Decision criterion
The code generator is free. The orchestrator is not.
When you compare Data Vault platforms, the “orchestration” line is often missing from the quote — because it's assumed to be covered by Airflow, dbt or an in-house scheduler. Yet that line very much exists in your budget: it's called half or a full FTE, every year, to maintain a piece of infrastructure that produces no business value.
- No clusters or schedulers to operate
- Dependency graph inferred from the model
- A single support chain in case of an incident
- Lineage from the source field all the way to the mart, with no gaps
Hidden costs
What an Outsourced Stack Truly Adds
None of these costs appear in a license comparison. All of them appear in your operations.
Infrastructure maintenance
Scheduler, workers, metadata database, upgrades, security: a self-managed orchestrator is a product in its own right that someone has to operate.
DAG Writing and Derivation
Loading dependencies are rewritten manually. With each model evolution, they must be updated — and an oversight leads to an inconsistent silent load.
Fragmented diagnostics
When a load fails at 3am, the team has to search across the modeling tool, the orchestrator and the target platform. Three interfaces, three logs, three support channels.
Rare skills
Maintaining Airflow correctly requires a platform profile. This profile is expensive, hard to recruit, and their departure creates an immediate operational risk.
Uncontrolled runtime cost
Poorly tuned parallelism multiplies your cloud bill. beVault computes the optimal scheduling from the model, with no manual tuning.
Comparative
Two architectures, identical scope
We deliberately assume the most favorable case for the outsourced stack: excellent tools, well mastered.
| Critère | beVault (integrated orchestration) | Stack Airflow / dbt |
|---|---|---|
| Initial go-live | A Few Days | Several weeks of installation and scoping |
| Annual maintenance effort | Included in the platform | 0.5 to 1 dedicated FTE |
| Model ↔ Execution consistency | Guaranteed by design | To maintain manually |
| Incident recovery | At the task level, without global re-run | Depends on DAG design |
| Support | A single point of contact | Community + multiple publishers |
Nuance
When the Outsourced Stack Remains Relevant
We do not claim that Airflow is a bad choice. If you already orchestrate hundreds of heterogeneous processes outside the data scope, and a platform team exists, sharing it makes sense. In that case, beVault fits in: our loads can be triggered and monitored from your enterprise orchestrator. What we dispute is building that infrastructure solely to run a Data Vault.
Next step