PGŽ Public Finance — transparent, verifiable, optimised
576+
Javnih tijela u RH
+12%/god
Prosj. rast proračuna JLS
2,5%
Ušteda — Base scenario
30 dana
Pilot angažman
PGŽ Budget — Overview 2023–2026
All figures drawn from official PGŽ sources. Click the icon to view the source.
Napomena: Konsolidirani proračun obuhvaća sve proračunske korisnike — škole, bolnice, komunalna i javna poduzeća. Platforma je primjenjiva na sve razine javnog sektora u Republici Hrvatskoj.
How Savings Are Generated
A transparent, reproducible model built on verified data.
PROJEKCIJA — godišnje uštede
~407 tis. € – ~5,09 mil. €/god.
Konzervativni (10% × 1%) do naprednog scenarija (25% × 5%). Pretpostavke na /methodology.
Measurement
Automated ingestion of budget data from CKAN, OpenCity and MFin portals. Standardisation according to the GFI chart of accounts.
Detection
A Z-score statistical model identifies deviations from expected execution. Thresholds: |z| ≥ 2.0 (high), |z| ≥ 3.0 (critical).
Forecasting
Linear extrapolation model with a 95% confidence interval for end-of-year (EOY) execution. Phase 2: Prophet / LSTM.
Action
Prioritised list of anomalies and recommendations for management. Citizen dashboard for transparency.
AI/ML/RL in Practice — PGŽ Context
Four concrete modules with verified applications and measurable outcomes.
Budget Control Tower
Centralizirani nadzor proračunskog izvršenja, KPI monitoring i automatski alerting za sve stavke i razine. Anomalije prikazane u realnom vremenu.
Anomaly Detection
Z-score statistički model identificira neobična odstupanja izvršenja od bazne linije. Prag: |z| ≥ 2,0 (high), |z| ≥ 3,0 (critical).
EOY Forecasting
Prognoza godišnjeg izvršenja (end-of-year) na temelju trenutnih trendova. Linearni model s 95%-tnim intervalom pouzdanosti. Phase 2: Prophet / LSTM.
Operational Optimization
Optimizacija komunalnih usluga — rute, raspored, potrošnja resursa. Reinforcement Learning modul za kontinuirano unapređenje operativa.
Pilot in 30 Days
A structured, low-risk onboarding process with clear deliverables for each week.
Tjedni 1–2
Onboarding & ingestija
Pristup podacima, setup data warehousea, automatski ingest CKAN / OpenCity izvora.
Tjedni 3–4
Detekcija & dashboardi
Aktivacija anomaly detection modela, prvi BI dashboardi s KPI prikazom i anomaly alertingom.
Tjedni 5–6
Prognoze & validacija
EOY forecast modeli, backtesting na prethodnim podacima, validacija s naručiteljem.
Tjedan 7
Predaja & prezentacija
Konačni izvještaj, prezentacija nalaza upravi, preporuke za fazu 2 i skaliranje.
Ready for Transparency
First demonstration on your own data within 30 days. No cost, no vendor lock-in.
Projection based on the savings model. Assumptions and methodology are available on the Methodology page. Actual results depend on data quality and operational implementation.