Back to Insights
AI in ERP: How SAP and Agentic AI Are Changing the Game
The companies that will fail with AI in ERP in the next five years will not fail because of SAP Joule, Microsoft Copilot or the EU AI Act.
/Herr_Dreher_500x500.jpg)
The companies that will fail with AI in ERP in the next five years will not fail because of SAP Joule, Microsoft Copilot or the EU AI Act. They will fail because of a specification that asks the wrong questions.
From our experience in over 1,200 ERP and digitalization projects in the DACH midmarket, a pattern is currently repeating itself: managing directors, IT managers and CFOs see demos of SAP Joule, Microsoft Dynamics 365 Copilot or Oracle Fusion AI Agents - and then start the same ERP selection they would have made in 2018. With function lists. With must/should criteria. With process diagrams that can't even describe agent-based behavior.
That no longer works. This article explains why - and what this means in concrete terms for your requirements elicitation, your ERP selection process and your AI strategy 2026.
Short answer: AI in ERP in one paragraph
AI is fundamentally changing ERP systems: instead of just managing data, modern systems make decisions independently. The example of SAP shows that companies need to rethink their requirements - away from function lists and towards decision logic, data strategy and agent-based processes. Traditional requirement specifications are no longer sufficient.
Executive summary
In the ERP context, AI is evolving from a pure analysis function to an operational control mechanism. The example of SAP - specifically Joule, Joule Studio and the Joule Agents rolled out from Q1/2026 - shows this:
-
ERP systems are evolving from transactional systems to agent-based platforms.
-
Decisions are increasingly automated and context-based.
-
Requirements are shifting from functions to capabilities and interactions (agent behavior).
-
Traditional requirement specifications fall short - AI-native requirements concepts are needed.
Our classification
Companies that do not actively shape this transformation risk structural inefficiencies and a strategic backlog that can no longer be made up in a normal investment cycle from 2027. By using SCOReX® or a similarly structured, vendor-independent methodology, you can get a head start at precisely this point.
1 Why are ERP systems evolving so fundamentally now?
For decades, ERP systems have stood for stability and transaction security. Traditionally, they fulfill three core functions:
Previously - System of Record
-Documentation(system of record)
-Processing (System of Transaction)
-Reporting(System of Insight)
→
From now on - System of Action
-AI agentsmake operational decisions
-Processesare orchestrated dynamically
-Decisionswithin defined guard rails
With the integration of AI - especially through solutions such as SAP Joule, supplemented by Joule Studio (configuration of own agents) and the Joule Agents rolled out productively from Q1/2026 - this logic is fundamentally shifting. ERP is moving from a "system of record" to a "system of action".
Practical example from production
In the past: dispatcher plans a production order manually, checks availability, calls sales, agrees a new delivery date.
Today: an AI agent recognizes the bottleneck, simulates three alternatives along delivery reliability, margin and stock, prioritizes automatically and triggers the adjustment. A person confirms - or escalates.
Our classification
The break is not gradual, but categorical. A "system of action" requires different requirements than a "system of record". Anyone who does not reflect this in their specifications in 2026 will be buying an outdated system in 2027.
2 Which AI developments are really relevant for ERP systems?
Not every "AI function" in ERP marketing is equally important. We differentiate - and recommend that you do the same in your evaluation matrix - between three maturity levels.
Level 1
Embedded intelligence
AI supports classic processes: Forecasting, scheduling, pricing, anomaly detection. Focus: increasing efficiency without structural change.
Now standard in SAP S/4HANA, Microsoft Dynamics 365, Infor CloudSuite, proAlpha.
Level 2
Augmented decision making
AI as co-pilot. Interaction via natural language. Example: "Which customers will jeopardize our cash flow in the next quarter?" - Answer in seconds, with source citation.
Differentiation sh ifts from AI model to data architecture.
