SAP AI Needs a Status Ledger, Not a Capability Count

Fund generally available value, learn through pilots, and keep roadmap promises out of the business case

Altivate | Published 26 April 2026 | Updated 26 July 2026

1. One portfolio count hides three investment decisions

SAP’s headline figure is 224 agents and 51 assistants across five domains, as framed since Sapphire in May 2026. It is a real number. It is also doing a lot of work, because it counts two categories of thing that a buyer should treat completely differently.

Speed one: narrow, embedded “AI-assisted” features. These sit inside a line-of-business application, do one bounded job and ship every quarter. Some are Base AI included with the cloud subscription; others are Premium AI and consume AI Units. Examples include creating a sales order from a PDF purchase order, explaining an error in plain language, and recommending a worker role on a statement of work.

Speed two: autonomous, multi-step “Joule Agents.” These carry the agentic narrative. In Forrester’s May 2026 assessment they are “mostly mixed GA, early-adopter, and preview status.” They are the thing being demonstrated on stage, and they are substantially not the thing you can put in a signed 2027 business case yet.

Two consequences follow, and most buyers get both wrong at once:

1. You may be under-using speed one. Inventory the GA features in your current entitlements before buying new scope, then verify whether each feature is Base or Premium AI.
2. You may be over-committing to speed two before the consumption model is mapped to your specific users and actions. See §5.

2. Untangling SAP’s own vocabulary

Three layers, used near-interchangeably in marketing, which is most of why buyers are confused:

The distinction that actually matters commercially is “AI-assisted” versus “Agent.” An AI-assisted feature does one step and hands back to a human. An Agent takes multiple steps toward a goal. SAP labels them differently in its release notes. When a vendor deck says “agentic,” check which word SAP used.

Overview: SAP Business AI and SAP Business AI on BTP.

3. The table

Status as published by SAP, with the source type for every row, checked 26 July 2026.

SAP primary means SAP itself applied that status label in its own release notes. Press only means the status comes from industry reporting and SAP has not labelled it. That distinction is the most useful column here.

Generally available – usable in a 2026 plan

Beta or Early Adopter Care – pilot, do not budget-commit

Roadmap

⚠️ That row needs a qualification, and it is the most easily misread line in this document. “Roadmap, Q4 2026” is the right label for A2A’s general availability, but it would be wrong to read it as “nothing works today.” A2A interoperability is directional right now, and only one direction is built:

• Outbound – Joule calling an external agent – exists and is documented today. SAP’s own reference architecture states that the current architecture “supports unidirectional (outbound) communication only,” and SAP community documentation walks through Joule invoking an externally hosted agent at runtime via an A2A Agent Card.
• Inbound – an external orchestrator calling Joule agents as peers – is not generally available. The Agent Gateway that would enable it is explicitly pre-GA, and SAP’s Sapphire 2026 guide states that “bi-directional A2A capabilities in Joule will be enhanced to enable third-party agents to securely call on Joule Agents and act within enterprise processes” – future tense, targeted at the same Q4 2026 date.

So the accurate one-line status is: Joule can call out; it cannot yet be called into. If you are making an orchestration decision now, that direction is the fact that matters.

⚠️ Claimed generally available, but not labelled by SAP – treat as unconfirmed

A separate and lesser case: Concur’s Policy Navigator appears in earlier coverage as an Early-Adopter capability, but we could not re-confirm it under that name in either quarterly digest. That is a naming or existence question rather than a labelling one, so we have not grouped it with the two rows above.

One row that deserves its own note

The Performance Preparation Agent in SuccessFactors – widely reported as SuccessFactors’ first Joule Agent, generally available since late 2025 – is confirmed GA in SAP’s own Q4 2025 digest under that name. Note that SAP’s post does not use the name “Performance and Goals Agent” that circulates in third-party coverage, and does not state a November 2025 date. If you are matching a vendor’s claim to SAP’s own record, match on SAP’s name.

Product detail: SAP Joule agents.

4. Where a confirmed status actually lives

SAP’s SuccessFactors 1H 2026 Release announcement – the document an HR buyer would naturally read for these four agent areas – names all four but attaches no GA, Beta or Early-Adopter label to any of them. The quarterly Business AI digest, by contrast, labels every capability it covers.

