SAP AI: Why the Return on AI Is Decided in the ERP

What SAP AI Means Today

The term SAP AI covers several building blocks that are often lumped together. SAP Business AI refers to artificial intelligence embedded in business processes and applications, for example in finance, procurement, sales or HR. Joule is the AI assistant in the SAP environment. It is operated in natural language and uses data from supported applications. The SAP Business Technology Platform (BTP) is the platform on which companies build extensions, integrations, services and their own AI applications.

For decision makers this distinction matters, because each building block has different prerequisites. Embedded AI functions need an ERP on a suitable release. An assistant needs clean, properly authorized data. Custom AI solutions need an architecture that brings data from several systems together.

All three share one condition: the intelligence works only as well as the data and processes it can reach. That is why the question of return on AI at SAP is first of all a question about the ERP.

Typical Application Areas

In practice, the application areas with many recurring decisions show up first. Machine learning and predictive analytics produce forecasts, for example for demand, liquidity or payment behavior. An AI agent takes over partial steps, such as the pre-check of invoices, and hands exceptions to people. Joule helps with analyses and reports. In data analysis across large volumes, AI spots patterns and deviations that disappear in a spreadsheet view.

All these applications access the system landscape, whether on-premise, in the cloud or hybrid. How well that works depends on the IT architecture: Is data consistent, are interfaces documented, is it clear who may change what?

Possibilities and Limits

The possibilities are real, but not unlimited. An AI agent can only take over tasks whose rules are described. A forecast is only as reliable as the historical data it learns from. And no vendor relieves you of deciding which processes in the company are suited to AI at all. That is a task for business and IT together.

Why the Return on AI Is Decided in the ERP

Many SAP AI projects start with the question of which tool to license. That is the wrong order. The white paper Return on AI: Why AI Only Pays Off in the Value Stream by CTG Consulting states that AI speeds up the single task, but time is lost at the handovers between departments. The first effect often comes from an organizational decision, not from a model. In the ERP this is especially visible, because that is where value streams run together, from the customer inquiry to the incoming payment.

Licenses Are the Visible Part

License and usage costs appear on the quote and can be budgeted. That is why they dominate the discussion. But they say little about everyday usage, and even less about whether a function has any effect. An AI function that is available but unused, because the prerequisites in the system are missing, delivers no return.

Data and Integration Are the Real Effort

The larger part of the effort arises elsewhere:

1. Data quality: Master data with duplicates, incomplete fields or inconsistent naming leads to unusable forecasts and suggestions.
2. Data migration: Which legacy data is carried over, cleansed or archived? Every decision here shapes what the AI can see later.
3. Integration: Between ERP, CRM, procurement platforms and specialist applications sit interfaces. Time is lost at these handovers, and that is exactly where AI comes in.
4. Authorizations and governance: An assistant may only use data the user is allowed to see. The authorization concept decides whether a use is permissible at all.
5. Operations and ownership: Who maintains models, rules and data quality once the project team has moved on?

None of these points is an AI question in the narrow sense. All five decide whether AI in the ERP pays off. There is no rule of thumb with fixed percentages for this, and anyone who promises one should be able to prove it.

This becomes especially clear with customer and supplier data. If customers exist in the ERP, in the CRM and in a sales spreadsheet under slightly different names, no AI can reliably say which customer brings how much revenue. The technologies are available. The consistent data foundation is missing.

S/4HANA Migration Before the End of Maintenance in 2027

SAP has announced that standard maintenance for ECC 6.0 ends at the end of 2027. According to SAP, a paid option for extended maintenance runs until 2030. You should check the terms directly with SAP or your partner. For many companies this means the move to SAP S/4HANA is decided, and only the how remains open.

This is the real link to AI. An S/4HANA migration that merely lifts the existing setup onto a new platform moves the data problems into the new system. If it is understood instead as an opportunity to build a shared data foundation, it creates the basis for end-to-end value streams. The move is therefore more than a technical upgrade: it is a transformation project that touches processes, data and responsibilities.

Three Paths, Three Consequences for the Data Foundation

• Greenfield: A new build with selected data carried over. High effort, but the biggest chance to clean up processes and data.
• Brownfield: Conversion of the existing system. Faster and less invasive, but it also carries over accumulated legacy.
• Selective transformation: A hybrid in which individual areas, entities or data sets are moved deliberately.

