Organisational Design · 2026

How to Build a SAP AI Centre of Excellence in 2026:
Team, Governance & Operating Model

June 28, 2026 12 min read SAVI AI Research Team
AI CoE SAP Governance Digital Transformation SAP BTP

A SAP AI project that succeeds in isolation is a proof of concept. A SAP AI project that scales across the enterprise is a Centre of Excellence. The difference between the two is not technology — it is organisational design.

Companies that deploy one successful SAP AI use case and then wonder why the next three projects stall, the fourth gets cancelled, and the fifth has to reinvent the wheel are missing a CoE. This guide gives you the exact structure, roles, governance cadence, toolchain, and 6-month launch roadmap used by the SAP AI programmes that consistently scale.

The multiplier effect: Enterprises with a formal AI CoE deploy new SAP AI use cases 3.2× faster, at 40% lower cost, with 28% higher first-year adoption than those without one. The CoE pays for itself on the second use case it delivers.

3.2×
faster deployment with AI CoE vs. without
40%
lower per-use-case cost after CoE is established
73%
of enterprises plan an AI CoE by end of 2026
6 mo
to minimum viable CoE from zero

Step 1: Choose Your CoE Operating Model

Before hiring a single person, decide which operating model fits your organisation. Each has fundamentally different implications for headcount, reporting lines, and decision-making authority.

🏛️

Centralised CoE

One central team owns all AI development, deployment, and governance. Business units submit requests; the CoE delivers. Common in highly regulated industries.

Maximum quality control and consistency
Easier EU AI Act compliance
Can become a bottleneck at scale
Risk of low business-side ownership
🌐

Federated CoE

Each business unit has its own AI team, loosely coordinated by a central standards body. Maximum speed but minimum consistency — common in large multinationals.

Business units move fast
High local ownership
Inconsistent quality and duplicated effort
Governance gaps — compliance risk

Step 2: The Core Team — Roles & Hiring Guide

The minimum viable SAP AI CoE for a mid-market enterprise needs these 7 roles. Not all need to be full-time from day one — the phasing section shows how to build incrementally.

AI CoE Lead

Head of AI / Director Level

Owns the AI strategy, budget, roadmap, and board-level reporting. Sets priorities across use cases. The single accountable leader for AI business outcomes.

SAP strategy Stakeholder mgmt P&L accountability
1.0 FTE

SAP BTP Architect

Senior / Principal Level

Owns the technical architecture on SAP BTP — integration patterns, security model, AI Core configuration, and the API layer between AI and SAP backends.

SAP BTP AI Core OData / APIs
1.0 FTE

AI / ML Engineer

Mid–Senior Level × 2

Builds, trains, and monitors AI models. Handles prompt engineering for LLM use cases, fine-tuning, RAG pipelines, and model performance optimisation post-go-live.

Python / ML LLMs / RAG SAP AI Core
2.0 FTE

SAP Data Engineer

Mid–Senior Level

Owns the data pipeline from SAP (S/4HANA, ECC) into the AI layer. Responsible for data quality remediation, SAP HANA Cloud, and the vector store for RAG use cases.

SAP HANA Data quality ETL / pipelines
1.0 FTE

Business Analyst (Process)

Mid Level × 2 (Finance + Procurement)

Bridges between AI capability and business process. Documents current-state workflows, defines exception logic, writes user acceptance test scripts, and owns adoption metrics.

SAP FI/CO Process mapping UAT / training
2.0 FTE

AI Governance & Risk

Senior Level (50% FTE initially)

Owns EU AI Act compliance, model risk management, audit trail governance, and the ethical AI framework. Interfaces with Legal, Compliance, and the DPO.

EU AI Act Risk management GDPR / DPA
0.5 FTE

Change Management Lead

Mid–Senior Level

Drives adoption across business units. Runs the champions network, designs role-specific training, monitors adoption KPIs, and surfaces resistance before it kills go-lives.

Change management Training design Comms & engagement
1.0 FTE

Hiring order matters: Hire the AI CoE Lead first — they hire everyone else. Hire the BTP Architect second (they unblock technical progress). Hire ML Engineers third. Add the Change Management Lead before your first go-live, not after. Skipping change management is the most expensive false economy in an AI CoE build.

Step 3: The Governance Model — Who Decides What

Governance is not bureaucracy — it is the mechanism that stops every use case from becoming a political fight. Clear decision rights prevent the two most common CoE failures: the project that launches without IT sign-off, and the project that IT kills without business input.