We are deliberately not stretching that into a claim about SAP’s house style. We did not find a second SAP product-release write-up to compare it against, so treat it as a demonstrated single instance rather than a proven policy – and note the counter-example sitting in our own source list: SAP’s Sapphire 2026 coverage is a comparable non-digest announcement, and it does label its headline items.

The practical rule holds regardless of which explanation is right. For a confirmed status, go to the quarterly Business AI release highlights – not the product release announcement, and not the press coverage. The digest is the one place SAP has consistently attached a status word to a capability name. If the capability isn’t there with a status next to it, you do not have a confirmed status; you have a plan. That is also how press coverage goes wrong: when a headline reports “SAP ships agents across X,” it is frequently filling a labelling gap with an inference.

5. The commercial model is public; your monetary case is not

SAP’s current pricing page separates Base AI, included in SAP cloud subscriptions, from Premium AI, activated through purchased AI Units. AI Units are bought annually, expire after 12 months if unused and draw from a central pool.

SAP also publishes example consumption mechanics. Joule Premium packages use volume tiers ranging from 8 to 1 AI Units per user per month, while Document Grounding consumes 0.005 AI Units per record. Customers can track use in SAP for Me and use SAP’s estimator before requesting a quote. Premium activation uses SKU 8019164.

Those mechanics are not a monetary business case. The public page does not provide the customer-specific quote or map every agent action and promotion to a stable consumption rate. Promotional end dates may also differ by capability. Model each proposed capability separately:

1. Is it Base or Premium AI?
2. What unit measure applies: named user, record, action or another basis?
3. What promotion or included quantity applies, and when does it end?
4. What annual quantity, expiry exposure and monetary quote apply to the proposed scope?

6. The prerequisite that is actually a project

Every SAP AI capability in §3 has an entitlement checklist behind it, and one item on that checklist is not like the others.

Joule requires a single, shared SAP Cloud Identity Services (IAS) tenant across Joule and every integrated SAP product, with a consistent Global User ID. In a landscape assembled over a decade, with an ECC estate, a separately bought SuccessFactors and an Ariba that arrived with an acquisition, that is not a switch. It is an identity consolidation programme and a common dependency that can stall implementation. Establish where you stand on it before you size anything else.

The rest is procurement, and it is worth checking in one sitting because each item is a hard gate:

• An SAP BTP enterprise Global Account with Joule foundation-plan entitlements, plus SAP Build Work Zone in that same Global Account.
• An underlying application licence for each LOB app you want to enable – Joule and Joule Agents are add-ons, and there is no standalone Joule purchase.
• SAP AI Core on the Extended service plan, not Standard, for anything depending on Generative AI Hub. This one is routinely missed because Standard sounds sufficient.
• Some capabilities are gated to S/4HANA Cloud Public Edition and are simply not available on Private Edition. Check per capability – the split is not intuitive, and it cuts across the GA table in §3.

Treat this as a procurement checklist, not an entitlement statement. Confirm each item in the applicable order form, service description and current product documentation before it becomes a project dependency.

Background: what is the SAP Generative AI Hub on BTP.

7. Gulf and India: the question SAP hasn’t documented

For a buyer in Saudi Arabia, the UAE or India with a data-residency constraint, the first question about any AI capability is not what it does. It is where the inference runs.

What is documented, read directly from SAP’s own platform documentation:

• Saudi Arabia – two BTP regions, sa30 and sa31, both in Dammam, on Google Cloud, split between regulated and non-regulated workloads. Note this specifically: not Riyadh. A Riyadh presence is claimed by secondary sources and appears to relate to SAP Business Network rather than BTP.
• UAE – ae01, Dubai.
• India – in30, Mumbai.

What is not documented, anywhere we could reach:

• Whether Generative AI Hub / AI Core on the Extended plan can be provisioned in those regions at all. We read SAP’s own Generative AI Hub overview and its BTP regions documentation. Neither connects the two. SAP’s Discovery Center – the authoritative per-region service catalogue – is not machine-readable.
• Where inference actually executes for a customer in those regions. The only related published fact is that not all models are available in all regions.

