Joule Pricing & Consumption

As of 2026-05-29

Joule pricing/consumption is the FinOps of AI on SAP — generative AI is the workload whose cost surprises most sharply, since inference scales with usage and ambition (many users, and especially autonomous agents looping). Consumption-based like BDC (M032); units/bundling from the SAP contract — model drivers, never invent a per-call price (§0i). Drivers: adoption breadth, interaction complexity (narrative/agentic > simple query), agent autonomy (M049 multi-step = highest consumption), grounding retrieval (M047, non-negotiable). Value AND cost scale with ambition → make the ROI case per use case, not platform-wide. Govern with the same FinOps loop (M043): visibility (M045), optimisation (not every workflow needs an agent), guardrails (budgets; agent loops especially need limits + human-in-the-loop M049). ROI anchors (EY ~30%, KPMG ~20%, Daimler 10→40%+, M046/§17.11) are those firms' results, not guaranteed (§0i). Goal: each unit of consumption buys proportionate value — neither runaway nor under-investment.