Artificial Intelligence in SAP environments has shifted from being an “external accelerator” to becoming an embedded capability across the SAP portfolio : especially in S/4HANA-centric landscapes and SAP Business Technology Platform (SAP BTP).  In parallel, SAP’s AI operating model has matured through services and patterns such as SAP AI Core (runtime/orchestration), SAP AI Launchpad (central lifecycle management), the unified AI API, and Joule as the conversational copilot experience across SAP applications.  Yet, many SAP AI initiatives still stall after pilots because teams start by debating “how to implement” rather than validating “what is worth implementing” and “where AI is actually valuable and feasible” in the enterprise process fabric.

A disciplined way to avoid this trap is a discovery‑first AI assessment – a structured approach that stays implementation‑agnostic while still being technology‑aware, so outcomes remain valid regardless of later platform or deployment decisions.  This essay reframes a six‑step approach for SAP AI discovery into a technical, architect-friendly method. It also overlays SAP guidance and best-practice assets (Discovery Center, AI BTP Best Practices, AI Core/Launchpad, Joule, and RAG patterns with SAP HANA Cloud Vector Engine) to ensure the assessment aligns with SAP’s recommended adoption pathways.

This essay is based on the framework introduced on the original SAP Community article: 6 Steps to an AI Assessment in SAP: Discovery‑First Approach 

Fig:  6 Steps to an AI Assessment in SAP: Discovery-First Approach

Step 1: Frame the Assessment Scope

Define the Assessment Boundary in an S/4HANA‑Centric System Context

Every AI assessment must begin with scope governance: identify the SAP domains (Finance, Supply Chain, Manufacturing, Sales, Asset Management, HCM, etc.), the adjacent applications, and the integration boundaries that matter – without drifting into build-versus-buy or tool comparisons.  In SAP programs, a reliable scoping abstraction is the business capability map or value-stream decomposition because it remains stable even when the underlying SAP configuration varies.

This discovery discipline mirrors the intent of SAP’s Discover phase practices – where structured discovery is separated from delivery execution – similar in spirit to the Digital Discovery Assessment (DDA) concept that captures scope and integration needs before implementation handover.  The key is to frame the AI assessment as a portfolio discovery exercise, not an architecture design workshop.

Typical Discovery outputs :

• Capability/value-stream boundary document (what’s in/out)
• Stakeholder map and process ownership inventory (who “owns” outcomes)

Step 2: Map Capabilities to Processes

Map Business Capabilities to S/4HANA Process Realization

AI opportunities in SAP do not originate from modules or transactions; they emerge where a business capability is operationalized through an end‑to‑end process and its associated decision points, exceptions, and data flows.  The technical architect’s job in discovery is to translate capability language (eg., “Forecast‑to‑Stock”, “Hire‑to‑Retire”) into S/4HANA process realizations (document flows, validations, approvals, master/transactional data dependencies).

This mapping also surfaces the “shadow steps” around SAP – manual spreadsheets, offline reconciliations, or email-based approvals – that frequently indicate automation or augmentation potential.  Importantly, the mapping step sets up a repeatable method: once you anchor AI discovery in processes, you can evaluate opportunities consistently across multiple LoBs.

Typical Discovery outputs :

• Capability process decomposition tree
• Process pain-point register (manual effort, exception hotspots, latency)

Fig: Business Capability Map Model from LeanIX

Fig: Representation of End-to-End Planning Process Decomposition 

Step 3: Identify AI Opportunity Areas

Apply SAP‑Relevant Heuristics to Identify AI Use‑Case Candidates

With process decomposition in place, candidate use cases can be detected using heuristics that align with SAP’s common AI scenario types:

1. High manual effort / repetitive validations : intelligent automation opportunities (e.g., checks, approvals, triage).
2. High exception rates : anomaly detection, recommendation, or guided resolution.
3. Decision-intensive planning : predictive/optimization scenarios (forecasting, risk scoring, demand signals).
4. Classification-heavy work (documents, tickets, master data enrichment) : ML classification/extraction candidates.
5. Conversational access needs : natural language interfaces over SAP process data and knowledge.

A key technical nuance: discovery should consider evidence sources – historical transactional data, exception logs, process mining signals, and operational telemetry-to quantify where work concentrates and where variance is high. This prevents “idea-based” AI and anchors the pipeline in measurable process reality.