We are stating this gap rather than filling it, because for a regulated Gulf buyer this is the whole question. A data-centre in Dammam does not tell you your AI inference runs in Dammam. If your data classification requires in-Kingdom processing, get the answer in writing, per capability, before you enable anything – and treat any consultant who answers it confidently from public documentation with suspicion, because that documentation does not exist.

8. Run SAP AI as a three-book portfolio

Maintain three separate books:

1. Production value: generally available capabilities, with entitlement, adoption owner, baseline, target and realised outcome.
2. Option value: beta and Early Adopter Care capabilities, with a bounded pilot, exit criteria and no recurring value booked.
3. Dependency watch: roadmap and unlabelled capabilities, with the decision they may affect and a dated recheck owner.

Do not combine the books in one “AI value” number. A roadmap capability can justify preserving an option; it cannot justify signing a production benefit into the business case.

9. The claims that stay outside the ledger

Status decays on SAP’s quarterly cadence. Five capabilities in §3 are described as generally available by industry press but do not carry an SAP status label in the material reviewed; they remain unlabelled, not GA and not dismissed. Customer-specific monetary quotes, per-agent consumption mapping and the Gulf/India inference-location question remain outside this ledger.

Vendor performance claims and sponsored ROI studies are excluded from the investment case until the enterprise can reproduce the baseline, intervention and result in its own environment.

10. The executive agenda

• Inventory generally available capabilities and verify whether each one is Base or Premium AI.
• Assign an adoption owner and measurable process outcome before enabling each one.
• Put beta and Early Adopter Care capabilities behind pilot budgets and explicit exit gates.
• Obtain the customer quote, capability-level consumption mapping, promotional terms and inference location in writing.
• Refresh the ledger after every quarterly digest and before every board or investment decision.

Altivate provides SAP advisory and implementation services and therefore has a commercial interest in this market. The first useful conversation is still not what to buy. It is which current, contracted capability can produce measurable value now.

Start with SAP Business AI, or return to White Papers.

Sources and verification status

Every status in §3 was checked on 26 July 2026 against SAP publications.

SAP primary

• SAP News Center, SAP Business AI: Release Highlights Q2 2026 (20 July 2026), the most recent quarterly digest and source for every Q2 status above. No Q3 2026 digest existed as of 26 July 2026; SAP’s cadence puts the next one in October.
• SAP News Center, SAP Business AI: Release Highlights Q1 2026 (14 April 2026).
• SAP News Center, SAP Business AI: Release Highlights Q4 2025 (14 January 2026).
• SAP, Business AI packages and pricing, documents Base and Premium AI, annual purchase, 12-month expiry, example consumption measures, SAP for Me tracking, the estimator and SKU 8019164.
• SAP reference architecture, Build and deliver AI agents (last updated 6 May 2026), is the source for “the current architecture supports unidirectional (outbound) communication only” and for the Agent Gateway being pre-GA. This is what establishes that A2A is directional today.
• SAP, Sapphire 2026 Innovation News Guide (12 May 2026), covers the bi-directional A2A enhancement and the Q4 2026 GA target for Joule Work and Joule A2A.
• SAP community documentation, “Joule A2A: Connect Code-Based Agents into Joule” (16 February 2026) – the runtime walkthrough of Joule invoking an externally hosted agent via an A2A Agent Card. Community-authored, not SAP product documentation – cited for the existence of a documented outbound path, which SAP’s own reference architecture independently confirms.
• SAP News Center, Sapphire 2026 coverage (13 May 2026) – AI Agent Hub general availability, the Joule Work glidepath, and A2A targeted for Q4 2026.
• SAP News Center, SAP SuccessFactors 1H 2026 Release (April 2026) – names the four new agent areas without status labels. This absence is itself a finding (§4).
• SAP Help Portal, the current BTP region list and Generative AI Hub overview, supports §7’s region correction and documented-gap finding.

Analyst

• Forrester, SAP Sapphire 2026 analyst blog (12 May 2026) – [read] the 224-agents/51-assistants mixed-status framing, the 31 December 2026 free-runtime expiry, and the unmodelled-2027-pricing assessment. Quotes verified word for word.

Industry press used only where explicitly labelled as unconfirmed in §3

• SAPinsider (June 2026) on Ariba and Fieldglass Joule agents.

Not confirmed for the investment case: customer-specific monetary prices, per-agent post-promotion consumption mapping and Generative AI Hub regional availability.