Most analytics platform decisions are made against a vendor quote. The quote is rarely wrong; it is just incomplete. By the time a platform reaches its second renewal, the line items missing from that original quote often exceed it. Below are the costs that consistently get missed in TCO models we are asked to review, and the questions that surface them before the contract is signed.
Egress and inter-region traffic
Compute pricing is legible. Egress is not. A modeling layer in one cloud reading from a warehouse in another, a BI tool in a third region pulling extracts, a reverse-ETL job pushing into a SaaS in a fourth — every hop is metered, and the meters are not on the invoice you negotiated. Mature platforms either co-locate compute with storage or budget egress as a first-class line item. The rest discover it as a quarterly variance.
The diagnostic question: what is the dollar value of cross-region and cross-cloud traffic generated by the platform on a representative day? If no one can answer in under a week, that is the cost.
Model and pipeline rebuild time
A schema change upstream is a labor event downstream. The TCO model that prices “an analytics engineer” at a fully-loaded rate is not wrong, but it usually omits the rebuild tax: how often models are restructured because a source migrated, a dimension was renamed, a metric definition changed, or a vendor deprecated an API. On platforms with weak lineage and weak contract enforcement, rebuild work can consume 20 to 35 percent of analytics engineering capacity. That is a recurring cost masquerading as project work.
Reverse ETL and activation infrastructure
The warehouse-as-source-of-truth pattern only delivers if the data leaves the warehouse and arrives in operational systems. Reverse ETL tooling, the identity resolution that sits in front of it, and the operational monitoring that catches a stalled sync before a campaign goes out with stale audiences — these are infrastructure, not optional. They are also routinely absent from the original platform business case, then bolted on for an additional six- or seven-figure annual run rate.
Observability and data quality
A platform without observability fails silently. Incidents are discovered by stakeholders, which is the most expensive way to discover them. The cost of observability is not the tool license; it is the engineering time to instrument freshness, volume, schema, and distribution checks against the assets that matter, and to route alerts to humans who will act. Budget this explicitly. Platforms that treat observability as a Q3 initiative tend to absorb a corresponding loss of trust in Q1 and Q2.
Governance, access review, and audit response
Every platform inherits the regulatory surface of the data flowing through it. Quarterly access reviews, audit evidence collection, lineage attestations for regulated metrics, retention enforcement, and the engineering work to make any of this automatable rather than manual — these are not free, and they grow with headcount and data volume rather than with platform spend. A useful proxy: how many engineering hours did the last SOC 2 or internal audit consume?
The integration tax
A modern data platform is rarely a single vendor. It is a warehouse, a transformation layer, an orchestrator, a catalog, an observability tool, a BI tool, a reverse-ETL tool, and a semantic layer, each with its own auth, its own metadata model, and its own upgrade cadence. The work to keep these in coherent alignment — shared identifiers, consistent naming, shared lineage — is continuous. Underweighting it produces a platform that is technically operational and practically incoherent.
The exit cost
The cost most often omitted from TCO is the cost of leaving. Proprietary SQL dialects, vendor-specific UDFs, semantic models defined in a tool’s DSL, dashboards built against a query engine that only exists inside one product — each of these raises the cost of a future migration. This is not an argument against any specific vendor. It is an argument for knowing the number, putting it in the model, and revisiting it annually.
What to do with this
Build the TCO model with these line items present even if the values are estimates. An estimate forces a conversation; an omission ends one. The platform that wins on a complete TCO is almost always different from the platform that wins on a vendor quote, and the gap between those two answers is the value of doing the work.