nuMetrix · First Use Case
I/O Imbalance
Measurement
Material inflow vs. patient-level consumption.
Where the gap is, why it exists, and what to do about it.
SZO (Spitalzentrum Oberwallis) · Brig & Visp
Data: 2023–2024 · 5,393 materials · CHF 1–99 band
The Challenge
From 40+ use cases,
one was chosen as the nucleus
The Agreement
- Measure the ratio of material consumption (Output) against warehouse-to-ward transfers (Input)
- CHF 1–99 — the middle ground where high-volume untracked leakage occurs
- Fully automated evaluation against real data (SZO) and synthetic tenants
- Revenue Leakage is four iterations later — I/O must be stabilized first
Why This First?
- Items over CHF 1,000 are already tracked manually
- Items under CHF 1.00 are irrelevant
- The CHF 1–99 band is where nobody looks — and where volume hides
- Count vs. cost: same coefficient (same price on both sides)
Key insight: Structural data issues (e.g., mapping 4-digit location codes to 5-digit cost centers) must be resolved before tackling billing tariffs.
The Approach
End-to-end, from CSV intake to verdict
Bronze
CSV intake
CSV intake
→
Silver
Validation
Validation
→
Gold
I/O model
I/O model
→
Probes
Anomaly detection
Anomaly detection
→
Hypothesis
Verdict
Verdict
The Model
- Input = SUM(quantity) of TRANSFER movements (warehouse → ward)
- Output = SUM(quantity) of patient-level usage
- Grain: material × month
- Filter: standard_price CHF 1–99
- Overlap detection: auto-exclude periods without both-sided data
The Probes
- I/O coefficient ratio — flags >20% deviation from 1.0
- I/O trend — flags worsening drift over time
- Weekly variants — same logic, finer granularity
- Population filter — only materials with both transfer AND usage
Severity: high (>50% imbalance), medium (>20%)
nuMetrix
First Discovery
Three Populations
Materials in the CHF 1–99 band fall into three cleanly separated groups.
25%
43%
32%
Transfer-only (1,323)
Both present (2,343)
Usage-only (1,727)
Transfer-only — 25%
Delivered to wards, never scanned at patient level.
Disinfectant wipes, isolation gowns, garbage bags, hand towels. Ward consumables. Nurses take from stock without scanning.
Usage-only — 32%
Patient consumption recorded, zero warehouse transfers.
Almost entirely oncology medications — Xtandi, Kisqali, Venclyxto, Brukinsa. Dispensed directly from pharmacy.
Both present — 43%
The only population where I/O coefficient is meaningful.
Overall coefficient: 0.47
Half of what enters the ward never gets documented at patient level.
Implication: The probe must filter to "both present" only. Transfer-only and usage-only are structural, not anomalies. The raw CHF 30.6M shrinks to CHF 4.3M after this sharpening.
The Findings
After sharpening: overlap period, both-present only
1,875
Findings
CHF 4.3M
Value at risk
923
Materials (≥100 transfers)
0.47
Overall coefficient
| Category | Materials | % | Avg. coeff. | CHF at stake |
|---|---|---|---|---|
| ■ Under-recorded (<0.80) | 657 | 71% | 0.37 | CHF 2.20M |
| ■ Balanced (0.80–1.20) | 220 | 24% | 0.93 | CHF 0.25M |
| ■ Over-recorded (>1.20) | 46 | 5% | 2.77 | CHF 0.32M |
24% are perfectly balanced — this proves accurate tracking is possible in this hospital. The question is what separates balanced from unbalanced.
The Split
What separates balanced
from unbalanced?
Balanced (~1.0)
Every unit delivered is documented at patient level
| Ringerfundin 500ml | ~1.0 |
| NaCl 0.9% Ecobag | ~1.0 |
| Cefuroxim 1.5g | ~1.0 |
| Esmeron 50mg | ~1.0 |
| Sufenta 50mcg | ~1.0 |
| Set Desinfection | ~1.0 |
OR / anesthesia / imaging
Per-patient documentation by protocol
Per-patient documentation by protocol
vs
Under-recorded (<0.50)
Less than half reaches patient records
| Ringerfundin 1000ml | 0.31 |
| Tena Pants | 0.13 |
| Kenacort A 40mg | 0.19 |
| EasyPump (infusion) | 0.32 |
| Octenisept | 0.23 |
| Set sondage vésical | 0.30 |
Ward consumables
Taken from cabinets without scanning
Taken from cabinets without scanning
The Ringerfundin split: 500ml (anesthesia, per-patient) = coefficient 1.0. 1000ml (ward use, from cabinet) = coefficient 0.31. Same product family, same hospital, different workflow.
nuMetrix
Root Cause
The 0.47 coefficient is an
architectural constant
Warehouse
(TRANSFER events)
(TRANSFER events)
↓
Ward stock / floor stock
↓
Scanned
at patient level
at patient level
Coeff ≈ 1.0
24% of both-present
24% of both-present
↓
Not scanned
(ward use)
(ward use)
Transfer-only or
coeff < 0.50
coeff < 0.50
↓
Direct from
pharmacy
pharmacy
Usage-only
32% of all materials
32% of all materials
- OR / anesthesia / radiology: tight coupling (coeff ~1.0). Every item documented by clinical protocol.
