Seat-to-Consumption Pricing Transition

Why this is in the vault

Five-plus vault documents over three months independently converge on the same pattern: enterprise software vendors are moving billing away from per-seat licenses toward usage/consumption meters, and that move is a one-way door with a specific, now well-documented failure mode on retention metrics. This has come up so often, from so many angles (NDR benchmarking, Agentforce's actual credit mechanics, Databricks sales-comp redesign, a general usage-vs-subscription survey, and now a dedicated trajectory brief), that it is a recurring concept rather than a one-off observation, and it bears directly on how RDCO should think about pricing its own services and Squarely/MAC surfaces.

The core finding, synthesized across sources: consumption pricing raises the ceiling and removes the floor. Under seat licenses, revenue is sticky (a purchased seat renews whether or not it's used); under consumption, revenue tracks actual usage, so it can expand faster in a genuine expansion case but also compresses immediately when usage drops — no renegotiation lag to soften the blow. [[2026-08-30-saas-seat-to-consumption-ndr-trajectory]] frames this precisely: the "130% club" of high-NDR SaaS names has effectively collapsed (P75 NDR fell from ~130% to ~116%), and consumption-native names bled hardest — Snowflake compressed 53 points (177%→124%) because "customers throttled usage faster than license renegotiations."

Three underlying mechanisms recur across sources:

• The disclosure problem: vendors that don't want to show the compression simply stop reporting true NDR, or swap it for a softer metric ([[2026-06-21-mostly-metrics-ndr-benchmarks]]).
• The accounting problem: "consumption ARR" is a trailing run-rate annualization, not contracted revenue, so it de-annualizes exactly as fast as it annualized ([[2026-06-04-mostly-metrics-consumption-based-arr]]).
• The repackaging problem: nobody actually kills seats outright. Salesforce's Agentforce Flex Credits still price around a seat-shaped floor ($125/user/mo to $550+/user/mo pre-loaded with credits) with a consumption meter bolted on top — the honest shape is "seat floor + consumption ceiling," not a clean replacement ([[2026-06-17-agentforce-flex-credits-consumption-pricing]]).
• The organizational problem: switching models rewrites sales comp, revenue recognition, and forecasting cadence — Databricks' own sales-comp redesign is cited as evidence this is a genuinely painful internal migration, not just a billing toggle ([[2026-07-14-mostly-metrics-databricks-consumption-sales-comp]]), and CFOs at Confluent, Figma, and Couchbase independently describe it as a one-way door ([[2026-08-18-mostlymetrics-usage-based-pricing-one-way-door]]).

Mapping against Ray Data Co

RDCO's own services-pricing thinking ([[2026-05-20-services-pricing-model-for-rdco-future]]) sits in the same design space this concept describes, and the lesson generalizes directly: a pure consumption model on any RDCO surface (Squarely, MAC, future agent-delivery services) would look great during genuine engagement growth but offers no cushion when usage dips — there is no seat-renewal lag to hide behind. The "seat floor + consumption ceiling" hybrid that Agentforce actually shipped, rather than the "seats are dying" headline framing, is probably the more honest reference pattern if RDCO ever prices a metered surface: keep a floor that survives a slow month, let the ceiling track real usage. The disclosure-gaming pattern (softening or hiding NDR when it turns unflattering) is also a useful tell to watch for when evaluating phData or Anthropic-partner-ecosystem vendors as comparables — a vendor's silence on a metric is itself the signal, per [[2026-06-21-mostly-metrics-ndr-benchmarks]].