ForumAttendeesDecisions MadeCadence
AI Steering Committee CIO, CFO/COO, AI CoE Lead, CISO Strategic priorities, budget allocation, portfolio approval, risk escalations, board reporting Quarterly
CoE Operating Board AI CoE Lead, BTP Architect, BAs, Process Owners Use-case backlog prioritisation, go/no-go decisions, resource allocation, milestone reviews Monthly
Technical Review BTP Architect, ML Engineers, Data Engineer, IT Security Architecture decisions, vendor selection, security sign-off, model accuracy thresholds Bi-weekly
Project Stand-up Delivery team (implementation-specific) Daily blockers, task progress, escalations within implementation sprints Daily (in-flight)
Adoption Check-in Change Management Lead, Business Champions, Process Owners Adoption rate review, resistance identification, training gaps, feedback loop Weekly (first 90 days)
Model Health Review ML Engineers, Data Engineer, Governance & Risk Accuracy drift alerts, retraining decisions, EU AI Act audit evidence collection Monthly

Step 4: The AI CoE Toolchain

Standardise your toolchain early — it is significantly harder to migrate between platforms mid-programme than to start on the right stack. These are the tools used by the most efficient SAP AI CoEs in 2026.

AI Platform

SAP BTP AI CoreModel hosting & inference
SAP AI LaunchpadModel lifecycle management
SAVI AI Agent LayerAgentic orchestration
SAP JouleNative SAP AI assistant

LLM & AI Models

Claude (Anthropic)Complex reasoning & analysis
GPT-4o (Azure)Document processing
SAP Document AIInvoice & contract OCR
Custom ML modelsClassification / anomaly

Data & Storage

SAP HANA CloudPrimary data store + vector
SAP Data IntelligenceData quality & pipelines
SAP Integration SuiteAPI & event integration
SAP DatasphereEnterprise data fabric

Monitoring & Governance

SAP AI Launchpad metricsModel accuracy dashboard
Azure Monitor / DatadogSystem health & alerts
JIRA / Azure DevOpsUse-case pipeline tracking
Power BI / SAP AnalyticsSteering committee KPI board

Step 5: The 6-Month CoE Launch Roadmap

Mo 1–2

Foundation: Team & Platform

  • Hire AI CoE Lead & BTP Architect
  • Stand up SAP BTP tenant (AI Core, Integration Suite)
  • Draft AI governance charter & decision-rights RACI
  • Convene first Steering Committee meeting
  • Shortlist Wave 1 use cases (score & prioritise)
  • Conduct SAP data readiness assessment
Mo 3–4

First Delivery: Wave 1 Use Case in Shadow Mode

  • Hire ML Engineers & Data Engineer
  • Wave 1 use case: data extraction & model training
  • Shadow mode enabled — accuracy tracking begins
  • EU AI Act risk classification documented
  • Publish AI standards & development playbook v1
  • Launch champions network in Finance & Procurement
Mo 5–6

First Go-Live & Wave 2 Pipeline

  • Wave 1 production go-live
  • First ROI measurement report to Steering Committee
  • Hire Change Management Lead & BAs
  • Wave 2 use-case scoped & approved
  • Reusable component library documented (accelerators)
  • CoE operating model presented to board for funding renewal

Step 6: CoE-Level KPIs to Report to the Board

A CoE that cannot prove its business impact will lose funding within 18 months. Report these five metrics to the Steering Committee every quarter — and publish them to the wider organisation.

KPITarget (Year 1)Target (Year 2)Reporting Cadence
Use cases in production2–38–12Quarterly
Total annualised ROI delivered3–5× CoE cost8–12× CoE costQuarterly
Avg. use-case delivery time (pilot→prod)≤90 days≤60 daysQuarterly
Portfolio AI accuracy (weighted avg.)≥92%≥95%Monthly
User adoption rate (live tools)≥70%≥85%Monthly

What Does a SAP AI CoE Cost?

Budget transparency prevents the most common CoE failure mode: being funded for the platform but not the people. Both are essential. Here are realistic cost ranges for a mid-market enterprise.

