Data analytics licensing: consumption versus the seat.

01 Key findings

1. Two commercial models, not one. Warehouses and lakehouses meter consumption (Snowflake credits, Databricks DBUs); BI tools meter seats (Tableau, Power BI, SAC). Each fails commercially in a different way and needs a different control.

2. Consumption elasticity is the primary cost risk. Auto-scaling warehouses can burn credits at 10× a minimum-size warehouse, and 20–30% of Snowflake credit consumption is typically pure waste — idle, oversized or dev-on-prod compute.

3. Platform expansion multiplies DBUs silently. Organisations that move Databricks from data engineering into ML training can see DBU consumption rise four to six times with no new users, driven by GPU-accelerated instance types.

4. Seat misclassification cuts both ways. Tableau Viewers who need Explorer capability create compliance risk; Explorers who only ever consume dashboards are pure overpayment. Both hide inside the same licence count.

5. Bundling is leverage and trap. Tableau via a Salesforce EA, Power BI via a Microsoft EA and SAC via an SAP renewal all gain pricing latitude — but treated as a passive add-on, the bundled product typically costs 20–30% more than a benchmarked standalone deal.

02 Licensing models compared

The single most useful lens on analytics cost is the billing meter. Consumption platforms charge for compute you run; seat platforms charge for users you license. The committed-discount lever and the dominant expansion risk follow directly from which meter applies.

03 Snowflake compute economics

Snowflake separates compute from storage — a decision that makes cost highly elastic. Compute is charged in credits consumed by virtual warehouses that run queries; storage is charged per terabyte of compressed data. On-demand credits run roughly $2–$4 depending on cloud and region; pre-purchased capacity in 12- or 24-month blocks falls to $1.50–$2.50 for large commitments.

That elasticity is commercially dangerous without governance. Auto-scaling warehouses spin up extra clusters as concurrency rises, protecting user experience while generating credits at 10× a minimum-size warehouse. Dev and test warehouses left running overnight consume credits for no business value, and loading jobs run on oversized warehouses are direct waste.

Negotiation centres on committed capacity. On-demand is never the right model for a committed consumer — the discount versus on-demand is 25–40%, and the break-even is typically four to five months of consumption. Snowflake’s enterprise team has meaningful flexibility above $500K annually, including non-standard storage discounting and rollover for unused pre-purchased credits. A credible evaluation of BigQuery, Redshift, Athena or Synapse unlocks a further 15–25% unavailable in an assumed-renewal conversation.

04 Databricks DBU model

Databricks meters Databricks Units (DBUs) — a compute unit reflecting cluster processing capability per hour. Workload types consume DBUs at different rates, and each platform expansion into ML, MLOps, Unity Catalog governance and generative AI (DBRX, Mosaic AI) tends to add new SKUs at premium DBU rates. Move from data engineering into GPU-accelerated ML training and consumption can multiply four to six times with no new users.

Negotiation centres on committed DBU packages: roughly 20–30% off on-demand for 12-month commitments and 30–40% for 24-month terms. The decisive points are the DBU rate per workload type (all-purpose versus jobs versus SQL versus ML), the minimum consumption floor relative to realistic projected usage, and rollover provisions for unused DBUs within the committed period.

05 Seat-based BI platforms

Tableau, Power BI and SAP Analytics Cloud all meter users rather than compute — and all three sit inside a larger vendor relationship that supplies leverage and complication in equal measure.

Tableau, acquired by Salesforce in 2019, now spans legacy Desktop/Server and the Tableau Cloud SaaS offering Salesforce is steering customers toward. Its Creator, Explorer and Viewer tiers map to authoring, interaction and consumption — but misclassified users (Viewers needing Explorer rights, Explorers who only consume) create compliance risk or overpayment. On a Salesforce EA, Tableau can be pulled into a broader multi-product negotiation, though Salesforce account teams often treat it as secondary to CRM and Service Cloud renewals.

Power BI is the default in Microsoft estates because Pro is bundled into M365 E3/E5. Cost escalates at the Premium boundary: Premium Per User adds roughly $20/user/month above Pro, while Premium Capacity — priced by dedicated compute block — makes sense above ~500 report consumers. Standalone Premium Capacity carries stiff list pricing; folded into a Microsoft EA alongside Fabric, Azure and Teams it gains real latitude. See our Microsoft Fabric licensing guide for Power BI’s evolution within the Fabric family.

SAP Analytics Cloud is user-based across BI, planning and viewer tiers and is positioned as the native analytics layer for S/4HANA estates. Negotiated as an add-on to a broader SAP renewal — where SAP holds maximum leverage — customers typically pay 20–30% more than a standalone SAC deal with live competitive alternatives. See our SAP Analytics Cloud licensing guide and, for Tableau ownership in context, the Salesforce licensing guide.

06 Consumption traps

The costliest analytics mistakes are structural, not negotiated — they recur every billing cycle until governance closes them. Three traps account for most avoidable spend.

07 Optimisation framework

Beyond individual negotiations, a governance approach produces sustained savings that one-off discounts cannot. Four mechanisms carry most of the durable result.

08 Our recommendations