Typical Discovery outputs :

1. Use‑case hypothesis cards (problem, trigger, decision, data anchors, KPI)
2. Candidate taxonomy (predictive vs classification vs GenAI assistant vs automation)

Step 4: Score Opportunities (Qualitative)

Qualitatively Score Value, Feasibility, and Adoption Readiness (Lightweight, Not a Full Business Case)

Discovery must prioritize, but it should do so with qualitative scoring rather than prematurely engineering a full ROI model.  The practical approach is to score each candidate along three vectors:

1. Business value (cycle time reduction, compliance uplift, cost, working capital impact).
2. Feasibility (data availability in S/4HANA, integration complexity, process standardization, automation baseline).
3. Adoption readiness (change impact, user trust, ownership clarity, operating model maturity).

In SAP programs, the feasibility discussion is still “discovery-grade,” but it can be informed by SAP adoption assets : such as SAP Discovery Center’s Discover >> Evaluate >> Adopt structure and its estimators and missions that help teams understand effort patterns without locking design choices.

Typical Discovery outputs :

• Scoring matrix (relative ranking, not absolute valuation)
• Constraints register (data gaps, master data quality, authorization concerns)

 Fig: Discovery First Approach: Qualitative Assessment Template

Step 5: Build AI Impact Heatmap

Consolidate Findings into an “AI Impact Heatmap” Across SAP Processes

Once candidates are scored, the assessment should converge into a single portfolio artifact: an AI impact heatmap – a quadrant-style visualization that positions use cases by value vs feasibility (and optionally overlays adoption readiness).  In SAP engagements, this heatmap becomes a common language between business stakeholders, functional consultants, and enterprise architects because it translates process discussions into prioritization outcomes.

The architectural strength of the heatmap is that it naturally forms a two-speed backlog:

1. High value / high feasibility : prime candidates for early proof-of-value.
2. High value / lower feasibility : strategic backlog tied to data/process remediation.

Typical Discovery outputs :

1. Heatmap with quadrant logic and narrative rationale
2. “Dependency notes” for backlog items (data quality, process harmonization)

Fig: Discovery First: Impact Heat Map template 

Step 6: Select Top 3–5 Discovery Candidates

Once Winners Use Cases are identified and backlog created, Explicitly End Discovery

The final discovery step is to select a small set of 3–5 prioritized candidates that have clear business relevance, process anchoring, and feasible starting points.  The discovery assessment should then stop; solution architecture, tool enablement, and build plans belong to subsequent phases.

SAP Discovery Center can then serve as the “bridge” into execution by providing guided missions and reference assets for moving from decision to adoption in a controlled way.  This ensures the discovery team does not overstep into engineering prematurely while still giving the delivery team a credible launchpad.

Fig: Worked Example: “Facilitate Invoice Validation” in S/4HANA Finance

What do we do next? 

SAP AI Best‑Practice Overlay (What to Reference Once Discovery Hands Off)

While discovery remains implementation-agnostic, SAP best practices can (and should) shape the guardrails for what comes next:

1. Use SAP Discovery Center as the adoption backbone : it is explicitly structured to “Discover, evaluate, and adopt” SAP BTP and SAP Business AI, and it provides missions, reference architectures, and estimators that help teams progress methodically.
2. Operationalize AI via SAP AI Core + SAP AI Launchpad + AI API : SAP positions AI Core as the execution engine for AI workflows and model serving, AI Launchpad as the central lifecycle management experience, and AI API as a standardized lifecycle interface across runtimes.
3. Apply SAP BTP AI Best Practices for GenAI patterns : SAP provides repeatable patterns such as secure model access, prompt templating, data masking, content filtering, and vector-based RAG pipelines—explicitly designed to reduce risk and standardize enterprise implementation.
4. Adopt Joule where conversational augmentation fits : Joule is positioned as SAP’s copilot for conversational interaction and process assistance across SAP applications, with enterprise readiness and compliance considerations (including GDPR and AI ethics) called out in SAP documentation.
5. If grounded GenAI is required, use RAG with SAP HANA Cloud Vector Engine : SAP documents vector engine capabilities and the role of vector databases in supporting RAG to improve response quality and mitigate hallucinations, and SAP Architecture Center provides a RAG reference architecture.

Why Discovery‑First Works in SAP Programs?

A discovery-first assessment is effective because it treats SAP AI as a process portfolio problem, not a tool problem.  By scoping carefully, mapping capabilities to S/4HANA processes, applying consistent heuristics, ranking candidates with lightweight scoring, and consolidating into an impact heatmap, SAP architects create an evidence-based AI backlog that can survive platform changes and implementation debates.  Then, and only then, SAP’s adoption assets – Discovery Center missions, AI Core/Launchpad lifecycle governance, Joule where conversational augmentation fits, and SAP BTP AI Best Practices for secure GenAI design – can be activated to move from discovery to execution with reduced risk.

Some quick references that you should go through before such a discovery first exercise: