Business AI: Use Cases and Decision Patterns

JinJin48

SAP Consultant | BTP · S/4HANA · US Global Rollout

May 9, 2026 · 15 min read

Where This Article Sits#

References#

1) The Use-Case Landscape#

Business AI on BTP is not a single feature per LoB. Each LoB has its own mix of (a) Joule conversational/agentic capabilities, (b) embedded ML inside the application, and (c) custom extensions on AI Core / GenAI Hub. Knowing which mix a use case sits in determines licensing, prerequisites, and the realistic GA boundary.

They are not substitutes.

2) How It Connects to S/4HANA#

Every Business AI use case ultimately reads from or writes to a system of record — S/4HANA for Finance, Ariba and S/4HANA for Procurement, SuccessFactors for HR. The integration shape is the same: the LoB UI surfaces Joule, Joule calls AI Foundation through the BTP control plane, and the transactional posting happens in the application of record.

                     ┌──────────────────────────────┐
   User in           │ LoB UI                       │
   Fiori / Ariba ───►│  Joule chat / agent surface  │
   / SuccessFactors  └──────┬───────────────────────┘
                            ▼
                     ┌──────────────────────────────┐
                     │ BTP — AI Foundation          │
                     │   AI Core + GenAI Hub        │
                     │   Joule orchestration        │
                     └──────┬───────────────────────┘
                            ▼
   ┌───────────────────────────────────────────────────────┐
   │ System of record (one of)                             │
   │   • S/4HANA Cloud (Finance postings, MM, …)           │
   │   • SAP Ariba (sourcing, contracts, SLP, Buying)      │
   │   • SAP SuccessFactors (EC, RCM, P&G, LMS, TIH)       │
   │ Audit log, role-based access, master data live here.  │
   └───────────────────────────────────────────────────────┘

For RISE with SAP and GROW with SAP customers, the Joule assistant tier is included; agents and custom skills meter separately through AI Units. For S/4HANA private edition or non-SAP-cloud landscapes, BTP entitlement, IAS/IPS, and identity propagation must be set up before any LoB feature can call Joule.

Key point: Business AI does not replace the LoB transaction; it sits in front of it. The audit trail, role-based access check, and master data come from the application of record. Treating Joule as the system of record is the most common architectural mistake in 2026 AI scoping conversations.

3) Maturity Map#

Maturity in 2026 is uneven across LoBs and within each LoB. Judge each agent on its own — the same project can ship one GA and one limited-release pilot side by side.

4) Requirements Checklist#

The §3 map answers “is it ready?” This section answers “are we ready?” Build one per-domain checklist into every scoping document, plus the cross-cutting BTP one.

Finance use cases (Joule + Cash Management Agent + embedded ML)#

Procurement use cases (Joule in Ariba + Bid Analysis + supplier 360)#

HR use cases (Joule HR + Talent Intelligence Hub + recruiting)#

Cross-cutting (any Business AI use case on BTP)#

5) Decision Criteria#

The three questions to ask first#

Q1. What measurable inefficiency in this LoB are we removing, and what is its current cost?

If the customer cannot describe the pain in numbers — DSO days, sourcing cycle time, time-to-fill, % of cash auto-cleared — Business AI is not the right project to start. A vague “we want AI in finance” stalls in PoC.

Q2. Where does the data sit, in what regions, and is the master data clean enough for grounded answers?

Answer this before talking about agents. Dirty supplier master and ungoverned skills taxonomies are not rescued by a model. If the data must remain in a region where the agent is not provisioned, the conversation is about regional rollout, not AI.

Q3. Who owns the use case after go-live — suggestions, review queue, fairness checks, AI Unit budget — and do they have year-two budget?

Business AI is not a one-time install. AI Units consume monthly, agents update each release, fairness expectations tighten with regulation. Without a named LoB owner with year-two budget, the use case decays silently.

Domain selection by project context#

Licensing and cost#

6) Common Failures#

These are requirements-stage failures, not implementation bugs — every one is preventable in scoping.

Failure 1: Use case promised on a limited-release agent as if it were GA

A customer hears “Joule does X” from a roadmap deck; the consultant commits; X turns out to be limited release for another cycle. Always check the current Joule Capabilities Guide PDF version and flag NDA pilots separately.

Failure 2: Master data quality assumed at the level required for grounded answers

Ungoverned skills taxonomies yield inconsistent Joule HR suggestions; duplicates in SLP produce noisy supplier 360 summaries; a fragmented chart of accounts makes Cash Management Agent contradict Fiori reports. Master data quality is part of the AI requirement.

Failure 3: Auto-execution turned on without a fairness or HITL plan

Joule HR’s shift from advisor to executor in 1H 2026 invites this failure. Auto-execution on hiring shortlists, calibration, or contract changes without fairness validation and a human-review fallback is not defensible in 2026.

