Nabla-X
ServicesCase StudiesContact
Healthcare·8 minute read

How we recovered $1.0M in hidden revenue leakage

A multi-facility healthcare network used QueryMind and DataSentinel to detect and explain revenue drift across billing systems—before it compounded into regulatory exposure.

Discuss Your Environment
$1.0M
Revenue recovered in Q1
35%
Reduction in under-coding errors
24h → 0
Real-time diagnostic monitoring

Healthcare context — why this is hard

Fragmented systems

Clinical (EHR), billing, lab, PACS, and external payer feeds often disagree on encounters and timestamps. Reconciling these reliably requires lineage, mapping, and conservative joins.

Coding & documentation variability

CPT/DRG use differs across facilities and coders. Small documentation differences can produce large revenue changes. Detection requires semantic matching across clinical notes and invoice records.

Payer & contract complexity

Different payer contracts apply different remittance rules; revenue shifts may be payer-driven rather than clinical. Our solution surfaces payer-driven vs clinical-driven causes.

Problem statement

Finance noticed recurring drops in revenue for several service lines. Surface-level metrics (totals) were available, but leadership needed causality: which service, which site, which documentation gap, and what payer behavior caused the change. Analysts were reluctant to run complex joins for fear of producing misleading results because some sources (e.g., lab, outpatient) were partial or delayed.

Symptoms
  • Revenue down 8% month-on-month in elective procedures
  • Inconsistent encounter counts between EHR and billing
  • Delayed lab and PACS feeds causing incomplete cohorts
Constraints
  • Regulatory & payer compliance required for audits
  • No risky wide joins across sensitive PHI without controls
  • Need reproducible, auditable query plans and explanations

Technical approach

1. Intent capture (diagnostic)

We classify the user's question as diagnostic (causal) rather than an aggregation. This changes downstream logic: the planner avoids naive aggregations and instead reasons about segments, baselines, and required evidence to support causal claims.

2. Semantic grounding

Map natural-language entities (e.g., "revenue", "elective procedures", "site X") to canonical tables, columns, and glossary terms. Use column lineage and historical validated queries to avoid ambiguous mappings.

3. Safe query planning & verification

The planner decomposes analysis into sub-queries, estimates cost, computes blast radius, and consults DataSentinel signals for data completeness and trust. High-risk plans (wide cross-facility joins, cross-PHI joins) are flagged and either rewritten into safe alternatives or require explicit approval.

4. Guarded execution & observability

Execute with row limits, cost ceilings and timeouts. Every execution records: intent, plan, data sources, result shape and trust scores — all available in a searchable execution log for audit and model feedback.

5. Explanation & learning loop

Return narrative explanations (what changed, what was excluded, alternative interpretations). Confirmed corrections (only from authorized reviewers) are used to update metadata and mappings — not to blindy retrain models — preserving correctness.

Implementation details & integrations

Data sources

  • EHR: encounter, ADT, clinical notes
  • Billing: invoices, claim edits, remittance
  • Payer feeds: adjudication timing & reason codes
  • Operational: OR schedules, staffing, supply usage

Platform & runbook

  • Data ingestion (CDC/ETL) into governed warehouse (Snowflake/Redshift/BigQuery)
  • Metadata store with column lineage & business glossary
  • Query planner service (stateless) that calls execution engine via API
  • Observability: execution log, trust scores, and health dashboards
Security & compliance: PHI is never exfiltrated to third-party LLMs. Any ML that touches PHI runs within your VPC on approved runtimes. Audit trails and RBAC are enforced for correction and learning loops.

Client

Multi-facility healthcare provider network serving 250,000+ patients annually across clinical, billing, lab, and payer systems with strict HIPAA compliance requirements.

The Challenge

Fragmented systems

EHR, billing, lab, and PACS systems disagreed on encounters and timestamps. Reconciling reliably required lineage, mapping, and conservative joins.

Revenue drops

8% month-on-month decline in elective procedures revenue with inconsistent encounter counts and incomplete cohorts from delayed feeds.

Coding variability

CPT/DRG usage differed across facilities. Small documentation gaps produced large revenue changes requiring semantic matching.

Compliance constraints

No risky wide joins across sensitive PHI without controls. Required reproducible, auditable query plans for regulatory compliance.

Our Solution

01

Intent capture

Classify questions as diagnostic (causal) rather than aggregation, changing downstream logic to reason about segments and required evidence.

02

Semantic grounding

Map natural-language entities to canonical tables using column lineage and historical validated queries to avoid ambiguous mappings.

03

Safe query planning

Decompose analysis into sub-queries, estimate cost, compute blast radius, and consult DataSentinel for completeness and trust.

04

Guarded execution

Execute with row limits, cost ceilings and timeouts. Record intent, plan, data sources, and trust scores for audit trails.

05

Explanation loop

Return narrative explanations with confirmed corrections used to update metadata—preserving correctness over blind retraining.

Results

$1.0M
Recovered revenue via targeted remediation
Real-time
From 24hr delays to instant diagnostics
35%
Reduction in under-coding errors

These numbers reflect actual impact from a 3-month deployment. Results depend on data maturity, scope, and remediation workflows. We always run discovery to produce realistic estimates before any PoC.

Technology Stack

QueryMind

Intent-first analytics with safe, schema-aware SQL generation

Learn more

DataSentinel

Continuous data reliability monitoring and trust scoring

Learn more

RootSense

Causal analysis for explainability and recommendations

Learn more

Ready to prevent revenue leakage?

Let's discuss how Nabla-X can help you detect and explain hidden patterns in your healthcare data.

Schedule Architecture ReviewView All Case Studies
© 2025 Nabla-X. Production-grade AI observability for regulated environments.