1
0–30 days
- Stand up registry + integrations for top 5 purposes (stroke, PE, sepsis, ambient, CAC)
- Import Overlook research; confirm owners; baseline KPIs
- Enable kill switches and incident capture
The control tower for hospital AI
See every AI in production, enforce guardrails, and turn alerts into outcomes—safely, system‑wide.
Audience: COOs, CMOs, CMIOS, CNOs, service line leaders, AI governance, quality & risk.
Sepsis TTA
-24 min
7d • target -15
Stroke D2A
3.2 min
7d • target 4.0
Ambient Minutes Saved
+18%
30d • watch
Circuit breaker tripped
Status colors are color‑blind safe. Red indicates action, amber indicates watch, green indicates safe, blue indicates information.
One place to see every operating AI
Overlook inventories each AI by purpose (e.g., CT‑PE triage, ICH triage, CAC quant, sepsis alerts, ambient docs, virtual nursing), across all sites and vendors. Leaders get live adoption, safety, equity, and ROI—by system, service line, hospital, and unit.
- Registry of AI Purposes, Sites, Owners, FDA status, Version, Vendor/Internal
- KPI tiles (adoption, time, quality, ROI) with drill‑downs
- Equity and disparity lenses (stratified outcomes)
- Evidence locker (validation, FMEA, IRB, SLAs)
Filter
Service line • Site • Vendor • Risk tier
KPI tiles
Heatmap (purpose × site)
| Purpose | Site | Owner | Vendor | FDA | Version |
|---|---|---|---|---|---|
| CT‑PE triage | Main | Radiology | Aidoc | 510(k) | 2.3 |
| Sepsis alerts | System | CMO | Internal | n/a | 1.8 |
Guardrails you can enforce
Standardize go‑live gates and run‑time safety everywhere—without slowing care.
- Pre‑go‑live gates: indication fit, local validation, equity checks, security & privacy
- Runtime guardrails: Kill switch, Circuit breaker on drift/PPV drops, blue‑green & shadow modes
- Threshold management with audit (who changed what & why)
- Version pinning & update approvals (AIBOM + changelog)
Policy cards
Audit timeline with diffs
Work together where the work happens
Each AI Purpose gets a workspace: runbook, issues, decisions, evidence, and metrics in one place.
Tabs: Runbook • Issues • Decisions • Evidence • Metrics
Decision template and comments
A cadence that sticks
Overlook bakes in the weekly site huddle, monthly safety review, and quarterly value review—plus a clear RACI so nothing falls through the cracks.
- Meeting packs auto‑generated (agenda, metrics, outliers, open issues)
- RACI badges (Business Owner, Clinical Champion, Nursing Lead, IT/Data, Safety, Privacy, Vendor)
- Action items tracked to closure with owners & due dates
Start with what you already run
We preload your control tower with a clean purpose‑level inventory—sepsis/EWS, stroke & PE triage, CAC & VCF detection, ambient docs, virtual nursing, RPM, digital front door, and more—so leaders see value on day one.
Starter Registry (CSV import + vendor mappers)
Outcomes that matter, out‑of‑the‑box
Choose a pack to auto‑wire the right metrics, targets, and visuals.
Sepsis/EWS
Stroke/PE
Imaging quant
Ambient documentation
Virtual nursing/RPM
Digital front door/voice
Built for your stack
Connect EHR, imaging, ambient, voice, RPM, and your analytics cloud.
Integration catalog and connection drawer
Know the risk. Act on it.
Overlook scores each AI by Clinical Criticality × Autonomy × Data Sensitivity × Change Velocity. The risk tier drives required gates, review cadence, and runtime controls.
Risk formula explainer
Policy mapping table by tier
- Clinical decision support & early‑warning (e.g., sepsis risk, deterioration alerts)
- Medical imaging triage/assist (radiology, pathology)
- Patient flow & bed/OR scheduling optimization
- Readmission & length‑of‑stay prediction
- Revenue cycle & prior‑auth automation
Solution playbook
Key AI behaviors to monitor
- Alert precision/recall and time‑to‑alert (e.g., sepsis, deterioration)
- Imaging triage accuracy & escalation logic; turnaround time
- Bed/OR schedule fit rate; conflict resolution behavior
- Prediction calibration & drift (LOS/readmission)
- Prior‑auth automation success & exception handling
KPIs (impact)
Time‑to‑treatment ↓, adverse events ↓, imaging throughput ↑, LOS/readmissions ↓, denial rate ↓, OR utilization ↑.
Overlook impact
- Impact Score: quantifies likelihood of hitting care‑quality & throughput targets per pathway.
- Operating AI Record: provenance for each model (vendor/internal), evidence, approvals, guardrails, runtime behavior and overrides.
- Impact Intelligence: cohort drill‑downs (unit, service line), drift and false‑positive hotspots, suggested playbooks.
- Teams: clinical ops runbooks; accountable owners; escalation loops.
Rollout plan (30/60/90)
2
30–60 days
- Expand to remaining purposes; publish equity dashboards
- Stand up evidence locker; standardize pre‑go‑live pack
- Launch monthly safety review cadence
3
60–90 days
- Automate value ledger and cost allocation
- Align to payer/quality measures; publish systemwide playbooks
- Retire low‑value AIs; scale high‑impact ones