Failure 4: Region constraint discovered after entitlements are raised

The project assumes a Joule agent is GA in the customer’s region; the integration guide is read in week six and reveals EU-only or phased rollout. Read the region matrix in week one.

Failure 5: AI Unit budget modeled on go-live volume only

AI Units do not roll over; query volume scales with adoption. A go-live forecast under-allocates by year 2 and forces emergency procurement. Model year-2 volume per use case with at least 30% headroom.

Failure 6: No LoB owner for the AI suggestions after go-live

IT owns the platform; nobody owns the suggestions. The category manager, controller, or HRBP must own bad-suggestion suppression, SAP feedback, and year-two budget. Without a LoB owner, the suggestion stream becomes noise within two quarters.

Failure 7: Customizing Joule with Joule Studio before validating standard agents

Joule Studio is GA, so the temptation is to jump to custom agents. Standard agents (Cash Management, Bid Analysis, Joule HR) cover most pain points with SAP-owned lifecycles. Start standard; reach for Joule Studio only when one is genuinely missing.

FAQ#

Q: Do these Business AI use cases work on SAP S/4HANA on-premise, or only S/4HANA Cloud?

The cleanest fit is S/4HANA Cloud (public and, for many capabilities, private edition) plus cloud SuccessFactors and Ariba. On-prem S/4HANA can still use embedded Finance ML, but Joule and most LoB agents target cloud editions; hybrid landscapes need explicit BTP, IAS, and connector setup, and several agents will not light up without a cloud system of record. Read the integration guide per use case before promising parity.

Q: How is Business AI on BTP licensed across LoBs, and where is the surprise cost?

Joule’s assistant tier is included with RISE and GROW with SAP. Beyond that, agents, custom skills, GenAI Hub invocations, and embedded Document AI meter through AI Units (0.1 AI Units per Document AI request as a public reference point). AI Units do not roll over. The two surprises: year-two volume scales faster than year-one with adoption, and bursty Procurement events / HR cycles spike consumption that flat-rate budgets miss. Hard pricing is account-team-negotiated; build year-1 and year-2 forecasts per use case.

Q: When should we descope a Business AI use case from a project?

When any of the following is true: the customer cannot quantify the LoB pain; master data is not at the quality required for grounded answers; the use case sits on a limited-release / NDA agent the customer has not enrolled in; the region does not yet support the agent; auto-execution is required but no fairness or HITL plan exists; or no LoB owner has year-two budget. Promising free-form contract auto-execution or fully automated hiring is malpractice in 2026.

Q: Can Joule replace embedded ML like Cash Application or Predictive Accounting?

Not as of 2026, and SAP does not market it as a replacement. Embedded ML predates Joule, has its own training and feedback loops, and remains the production path for high-volume AR matching and predictive postings. Joule sits on top — answering queries, summarizing dispute root causes, translating e-invoicing errors — without retraining the embedded models. Treat them as a pair, not a migration.

Q: How do we document a Business AI use-case requirement so it survives handover to delivery?

Write it as a quintet: (1) the named capability (Cash Management Agent / Bid Analysis Agent / Joule HR career plan / etc.), (2) the LoB pain metric and target, (3) master-data and identity prerequisites with named owners, (4) the HITL or fairness pattern with a written threshold, (5) AI Unit budget with year-1 and year-2 forecasts. Without all five, the requirement gets reshaped during delivery — usually by removing scope.

Q: Should we wait for Joule Studio custom agents before starting?

No. Joule Studio is GA in Q1 2026, but standard agents cover most Finance, Procurement, and HR pain points with SAP-owned roadmaps. Start with standard agents, capture remaining gaps after a quarter of real usage, then design Joule Studio agents for the residual cases — owned by the LoB.

Q: How do regulatory considerations (EU AI Act, pay transparency, audit) shape use-case selection?

They tighten the boundary on auto-execution and on HR. Hiring, calibration, and pay-transparency use cases must keep an explicit human-accountability layer; SAP commits to audit trails and explanation surfaces, and the customer’s governance review (often ISO 42001-aligned) closes the loop. In Finance and Procurement the constraint is lighter but still real: AI suggestions are advisory until a named role accepts them.

Summary#

Business AI on BTP in 2026 is best read as three LoB-shaped portfolios on the same AI Foundation runtime — Finance (Cash Management Agent, Dispute Resolution Agent, Joule in S/4HANA, embedded ML), Procurement (Bid Analysis Agent, supplier 360, sourcing copilot in Ariba), and HR (Joule HR, Talent Intelligence Hub, recruiting copilot, Learning AI). The consultant value is reading the customer’s master data, region, role-based access, and human-review pattern and saying — accurately — which agents are GA today, which are limited release, which need prerequisites, and where the answer is “not in this phase.” Promise inside the GA envelope, refuse outside it, and ship the LoB owner and AI Unit year-two budget every time.

What to Read Next#