Cost CategoryYear 1Year 2+
People (7 FTE fully-loaded, blended rate €100k)€700,000€750,000
SAP BTP platform & AI Core licences€120,000€140,000
LLM API access (Claude, GPT-4o via Azure)€40,000€60,000
Tooling (monitoring, DevOps, BI dashboard)€30,000€25,000
Training & certifications (SAP BTP, AI Core)€25,000€15,000
SI / implementation partner (Wave 1 support)€100,000€50,000
Total Annual CoE Investment€1,015,000€1,040,000

With 3–5× ROI in Year 1 as the target, the CoE should deliver €3–5M in measurable business value to fully justify its cost. Use-cases with the fastest payback (invoice processing, GR/IR) fund the CoE's continued existence — sequence them first.

5 AI CoE Anti-Patterns That Guarantee Failure

Anti-pattern 01

The "Build It and They'll Come" CoE

Team built, platform deployed, governance documented — then waiting for business units to submit requests. Without proactive engagement, the queue stays empty.

The CoE Lead must actively identify and sponsor Wave 1 use cases from day one. Don't wait to be asked.
Anti-pattern 02

Platform Without People

Fully funded SAP BTP deployment with two engineers and no change management, no business analyst, and no data engineer. The platform sits idle.

Budget 65–70% of CoE cost to people, 30–35% to platform. The ratio reverses many organisations' instincts.
Anti-pattern 03

Governance Theatre

Governance documents exist but decisions are still made informally. The Steering Committee meeting becomes a status update with no actual decisions.

Every Steering Committee meeting must make at least one prioritisation decision — force it into the agenda template.
Anti-pattern 04

The Invisible CoE

The CoE team works hard but nobody in the business knows what they've delivered. No internal comms, no success stories, no visible KPI dashboard.

Publish a monthly "AI Impact Report" to all senior leaders — even a 1-pager showing live use cases and ROI delivered.
Anti-pattern 05

Perfection Before Production

Spending 6 months refining the governance framework, the toolchain architecture, and the operating model before delivering a single use case to production.

Launch Wave 1 in parallel with governance build. The fastest way to improve your operating model is to run it.

Building Your SAP AI CoE? We've Done This 30+ Times.

SAVI AI can accelerate your CoE from zero to first production use case in 90 days — providing the BTP architecture, AI model library, and change management playbook so you don't start from scratch.

Book a Free CoE Design Session

FAQ

How big should a SAP AI Centre of Excellence be?
A minimum viable AI CoE for a mid-market enterprise needs 5–8 people: an AI Lead, 2 SAP BTP engineers, 1 data engineer, 1 change management lead, and 1–2 business analysts embedded from Finance and Procurement. Large enterprises (€5B+) typically run 15–25 person CoEs with domain-specific squads per process area.
Should the SAP AI CoE report to IT or the business?
The most successful model is hybrid: the CoE sits in IT for technical governance but has a dotted line to the CFO or COO for business accountability. A joint steering committee with both IT and business representation, meeting quarterly, balances both needs. Purely IT-owned CoEs struggle with adoption; purely business-owned CoEs struggle with technical quality.
What is the difference between an AI CoE and an AI factory?
An AI CoE governs standards, selects tools, owns the AI platform (SAP BTP), manages risk, and enables business units to build AI solutions. An AI factory is a delivery engine — a team that builds AI use cases at scale using repeatable patterns defined by the CoE. Mature enterprises have both: the CoE sets the rules and the factory builds at speed.
How long does it take to set up a SAP AI CoE?
A minimum viable AI CoE can be operational in 6 months: months 1–2 for team formation and toolchain setup, months 3–4 for governance documentation and first use case delivery, months 5–6 for Wave 2 pipeline and stakeholder reporting. Waiting for the 'perfect' team before starting is the most common mistake — launch with a core team of 5 and grow based on demand.
What tools does a SAP AI CoE need?
The minimum toolchain: SAP BTP (AI Core, AI Launchpad, Integration Suite), a vector database for RAG use cases (SAP HANA Cloud vector store), a model management platform, a data quality tool (SAP Data Intelligence), a project management tool (JIRA or Azure DevOps), and a central KPI dashboard. Add LLM access (Claude, GPT-4o) via BTP AI Core for generative use cases.
How do I measure the success of a SAP AI CoE?
Track five CoE-level metrics quarterly: (1) Number of AI use cases in production, (2) Total annualised ROI delivered, (3) Average time from use-case approval to production go-live, (4) AI model accuracy across the live portfolio, (5) User adoption rate. Present these to the steering committee each quarter — a CoE that cannot report its business impact will lose funding within 18 months.