Details and points of attention regarding the consumption of SAP Business AI
In this article, we take a closer look at what AI Units are, how their consumption is measured, and what points of attention need to be considered when defining the AI-related budget, both during contract signature and during credit sizing. AI Units are the currency SAP uses to charge for the consumption of its artificial intelligence portfolio: Joule (SAP's conversational assistant, which navigates applications and automates tasks in natural language), agents, and intelligent services embedded in its cloud applications. As SAP itself defines them on the official pricing page, they are a form of virtual currency that customers purchase to activate and use Premium AI features: consumption-based credits, flexibly usable across different solutions and not tied to a specific product or line of business, which customers draw from a centralized pool. They are purchased on an annual basis and, if unconsumed, expire after 12 months. The model is structured on two levels, as shown in the aforementioned page: Joule Base — navigation, simple functions — is free and does not require AI Units; everything beyond that, Premium AI, is paid for by purchasing a separate pool of AI Units:
One year after the introduction of the new commercial model (July 2025), the first customers who signed under the new conditions are now reaching their first annual contract renewal: it is the right time to take stock of how this currency works and what to check during the renewal phase. AI Units are therefore added to other existing SAP metrics: FUE (Full Use Equivalent, the metric on which RISE is sized: not a head count, but a value weighted according to each user's professional role), BTP credits, capacity units, without replacing them: it is a fourth pricing model, featuring a consumption-based logic that must be monitored carefully.
SAP flexibly applies two pricing models for Premium AI, depending on the nature of the feature:
Some mechanisms are common to both variants, only the numbers change: the included requests are shared among all users assigned to the same package, and unused capacity from one user compensates for the higher consumption of another. Billing follows the high-watermark model: it counts the maximum number of assigned users reached during the month, not the average. If you exceed the included requests, the overage is billed automatically at the end of the month, at an additional rate for every 1,000 extra requests (although, as we will see, the exact mechanism and applied rate vary from contract to contract). Joule agent actions draw from the same shared pool, but weigh disproportionately: a single agent execution consumes 5 to 10 times an interactive request; therefore, even a seemingly large pool runs out much faster when adoption shifts from a simple prompt to an agent executing actions autonomously.
There is no official price list for AI Units; everything is negotiated per contract. As a historical reference, an old SKU (prior to the new PUPM model post-Sapphire 2025) set a minimum purchase of 100 AI Units at a list price of around 7 euros per unit. Regarding overage — that is, the cost when exceeding the prepaid pool — sources do not agree. Several independent licensing advisors report punitive rates, typically 2 to 5 times the contracted rate depending on the source. SAP Learning sources cited in research instead describe, for some packages, a specific clause (Excess Use PPU): the discount the customer negotiated on the list price is reduced by 30 percentage points for excess units. Less dramatic than the 2-5x cited by advisors, but still a real surcharge, not a discount. It is an area where the specific clause in your contract matters more than any generic reference figure: it must be verified text in hand, not assumed. SAP provides two monitoring tools, each designed for a different question. The Business AI Consumption Dashboard, inside SAP for Me, answers "how many AI Units are we consuming": it shows the remaining balance, consumption for each Premium feature, and expiring units. The Joule Analytics Center instead answers "how are my employees using it": it tracks messages and conversations, but does not show AI Unit consumption. Neither covers both perspectives on its own: for a complete picture, they must be consulted together. The difficulty of having an overall view is compounded by the fact that consumption is indeed measured in real-time on the backend, but with a tenant-level balance and monthly reporting. Without internal chargeback — which can only be done by cross-referencing data — you do not know who is consuming what until the bill arrives.
We have seen how consumption works. For those defining the budget related to the use of Premium AI features, what really matters is knowing where the risks are hidden, in order to anticipate and manage them. There are two main points of attention: one contractual, one related to sizing.
