Two hundred bundled actions buy 20 to 40 agent runs per Advanced FUE once the agentic multiplier is applied

The bundled allowance sounds generous stated in actions and stops sounding generous stated in agent runs. Two hundred actions per Advanced FUE is 20 to 40 agentic runs, and since July 2026 use based pricing is the default posture at a cloud renewal rather than an option you can decline.

Prepared by Redress Compliance · August 16, 2026 · SAP advisory. 35 to 45 RISE and S/4HANA Cloud renewals benchmarked, 2024 to 2025.

AI Units are the single consumption currency across Joule and Joule agents, behaving like a prepaid balance drawn down as capabilities run. Base is bundled in RISE and S/4HANA Cloud since July 2025. Premium is a separately purchased add on that requires AI Units.

SAP bundles roughly 200 AI actions per Advanced FUE, pooled across licences. Stated that way the pool sounds ample. Restated against the agentic multiplier it looks quite different, and this is the arithmetic the original framing leaves out.

The pooled allowance is a starting balance rather than a ceiling. A company with a large Advanced FUE count starts with a large pool, and an agent roadmap that automates a handful of workflows on a schedule will clear it well before renewal, because scheduled agents run whether or not anyone is working.

The standard SAP account team framing is that Business AI is already included, so there is nothing to negotiate and buyers should simply switch it on. We disagree, and the mechanism is worth stating precisely because it is not really a pricing problem. In roughly seven out of ten SAP cloud estates modelled across 2024 and 2025, the bundled Base entitlement masked a Premium and AI Units run rate that would have landed as an unbudgeted true up within two renewal cycles.

What makes Base dangerous is not its price, which is zero, but the absence of a meter anyone watches. Interactive Joule sits largely inside the Base entitlement, so finance never sees a bill for chat and nobody stands up consumption reporting for a line that costs nothing. The estate then runs for a year building intuitions about AI usage from a period in which usage was structurally free and structurally light. When Premium is enabled, the workload changes shape at the same moment the billing does, and the organisation is forecasting a metered agentic run rate from a history of unmetered chat.

The multiplier is where a naive budget breaks. When a workflow that used to be one prompt becomes a five step agent, the same business task now draws five to ten times the units, and it runs on a schedule rather than on demand. Those two changes compound: higher draw per execution, and far more executions than a human would ever initiate. An allowance of 200 actions per Advanced FUE, which reads as generous, is 20 to 40 agent runs. On any estate with a real automation roadmap, that is a small number, and the overage rate that follows was set months earlier in an order form nobody costed against agentic volume.

The fix is contractual and it belongs at the current renewal rather than the next one. Treat the action allowance, the AI Unit price, and the agentic multiplier as three separately negotiable terms. Cap consumption in the order form rather than relying on governance to hold it. Instrument Base now, even though it costs nothing, so that when Premium is enabled there is a genuine baseline to forecast from. And govern agent enablement as a change process, because the difference between a pilot and a run rate is a scheduling decision made by someone who does not see the invoice. The metering mechanics sit in AI Units explained, the assistant against agent comparison in Joule agents versus the assistant, and the wider library in the SAP practice.

• Use based pricing is now the default posture for SAP cloud renewals rather than an opt in curiosity, so the AI Units meter is part of the core commercial model on the table.
• Negotiate the action allowance up front, because the effective overage rate you eventually pay depends entirely on how hard the allowance and unit price were argued in the order form.
• Cap consumption contractually. A cap in the paper survives a change of team, a change of roadmap, and a scheduling decision made in an unrelated project.
• Model the agent roadmap before enabling Premium, converting each planned automation into agent runs per period and then into actions, rather than reasoning in headcount.
• Build internal chargeback, because the balance is consumed tenant wide and an unowned pool is drawn down by whoever moves first.
• Instrument Base immediately, so that the baseline exists before the meter matters. This costs nothing and it is the input every later negotiation depends on.

Across roughly 35 to 45 SAP RISE and S/4HANA Cloud renewals benchmarked, three consumption patterns recur:

Overage ran roughly $0.08 to $0.18 per action in the quotes reviewed, and the FUE action allowance was rarely mapped against real usage before signature. Those two facts together are why the first true up is the moment most buyers learn what their agent roadmap costs.

The risk sits in agentic consumption rather than in interactive Joule. Chat is small, predictable, and mostly bundled. Agents read context, plan, call systems, and write results, and each of those steps draws from the same balance.

Watch the briefing · 3:48Optimize the Estate First: The SAP Work That Pays for the NegotiationThe estate work that has to happen before a consumption model gets priced at renewal.

What are SAP AI Units?

The single consumption currency SAP uses to meter Business AI across Joule and Joule agents. Any capability outside the bundled Base tier draws down a balance, so the unit behaves like a prepaid credit and the cost question moves from user count to how much the estate actually runs.

How many agent runs does the bundled allowance really cover?

Roughly 20 to 40 per Advanced FUE. SAP bundles about 200 AI actions per Advanced FUE, and a multi step agent draws 5 to 10 times an interactive prompt, so restating the allowance in the unit you consume gives 20 to 40 runs rather than 200 of anything.

What does an agent run cost once the pool is spent?

Between $0.40 and $1.80. Observed overage ran $0.08 to $0.18 per action in the quotes reviewed, and an autonomous agent draws 5 to 10 actions per run. That per run figure is the one that belongs in a run rate forecast.

What is included in Business AI Base?

The embedded AI features and interactive Joule, bundled in RISE and S/4HANA Cloud since July 2025 at no incremental charge. For many buyers it is enough for a first year, which is precisely why the meter goes unwatched and no consumption baseline gets built.

Why is the bundled tier a risk if it is free?

Because it is uninstrumented. Nobody stands up consumption reporting for a line that costs nothing, so the estate spends a year forming intuitions from usage that was structurally free and light. When Premium is enabled, the workload and the billing change shape at the same moment and there is no history to forecast from.

Why do agents consume so much more than chat?

An interactive prompt is one turn that answers and stops. An agent reads context, plans, calls systems, and writes results, and each step draws from the balance. It also runs on a schedule rather than on demand, so higher draw per execution and far more executions compound.

What changed in July 2026?

Use based pricing became the default posture for SAP cloud renewals rather than an opt in. Business AI consumption is now part of the core negotiation, so the action allowance, the AI Unit price, and the overage rate all sit on the table at renewal instead of surfacing at a true up.

Is the action allowance a ceiling?

No, it is a starting balance, and treating it as a ceiling is the trap. The pool is shared across Advanced FUE licences, so a large FUE count starts large, but scheduled agents draw it down whether or not anyone is working.

Who owns the consumption internally?

Usually nobody, which is the problem. The balance is consumed across the whole tenant rather than per department, so unless you build internal chargeback the pool is drawn down by whichever team automates first and no cost centre feels it.

What should be negotiated rather than accepted?

Three terms separately: the bundled action allowance, the AI Unit price, and the overage rate, plus a contractual consumption cap. Governance alone does not hold a run rate, because the decision that creates it is a scheduling change made by someone who never sees the invoice.