Negotiating SAP AI Contracts: How to Buy AI Units, Set Consumption Caps and Avoid Overruns

SAP's AI contracts are designed to maximise SAP's revenue flexibility, not your budget predictability. The shift to consumption-based AI pricing — formalised with the July 2025 commercial restructure — introduces commercial risks that have no equivalent in traditional named-user licence agreements. The fundamental difference: traditional SAP licences give you a fixed asset at a fixed price. AI Units give you a metered service where both the consumption rate and the price per unit can change during your contract term, if you allow SAP to include their standard terms without challenge.

Most enterprise procurement teams are negotiating SAP AI contracts using the same playbook they use for user licences — volume discounts, payment terms, renewal protections. That playbook misses the three commercial risks that actually matter in AI contracts: consumption rate manipulation, unilateral repricing, and overrun monetisation. This guide covers all three and provides the specific contractual language you need to protect your organisation.

Before diving into negotiation mechanics, understand the broader SAP AI Units model — the consumption currency that underpins all Joule, BTP AI, and Business AI charges.

Why SAP AI Contracts Are Different

Three structural differences make SAP AI contracts more commercially dangerous than traditional licence agreements.

1. The Metric Is Opaque by Design

A Professional Named User licence is a discrete, auditable asset. You either have a user assigned or you don't. An AI Unit is a computed quantity — the result of SAP's platform measuring the complexity, token count, and service calls associated with each AI interaction. You cannot independently verify AI Unit consumption using your own tooling. You are entirely dependent on SAP's measurement infrastructure, which SAP controls, can revise, and is not subject to third-party audit in the same way that USMM-based licence measurements are.

2. Consumption Is Not Predictable at Contract Signing

When you buy 1,000 Professional Named User licences, you know exactly what you've bought. When you buy 10 million AI Units, you have no reliable way to predict what business outcomes those units will deliver — because the consumption rate per outcome depends on which features you activate, which agents you deploy, and how SAP adjusts its consumption model over the contract term. SAP's own consumption modelling tools are optimistic by design: they model copilot usage rates, not agentic rates.

3. SAP Has Multiple Revenue Recovery Mechanisms

If you under-purchase traditional licences, SAP recovers revenue through the audit process — a process that is well-documented, extensively litigated, and has established norms. If you under-purchase AI Units, SAP recovers revenue through three different channels: automatic overage billing, renewal uplift based on consumption data, and competitive pressure from deployment dependency. The audit for AI overuse is SAP's usage data from your own BTP account — and they have it in real time.

Understanding AI Unit Pricing

SAP AI Units are priced through three commercial vehicles: standalone purchase through BTP Global Account, Business AI subscription bundles, and RISE/GROW entitlements. Each carries different unit economics and different commercial risks.

Standalone BTP purchase carries the highest per-unit cost and the least favourable terms. SAP's standard BTP service descriptions apply, including the unilateral repricing clause. Avoid purchasing AI Units as standalone BTP line items unless you have no other vehicle available.

Business AI subscriptions (Base and Premium) bundle AI Units with access to the full Joule and embedded AI feature set. The effective per-unit cost is lower than standalone, but the total commitment is higher and the specific unit allocation per use case is less transparent. SAP's commercial team will typically not provide a unit-by-unit breakdown of what the subscription includes — they prefer the "unlimited within fair use" framing, which is not a contractual commitment and can be used against you at renewal.

RISE and GROW entitlements typically include AI Units as a BTP credit allocation within the overall subscription. This is the most commercially efficient vehicle for most enterprises — the AI Units are effectively subsidised by the overall RISE contract value. Our RISE with SAP advisory team negotiates AI Unit inclusions as a core component of every RISE deal review.

Sizing Your AI Unit Purchase

The single most common commercial mistake in SAP AI contracts is purchasing AI Units based on SAP's consumption model without building in an independent consumption analysis. SAP's model will project copilot usage rates. Your actual deployment — especially if it includes Joule Agents — will consume at significantly higher rates.

A responsible consumption model requires three inputs:

1. Use case inventory — a complete list of every AI feature you intend to activate, categorised by copilot interaction vs. agentic workflow
2. Volume projection — estimated monthly frequency of each use case, with high/medium/low scenarios
3. Rate schedule confirmation — written confirmation from SAP of the specific AI Unit consumption rate for each use case category in your deployment plan

Most enterprises complete steps 1 and 2 and skip step 3 — because SAP resists providing written rate confirmations. Do not accept a rate schedule embedded in a URL reference to SAP's documentation. Documentation can be updated. You need the specific rates fixed in your Order Form as an exhibit or schedule.

Negotiating Consumption Caps

Consumption caps are the most important structural protection in any SAP AI contract. Without a hard cap, SAP's platform will continue accepting AI requests after your purchased allocation is exhausted — generating overage charges that SAP then uses to anchor the renewal conversation.

SAP resists consumption caps for a simple commercial reason: caps reduce their overage revenue and limit the consumption data they can use as renewal leverage. You should expect pushback and you should not accept it. A hard cap is a contractual right, not a technical capability limitation — SAP's BTP platform supports consumption limits at the Global Account level.