Stage 3
Agentic AI - the real disruption
AI acts independently within defined guard rails. End-to-end processes semi-autonomous or fully autonomous - from defect reporting to invoicing.
Structural break. The disruptive lever of the next three years.
In the competitive context, we are currently observing five serious movements on the market:
SAP Joule + Joule Agents
Tightest integration in S/4HANA, part of the SAP Business AI stack.
Microsoft Dynamics 365 Copilot
Strength in Microsoft 365-savvy SMEs.
Oracle Fusion AI Agents
Tier 1 competition, in DACH primarily with larger SMEs.
Infor Coleman AI - IFS.ai
Industry-specific depth - service and asset management.
abas - proAlpha
German SMEs, AI roadmaps 2026 in implementation - pragmatic.
Our classification
The provider question is of secondary importance today. Anyone who starts the selection process with the provider question has already set up the wrong funnel. We recommend - see section 6 - taking the opposite route via SCOReX®.
3 Why classic specifications are no longer sufficient for modern ERP projects
To this day, most ERP selection projects are based on function catalogs, process descriptions and must/should criteria. This logic is structurally incompatible with agent-based systems. We see this in every selection process.
Classic specifications ask
“What should the system be able to do?”
AI-native ERP requires answers to
•Which decisions may the system make on its own?
•How much autonomy is permitted per process step?
•Which data may be used — and which may not?
•How are wrong decisions detected, escalated and reversed?
•How is auditability under the EU AI Act ensured?
Core problem: Functions are static. AI systems are dynamic. A static requirements format cannot fully specify dynamic behavior.
Our classification from project practice
In a project with a medium-sized medical technology company, we experienced that a 280-page specification sheet had to be restructured three weeks before the contract was signed - because the process offered by the provider was completely agent-based and not a single escalation path was specified in the specification sheet. Our SCOReX® model, which we explain in section 8, closes precisely this gap.
4 Which requirement dimensions companies need to redefine today
A future-proof AI-native ERP requirements concept covers at least six dimensions - we work strictly MECE so that nothing is duplicated and nothing is forgotten.
4.1
Decision logic
Which decisions can the AI make independently? Where are the intervention and escalation points? What reversibility is required?
4.2
Agent behavior
Which triggers activate which agents? Which actions are permitted - order, delivery date, customer communication? Which escalation logic applies?
4.3
Data architecture
Which data sources, in which quality? Who is the data owner, who is the data steward? How is data governance ensured across the ERP boundary?
4.4
Integration capability
Connection to APS, MES, IoT platforms, supplier and customer portals. Real-time capability. Interface stability for AI-driven schema changes.
4.5
Governance & compliance
Traceability (decision logging). Auditability. AI Act compliance - operational obligations apply from Q3/2026.
4.6
Economic impact
Which KPIs does the AI demonstrably improve - delivery capability, inventory, DSO, EBIT margin? How is success measured - business case per agent, not per license.
Our classification
In every ERP selection process that we currently accompany, dimension 4.1 (decision logic) and 4.5 (governance) are the two that are missing in 90% of existing specifications - and that will become expensive later on.
5 Trends 2026: What's really moving in AI in ERP right now
Three movements on the market that will be felt by SMEs instead of just appearing in analyst reports from Gartner, IDC or Forrester:
Trend 1
Agentic ERP goes live
SAP Joule Studio, Microsoft Copilot Studio, Oracle AI Agents - building agent-based processes is becoming manageable for internal IT for the first time. Competition is shifting from model quality to the speed of agent production.
Trend 2
The EU AI Act takes effect operationally
What was adopted as a regulatory framework in 2024 will become operationally mandatory in 2026. ERP agents in procurement, HR or production-critical processes are often high-risk. Transparency, documentation, supervision - must be included in the specifications before the system is selected.
Trend 3
Composable AI stacks
Vector databases, RAG pipelines, agent orchestration and domain-specific models are replacing monolithic AI modules. Consequence: Openness of the data architecture beats depth of the AI function.