- Wards: loose coupling (coeff 0.13–0.50). Bulk supplies consumed without scanning. By design.
- Pharmacy / specialist units: no coupling (usage-only). Oncology drugs dispensed directly. Expected.
- Stable over 24 months — Ringerfundin 1000ml stays at 0.26–0.37. A process feature, not an incident.
Automated Diagnosis
Process failure:
inventory ↔ patient-usage disconnect
What the System Found
- Hypothesis: hyp_io_imbalance — Confirmed
- Evidence score: 5 probes evaluated, strong cross-probe agreement
- Diagnosis: process_failure / inventory_patient_usage_disconnect
- Confidence: 60% base + 15% boost (trend) + 10% boost (material health)
Interpretation: Inventory transfers significantly exceed patient-level usage for 1,875 materials. The I/O coefficient indicates a disconnect between what enters the ward and what is documented as used.
Evidence Chain
Primary
I/O coefficient ratio probe
Support
I/O coefficient trend probe
Support
Weekly I/O coefficient
Context
Weekly trend
Support
Material health assessment
All automated. No AI at render time.
Template-based interpretations in EN / DE / FR.
Going Deeper
Ward-level analysis
Clinical wards and logistics cost centers are structurally disjoint. We bridged them via service mandates.
The Bridge
- Clinical ward (requester) ↔ Sachkosten CC (provider)
- 222 service mandates, 134 unique ward–provider pairs
- 99% of usage CCs have a mandate
- But only 11 of 47 providers have mapped transfer data
- 23 wards routed to 7 providers — 33.4% of usage covered
Gap is on the Brig side (H3 master data gap). Visp has 90% coverage.
Ward Results
20,134
Rows computed
23
Wards analyzed
CHF 909K
Over-transferred
CHF 611K
Under-transferred
Top providers: V-Notfallstation and V-IPS serve 12 and 9 wards respectively, with strong positive deltas.
nuMetrix
Treatment
Recommended Actions
Immediate
- Focus on highest-volume materials with I/O < 0.5 — Ringerfundin 1000ml, Tena, Kenacort, EasyPump. These 8 materials alone account for CHF 300K+ in gap.
- Audit ward scanning workflow for those materials — identify where patient-level documentation breaks down.
- Use balanced materials as benchmark — the 220 balanced items prove the system works. Ask: what do OR materials have that ward materials don't?
Next Phase
- Classify materials by tracking expectation — OR-mandated (expect ~1.0), ward-optional (expect <1.0), pharmacy-direct (expect usage-only).
- Complete Brig mapping — when H3 master data is extended, ward coverage goes from 33% to ~80%.
Watch For
- Trend changes — A stable 0.30 is a process feature. A drop from 0.90 to 0.30 is an incident. Weekly probes detect this automatically.
- Over-recorded items — Freestyle glucose sensors (coeff 8.3), Novorapid insulin (coeff 21.2). These bypass warehouse. May indicate unbridged supply routes.
What this is NOT about: Revenue leakage. That requires tariff mapping and insurance billing logic — four iterations later per the agreed roadmap.
What this IS about: Making the invisible visible. CHF 4.3M in material value flows through wards with no patient-level documentation. Now we can see it, track it, and act on it.
What We Built
Fully automated. Every number is queryable.
Gold Models (6)
- ✓ I/O coefficient (monthly)
- ✓ I/O coefficient trend (3-month rolling)
- ✓ I/O by cost center
- ✓ I/O by ward (service mandate bridge)
- ✓ Weekly coefficient + weekly trend
Probes (4)
- ✓ Ratio probe (monthly + weekly)
- ✓ Trend probe (monthly + weekly)
Analytics Pyramid
- ✓ Hypothesis: hyp_io_imbalance (5-probe chain)
- ✓ Diagnosis: inventory process gap
- ✓ Platform layer: all models unioned
- ✓ Evidence page: drill-down dashboard
- ✓ Explorer: hypothesis detail + presentation
- ✓ Interpretations: EN / DE / FR
Documentation
- ✓ Spike design doc
- ✓ CC mapping analysis
- ✓ Deep analysis (3 populations)
- ✓ Ward model design
nuMetrix
CHF 4.3M
in undocumented material flow
We know…
- Exactly which materials are affected (1,875)
- Why the gap exists (ward scanning workflow)
- Where it happens (23 wards, 7 providers)
- That tracking works for 24% of materials
- That the gap is stable (not worsening)
Next steps…
- Audit top-8 materials (CHF 300K+ gap)
- Classify materials by tracking expectation
- Complete Brig site mapping
- Cross-hospital comparison (when data arrives)
- Revenue leakage (iteration 5, after I/O is stable)
The core engine is running.
Now we plug in the next use case.
nuMetrix · February 2026