Types of Consumption Controls to Negotiate

Hard stop caps — the platform refuses additional AI requests once the purchased allocation is exhausted. No automatic top-up, no silent overage. This is the strongest protection and SAP's most resisted position.

Alert-and-approval caps — the platform alerts your designated administrator at defined consumption thresholds (typically 75%, 90%, 95% of allocation) and requires explicit approval before any additional consumption is permitted. This is a reasonable middle position that SAP is more likely to accept.

Sub-account caps — consumption limits applied at the BTP sub-account level, allowing you to allocate AI Units by business unit or use case and prevent any single deployment from consuming disproportionate allocation. This controls internal consumption patterns without requiring SAP to enforce an overall hard cap at the account level.

Rate Lock and Price Protection

SAP's standard Business AI and BTP service terms contain a clause — typically buried in a supplemental service description — that permits SAP to revise AI Unit consumption rates with 90 days' notice. This clause is commercially extraordinary: it means SAP can increase the effective cost of your existing deployment mid-contract without requiring your consent to the price change.

The commercial rationale SAP uses to justify this clause is that AI infrastructure costs are unpredictable, requiring commercial flexibility. The reality is that SAP's major cloud infrastructure is contracted with hyperscalers at stable rates. The clause exists to give SAP commercial flexibility as AI pricing evolves — specifically, to benefit from the AI premium pricing that enterprise software vendors are currently extracting from buyers who lack commercial benchmarks.

The specific language you need to replace the standard rate revision clause:

Most SAP account teams will claim this language requires legal escalation. That is true — accept it. The escalation delay is commercially acceptable because the protection it provides is material. A mid-contract rate revision of 15–20% on a large AI Unit purchase is a seven-figure exposure for a major enterprise deployment.

Dangerous Clauses to Remove

Beyond the rate revision clause, SAP's standard AI terms contain several additional provisions that enterprise legal teams must review before signature.

The "Fair Use" Limitation

Some SAP Business AI subscription descriptions include a "fair use" qualifier on AI Unit access — language suggesting that extremely high consumption may trigger commercial review. "Fair use" is not a contractual limit — it is a commercial threat mechanism. SAP can invoke it to force a commercial conversation when your consumption is high, even if you are within the technical parameters of your subscription. Remove "fair use" language from all AI contracts and replace it with explicit, quantified consumption allocations.

The AI Model Substitution Clause

SAP's standard terms permit SAP to substitute the underlying AI model powering Joule with a different model at SAP's discretion. Model substitution can materially affect performance, consumption rates, and output quality — all of which can affect the business value of your AI deployment. Require SAP to commit to maintaining materially equivalent AI model performance and to provide 180 days' notice before any model substitution that materially changes consumption rates or output characteristics.

The Usage Data Provision

SAP's AI services collect detailed telemetry on how AI features are used in your environment. SAP's standard terms permit SAP to use aggregated usage data for "product improvement and benchmarking." This provision allows SAP's commercial team to access consumption patterns across the SAP customer base — and to use that data to benchmark what enterprise customers are paying and consuming. Your negotiating position is strengthened if you limit SAP's ability to use your organisation's consumption data for SAP's own commercial intelligence. Require that your data is excluded from benchmarking analytics used for SAP's pricing decisions.

The Automatic Renewal with Uplift Clause

AI Unit purchases that auto-renew with standard uplift (typically 3–5% annual) compound over time in a way that named user agreements do not — because consumption typically grows as adoption expands. An AI Unit contract that auto-renews with 5% annual uplift, while your consumption grows at 30% per year, creates a growing gap that SAP exploits at the next formal renewal. Require explicit opt-in renewal for all AI Unit commitments, with no automatic uplift and full right to right-size the allocation based on actual usage data.

Bundling Strategy

The most effective commercial strategy for AI Unit pricing is to negotiate AI as a component of a broader SAP deal rather than as a standalone purchase. SAP's AI Unit pricing responds to overall deal economics — a £5M RISE renewal with AI Units bundled will produce a materially better AI Unit effective rate than a standalone AI Unit purchase of equivalent size.

Bundling opportunities to explore in your next SAP contract negotiation:

• RISE with SAP AI inclusion — negotiate AI Unit allocations as a non-negotiable inclusion in your RISE subscription, not as an add-on. SAP's standard RISE packages include limited AI entitlements; pushing for 2–3x the standard inclusion at no additional cost is achievable in competitive situations
• S/4HANA Cloud PE AI bundling — similar approach for private cloud deployments where AI is increasingly positioned as a differentiator against competitors
• Enterprise License Agreement (ELA) AI components — enterprises with SAP Enterprise License Agreements can negotiate AI Unit pools as part of the ELA structure, providing the most flexibility in deployment without triggering per-use-case commercial conversations
• Hyperscaler credit funding — if your organisation has committed spend with AWS, Azure, or Google Cloud, investigate whether SAP AI Units can be purchased through the hyperscaler marketplace using existing credits. Rates available through marketplace channels can be 20–35% lower than direct SAP purchase in specific deal configurations

Complete Negotiation Checklist