Our classification
Anyone who chooses an ERP in 2026 that does not openly support these three trends is buying into structural path dependency.
6. Practice: How this development will have a concrete impact in the manufacturing industry
In discrete manufacturing - such as variant production in mechanical and plant engineering - we see the following pattern:
Initial situation
•28–42% of operational planning still runs in Excel today (SCOReX baseline 2025/26).
•High level of manual coordination between sales, scheduling and production.
•Reactive control instead of proactive planning.
→
AI-based target image
•Automated planning via S&OP and APS, integrated with the ERP master.
•Real-time adjustment in the event of disruptions (supplier failure, machine downtime).
•Consistent data foundation right through to the workbench — and back.
Concrete effects from our projects
20-35 %
Reduction in planning effort
+5-12 %
Increase in delivery capability (percentage points)
8-18 %
Reduction in inventories
Our classification
The added value is not created by a single AI function, but by the interaction between planning, scheduling, production and service. Introducing Agentic AI in just one functional silo raises 20-30% of the potential value.
→ See specific customer projects from DACH SMEs - anonymized, with figures.
7. AI in ERP for your sector — three practical views
AI requirements don't transfer one-to-one. Sector structure changes which dimensions need to carry the most weight.
8. How Dreher Consulting methodically addresses this development with SCOReX®
The central challenge is not the technology. The central challenge is the structured definition of requirements - before the system question is asked.
This is precisely why we developed our AI-supported SCOReX® model.
What SCOReX® does
•Structured elicitation of requirements beyond classic functions — including decision logic, agent behaviour and governance.
•Integration of process logic, architecture and AI capabilities into a single requirements view.
•Derivation of a future-proof, implementable specification that is compatible with agent-based systems.
Concrete added value from our project experience
25-35 %
less actual analysis time
Higher quality
through cross-validation with market and supply chain data
Less risk
through early clarity about escalation and governance paths
Our classification
SCOReX® is not "just another methodology". It is the structured answer to the fact that function lists cannot describe dynamic system behavior. And we use it as an owner-managed, vendor-neutral consultancy with no reseller obligations to SAP, Microsoft, Oracle or any other manufacturer.
9. Why the system question is the wrong first question - and how SCOReX® turns this around
From a First Principles perspective, an ERP selection is a chain of decisions. If you start with the system question, you optimize for one variable - the vendor - and ignore the other seven. We reverse the order:
1
What are your relevant decisions - operational, tactical, strategic?
2
Which of these decisions should be AI-supported or AI-autonomous today or in three years' time?
3
What data is required for this - and what is missing?
4
What governance requirements arise from the EU AI Act, industry regulation, internal risk policy?
5
Which integration architecture will support the next ten years?
6
This is where the provider question begins.
Which provider clusters fulfill exactly this profile - evaluated on a provider-neutral basis?
7
Which two to three providers will be included in a structured final selection?
8
Which provider decision is made at the end - on a reliable, comparable basis?
Only question 6 is the provider question.
The first five will determine whether the chosen system will still be viable in 2030.
Our ranking
This reverse order is not academic. We see in every project that the order determines whether a medium-sized company will be operationally scalable in five years - or stuck in an ERP lock-in.
→ Read more about vendor-independent ERP strategy.
10 Strategic implications for decision-makers
Development is not optional. It is structural. Today, managing directors in the DACH SME sector are faced with three realistic options for action:
Option A
Wait and see
Stable in the short term, risky in the medium term, structurally disadvantageous in the long term. Those who do not start defining AI-native requirements in 2026 will start from a weaker position later on.
Option B
React technology-driven
Fast, but risky. A selection that starts with the provider question optimizes on the front end - and underestimates architecture, data and governance implications.
Option C - recommended
Architecture and requirements-driven transformation
Structured requirements elicitation first. Then architecture. Then provider selection. Result: sustainable competitiveness and a comparable, reliable basis for decision-making.