1. Contractual
On the contract side, you must ensure that four conditions are explicitly fixed, starting with expiration:
Ora, con i primi anniversari contrattuali che arrivano proprio in questi mesi (i primi contratti col nuovo modello PUPM risalgono a luglio 2025), quel test arriva per davvero. È una clausola esplicitamente negoziabile: diversi advisor di licensing indipendenti segnalano che i diritti di rollover, soprattutto per impegni di volume elevato, si possono ottenere trattando col fornitore, ma solo se richiesti prima della firma, non dopo. Chi ha firmato un anno fa senza porre la questione si trova ora, al primo rinnovo, a scoprire il costo pratico di quella clausola mai discussa. È un punto da tenere in conto ad ogni rinnovo, e da inserire esplicitamente per iscritto nel contratto.
All four of these risks must be addressed during the negotiation phase, not after signing... we find them in the negotiation checklist at the end of the article.
2. Sizing
How many AI Units do you really need? Here the problem is that the estimate is not made by you — it is proposed by SAP — and the two numbers rarely match. Consumption projections provided by SAP, according to various independent consultants, tend to be abundant in the first year of adoption: if you sign based on that number, you risk buying capacity you will not use and that will expire. But the risk flips when adoption reaches a steady state. With more users, more use cases, and above all more active agents, the same budget that seemed abundant in the first year could be exhausted as early as March, well before renewal, if consumption growth is not factored in. You need to find the right balance between these two opposing risks: oversizing at the beginning wastes budget on units that expire; undersizing at steady state triggers overage, a real surcharge compared to the contracted rate, more or less heavy depending on the specific contract clause (see above). You are squeezed in the middle. To complicate everything, there are agents: a Joule agent consumes 5 to 10 times the units of an interactive prompt. The moment you transition from "AI that answers" to "AI that acts," the scale of consumption changes by an order of magnitude. And the balance is at the tenant level: without an internal chargeback and continuous monitoring — verified personally, not taken for granted (see above) — you do not know who is consuming what until the bill arrives. The most sensible move is to build your own consumption model, independent of the supplier's, and use it as a counterpart to SAP's projection during negotiation. Where real data is lacking — typically in the first approach, without historical usage — it is better to build realistic usage scenarios (by number of users, use cases, expected frequency) rather than relying solely on the supplier's projection. Sizing based on observed real data, or on the scenario built in its absence — at steady state, not just for the first year — helps strike a balance: neither wasting on an inflated estimate, nor exposing yourself to overage on an overly conservative estimate.
The negotiation checklist Before committing to a volume of AI Units, it is advisable to consider defining certain aspects during the negotiation phase, distinguishing between those to be fixed in the contract and those useful for properly sizing requirements:
On the contractual front:
On the sizing front:
Conclusion
For the first time, SAP spending becomes variable, and AI Unit overages automatically grow as AI adoption increases. It is no longer a budget fixed at the beginning of the year with the user count and contract escalator; it is consumption that must be anticipated and monitored month by month. The first renewals arriving in these weeks are the first concrete opportunity to see how this model really works: those who negotiated well a year ago are finding out now, and those who didn't, are too.
At WEGG we are expert SAP licensing consultants and, at the same time, FinOps consultants: a combination that allows us to follow clients on both fronts described in this article. We can support you in the right-sizing of your actual AI Units requirements, in contract analysis — what you actually signed, and above all what is missing — and in negotiation support, helping you define protection measures and build consumption models independent of the supplier's estimate.
Ne parleremo nel webinar “SAP nell’era dell’AI”, in programma per il 24 settembre. Here il link per iscriversi.
Insights
OUR OFFICES
OUR OFFICES
PADUA
Via Arnaldo Fusinato 42, 35137
MILAN
Viale Enrico Forlanini 23, 20134
ROME
Viale Giorgio Ribotta 11, 00144
Copyright © 2025 WEGG S.r.l. • P.I 03447430285 • C.F. 02371140233 • REA 311023
Azienda Certificata ISO 9001:2015 – ITA / ISO 9001:2015 – EN
