Benchmark Health

Benchmark Health runs behavioral health clinical operations on a data estate that pulls together clinical, operational, and patient-experience systems — from its core clinical records to call activity, workforce data, and patient feedback. Leadership relies on these numbers to manage clinician capacity, billing, and patient outcomes.

When the engagement began, that data moved through a patchwork of hand-built cloud jobs and loading logic buried inside dbt hooks. It was brittle to operate, hard to monitor, expensive to run, and difficult to trust. Critical loading logic ran invisibly on every command, key pipelines were managed by disparate tools with no central view, and there was no end-to-end lineage from raw ingestion to final models. Midwest Data was brought in to turn this into a dependable, governed, analytics-ready platform.

The business needed reliable numbers, but the underlying plumbing couldn't yet support confident decision-making:

Operational blind spots. Fragmented orchestration meant failures could pass silently and lineage was untraceable — nobody could see the full path from a source file to the dashboard that depended on it.

Data you couldn't fully trust. Joins on non-unique keys (phone numbers, emails) caused row duplication, and missing primary keys plus loose LEFT joins let orphaned and test records leak into core models.

Bespoke, non-repeatable work. Onboarding a new clinical practice and producing capacity and billing reporting were one-off efforts rather than a repeatable process.

Cost and compliance pressure. Full rebuilds were slow and Snowflake-expensive, and a healthcare data estate needed a stronger PHI-handling posture.

Impact came quickly: within the first few weeks the platform was already noticeably more reliable and trustworthy, and the team had clear visibility into data they'd previously had to take on faith. From there, Midwest Data delivered across three pillars — modernizing orchestration, fortifying data integrity, and raising engineering excellence — then built the analytics layer the business runs on, and folded the whole estate into Benchmark's central codebase.

A single, observable orchestration plane. We consolidated the scattered legacy jobs into one Dagster deployment that owns scheduling, dependencies, and alerting, with dbt for modeling and Snowflake as the warehouse. Every source across the clinical, operational, and patient-experience systems was rebuilt as a version-controlled, observable Dagster asset writing raw data to GCS and loading into Snowflake. By mapping dbt source references to Dagster asset keys, the full DAG is now traceable from ingestion to final model, pipelines materialize automatically when upstream data lands, and the brittle legacy orchestration was retired.

Data integrity as a first-class concern. We attacked the systemic causes of data-quality issues rather than patching symptoms: deduplication and surrogate-key fixes across telephony, practitioner, and care-team models, and referential-integrity fixes so test and orphaned records stop flowing downstream. Automated dbt tests and pull-request CI now catch issues before they reach reports, and repo-wide SQL linting keeps the codebase consistent.

A leaner, compliant, maintainable codebase. We pushed transformations upstream and removed redundant models to keep the layers DRY, re-architected how one of the largest sources is read — collapsing hundreds of practice-specific source tables into a small set of shared external tables — and added a workflow so onboarding a new clinical practice is now routine rather than a project. We introduced column-level PHI masking in Snowflake to support compliance, and implemented dbt "slim" CI so builds only run what changed, cutting both Snowflake cost and CI time.

The analytics layer the business runs on. On the foundation, we built the reporting clinical operations actually depends on: care-manager productivity and capacity planning, billing by patient and month, patient and referral lifecycle metrics, intake and outcome tracking, longitudinal clinical-assessment history, and patient-experience (NPS) reporting. We also stood up a secure, read-only analytics service so teams can self-serve insights.

Benchmark Health saw meaningful improvement within the first few weeks, and over the course of the partnership went from fragile, hard-to-trust data plumbing to a modern, governed analytics platform the business runs on every day — not incremental tinkering, but a new foundation.

From “hope it ran” to self-monitoring. Data flows end-to-end on a predictable cadence, failures surface proactively, and the team is freed from manual firefighting.

One trusted source of truth. Every clinical and operational source is unified into a consistent, well-tested analytics layer with single definitions for patients, encounters, referrals, and providers.

Decisions leadership can stand behind. Capacity, billing, and patient-outcome reporting now run on trustworthy dashboards instead of disconnected silos.

Leaner and ready to scale. Quality and compliance went up while build cost and time went down, and the entire estate now lives in Benchmark’s central codebase for long-term ownership — including a new team member who joined Benchmark from the engagement.

Business impact. Beyond the engineering, the work changed what the platform costs to run and what the business can do with it.

Runs lean. Self-healing pipelines, automated testing, and agent-assisted routines for recurring tasks mean the platform is now maintained by a single analytics engineer.

Lower cloud spend. Retiring the legacy orchestration and building only what changed in CI reduced Snowflake compute and build time.

Protected revenue. Corrected billing logic and trustworthy capacity reporting support accurate reimbursement and clinician utilization.

Faster to grow. Onboarding a new clinical practice went from a bespoke project to a repeatable workflow.