Our classification
Option C is slower to start with - and faster to implement. We know from over 1,200 projects: The speed of implementation is determined by the quality of the requirements, not by the speed of the vendor decision.
11 Conclusion: The real change is methodical - not technological
The introduction of AI in ERP systems is not purely an IT issue. It concerns decision-making logic, process design, organizational structures and control logic at its core.
Core thesis
The companies that will be successful in 2026-2030 are not those with the best software - but those with the best requirements concept for AI-based systems.
Those who focus on supplier competition have already left the actual race. The 2026-2030 competition will be decided by requirements quality, data architecture and governance maturity - not by model selection.
12 Next step: 30 minutes with Dr. Dreher
If you want to assess the extent to which your company is prepared for this development, the most sensible way to start is not with a system discussion. Rather, a structured assessment of your requirements, decision-making logic and processes in the context of AI.
On this basis, reliable decisions can be made - independent of software providers.
What you actually take away
An initial classification of your ERP and AI maturity along the six SCOReX® dimensions, three prioritised fields of action and a clear recommendation as to whether a preliminary project makes sense — or whether you don't yet have a need.
13. frequently asked questions about AI in ERP
What is agentic AI in the ERP context - and how does it differ from "classic" AI?
Agentic AI refers to AI systems that make decisions and trigger actions independently within defined guard rails - such as orders, rescheduling appointments and customer communication. In contrast, classic AI in ERP (embedded intelligence) provides suggestions or forecasts that are confirmed by a human. The difference is not gradual, but categorical: from decision supporter to decision maker.
How does SAP Joule change ERP requirements in concrete terms?
SAP Joule integrates generative AI directly into S/4HANA processes. Companies can configure their own agents with Joule Studio. Requirements must map decision logic (What is the agent allowed to do?), escalation paths (When does a person take over?) and governance (How is documentation auditable?) - no longer just functional scopes.
Why are traditional specifications for AI-native ERP systems no longer sufficient?
Traditional specifications describe static functions. AI-native ERP systems behave dynamically - they decide, learn and escalate. A specification sheet without an explicit specification of decision logic, degrees of autonomy and governance paths leaves the main risks open in the subsequent contract.
What obligations does the EU AI Act impose on ERP users in SMEs?
ERP agents in production-critical, personnel-related or financial processes often fall into the high-risk category. This results in transparency, documentation and supervisory obligations. Operational implementation obligations take effect from Q3/2026. This belongs in every 2026 specification - not in a later compliance retrofit.
What is the difference between embedded intelligence, augmented decision making and agentic AI?
Embedded Intelligence - AI as a functional component in existing processes (e.g. forecast optimization). Augmented decision making - AI as a co-pilot that prepares options for a decision-maker. Agentic AI - AI as an autonomous actor that acts independently within defined guard rails. The economic leverage increases along this staircase - and with it the requirements.
What is SCOReX® and how does it differ from conventional requirements elicitation?
SCOReX® is the model developed by Dreher Consulting for structured requirements elicitation for AI-native ERP selection projects. In contrast to traditional methods, it addresses not only functions, but also decision logic, agent behavior, data architecture and governance - along a MECE structure that is cross-validated with market, competition and supply chain data. Typically reduces as-is analysis time by 25-35%.
How long does an AI-native ERP selection in SMEs typically take?
From our experience: 4-7 months from initiation to contract signing - depending on data availability, organizational maturity and the number of processes to be considered. SCOReX® shortens the as-is analysis and requirements definition phase in particular. The longest variable is usually the internal decision-making organization, not the methodology.
How can I speak directly with Dr. Dreher about my ERP strategy?
Arrange a 30-minute initial meeting directly in the calendar. Based on ten targeted questions, you will receive an initial classification, three fields of action and a clear recommendation - without sales pressure. Make an appointment.
Found this helpful? Let’s explore how these insights could benefit your business.
Click the Contact Us button to connect with me directly.
