioning signal for university partners.
Aggregated skills-to-train lists (SQL basics: 1,900 people; AI assurance: 800) become co-designed micro-credentials with delivery partners, refreshed as the gap register moves.
Graduate intake weighted to forecast gaps 3 years out (data, cyber, AI oversight), not last year's org chart. Curriculum input flows from the skills inventory.
6 to 12 month conversion programs for the highest-volume pathways, delivered part-time alongside redesigned roles, credit-recognised toward postgraduate awards.
Universities evaluate pathway completion and role performance against the platform's baseline, so the reskilling investment case is built on measured outcomes.
Aligned with the APS Academy and the Data, Digital and Cyber Workforce Plan's coordinated approach to capability. Illustrative pathway volumes modelled from published data.
Driver-based what-if scenarios over the levers planners actually argue about. Sliders, not code. Every run is reproducible and auditable.
Presets, or set your own.
Assumptions and model version are stamped on every saved scenario.
Recomputed live as levers move.
Median with 50% and 90% bands against demand.
The strategy layer drafts; the analytics decide. A worked strategy exists for every one of the 102 agencies, grounded on that agency's own scorecard numbers, and a human owns the final document. Drafts below are pre-generated for this demonstrator; in production they are drafted live through the model gateway on the same grounding.
Standard reporting is automated so planners spend their time on decisions. The whole-of-government scorecard and per-agency scorecards below are computed live from the model; the three worked examples after them are pre-generated. Everything prints cleanly for the executive pack.
Outlook. Total headcount is stable to gently declining (−0.6% a year median), but composition shifts materially: service delivery roles decline 2.2% a year while data, digital and assurance roles grow 4–6% a year. The 90% interval on total FY2030 headcount spans 29,100–33,400, driven mainly by AI adoption speed.
Priority gaps. Unmitigated projections concentrate risk in three capabilities:
| Capability | Gap by FY2029 (FTE) | Time-to-impact | Primary treatment |
|---|---|---|---|
| Cyber & information security | 640 | 14 months | Reskilling pathway from ICT operations + targeted recruitment |
| Data analysis | 520 | 18 months | Adjacency pathway from reporting roles; graduate stream uplift |
| AI & automation oversight | 310 | 11 months | New role family; internal pathway from program evaluation |
Actions committed. Task-level job redesign in the two largest divisions; three funded adjacency pathways (first cohorts September 2026); surge pool formalisation; quarterly re-forecast with bias audit at each retraining cycle.
Movement this quarter. Two risks escalated, one retired. The attrition regime detected in FY2024 remains in force for service delivery families; separation rates have not reverted to pre-2024 behaviour, and forecasts continue to weight the post-break regime.
| Risk | Status | Change | Owner |
|---|---|---|---|
| Cyber capability shortfall ahead of legislated uplift | Critical | Escalated (was Serious) | CISO network |
| Data analysis demand outpacing supply | Serious | Unchanged | Chief Data Officers |
| AI oversight roles unfilled at EL1 | Serious | Escalated (was Warning) | HR / integrity |
| Procurement surge for major programs | Warning | Unchanged | CFO network |
| Records digitisation backlog capability | Retired | Automation absorbed demand | — |
Recommended attention. The cyber gap's time-to-impact (14 months) is now inside the typical recruitment-plus-clearance lead time (17 months). Internal pathway activation is the only treatment that closes in time; the pathway cohort should be doubled this quarter.
Finding: ready, with two conditions. The scenario is met from the pre-identified surge pool without breaching minimum staffing in any donor agency, provided (1) activation agreements with the three largest donor agencies are current, and (2) the 380-person training-lapsed cohort is recertified this quarter.
| Source | Available FTE | Activation time |
|---|---|---|
| Surge-ready pool (current certification) | 1,720 | 1–2 weeks |
| Adjacent-skills activation (short conversion) | 840 | 3–5 weeks |
| Recent leavers, re-engagement register | 310 | 4–6 weeks |
Coverage: 2,870 FTE against 2,500 required (115%). Donor-agency service levels degrade by at most 4% for the surge duration under the recommended draw profile.
Your data and the models that read it stay onshore. Horizon runs on infrastructure in Australia with locally hosted models — no offshore processing, ever.
All data and models stay onshore.
The models that read your workforce data run locally, not in the cloud.
The AI interface is grounded on deterministic analytics; provenance runs end to end.
General news and Australian parliamentary material — Senate estimates, Hansard and committee hearings where agencies answer for AI adoption and the workforce. Parliamentary transcripts are a key input data source, with the original source linked on every item.
Parliament — Hansard, Senate estimates and committee hearings. Government — ministers' media and portfolio announcements. News — press coverage. Reports and Evidence appear here when they occur in the feed. Same categories the search uses.
Loading the intelligence feed…