Which path fits depends on process variety, custom developments, data situation and timeline. From an AI perspective the test question is: which path gets us, by the cutoff date, the data we need for the first value stream?

Clean Core Instead of Sprawl

The more custom development sits in the core of the ERP, the more costly upgrades become and the harder it is to connect AI functions. The principle of a lean core, where extensions are built through interfaces and the BTP rather than in standard code, is therefore also an AI decision. It also creates flexibility for future releases and limits the technical debt that otherwise grows with every release.

Planning Migration and AI Together

Many companies treat migration and AI scenarios as separate projects with separate budgets. The result is that data models are touched twice. It is more sensible to define during migration planning which one or two value streams should benefit from AI first, and to prioritize data cleansing exactly there.

From Use Case to Value Stream: What It Means for SAP

The white paper describes four steps: choose, measure, redesign, collect. Applied to an ERP environment, this gives a clear sequence.

Choose. Instead of launching ten use cases in parallel, focus on one value stream with a results lever and an owner. The white paper scores candidates on five weighted criteria: results lever, friction from handovers, data situation, ownership and feasibility, meaning a first measurable effect within four to six weeks. No owner, no value stream. In SAP landscapes, processes such as invoice verification, quotation or demand planning are often good candidates because they span several departments.

Measure. Before building, take the baseline, separating working time from waiting time. The white paper works this out in an explicitly illustrative example from mechanical engineering (inquiry to quotation): of 8 days of lead time per quotation, only about 6 hours were real work. The rest was waiting at handovers. Without measuring, you automate the wrong step. Timestamps per document status from ERP and CRM are often enough as a data source, according to the white paper even with older systems such as SAP ECC.

Redesign. The order is simplify, automate, orchestrate, and for each process step the solution with the best business case counts, which is not always AI. In the illustrative example the quotation goes out on the same day instead of after 8 days, and handling time per quotation falls from about 6 to 2 hours. The first effect came from a higher approval limit, so from a decision by the owner. Only both together deliver the full effect.

Collect. The business case is calculated in win rate, margin and cash, not in hours saved. Freed capacity only counts once it is deliberately reassigned. For management decision making this means: instead of a promise, there is a figure that shows up in the result.

The link to the ERP is obvious: all four steps need reliable process data. Where data from five systems is maintained differently, even the baseline fails. In the white paper's maturity model, the return begins with the jump from level 2 to level 3, where AI no longer just speeds up one department but changes an entire process. For the step after that, it names a shared, maintained data foundation. It also points out that the value lies in the core business, while the better return often comes from the back office.

Return on AI in the ERP

From use case to value stream

1. 01ChooseOne value stream with an owner instead of ten use cases.
2. 02MeasureBaseline before building, separating working time from waiting time.
3. 03RedesignSimplify, automate, orchestrate.
4. 04CollectWin rate, margin and cash instead of hours saved.

Maturity model of the white paper

Level 2AI speeds up one department.

The return starts here

Level 3AI changes an entire process.

5 Checks Before the First AI Use Case in SAP

1. Is there an owner for the value stream? One person who is accountable end to end and may make decisions.
2. How good is the master data really? A sample across customer, supplier and material master often shows faster than a workshop where cleansing is needed.
3. Which interfaces are in the way? Which systems outside SAP send or receive data in this process?
4. Do release and operating model fit? Not every AI function is available in every release and operating model. Clarify this before planning.
5. Who runs the result? Maintaining data quality, authorizations and rules needs permanent ownership.

Which Roles Are Missing

Technology alone does not solve any of these problems. What matters is who understands the processes and who owns the data. In practice, three profiles are missing most often for SAP AI:

• SAP architects with process knowledge who think migration, clean core and integration together.
• Data owners who steer quality, responsibilities and authorizations in the ERP and, according to the white paper, ensure the shared data foundation is sound.
• Business process owners (called value stream owners in the white paper) who are accountable for a value stream across department boundaries.

Whether a role is filled permanently, by a freelancer or on an interim basis depends on how long the need lasts. For the migration phase, external support often makes sense, while data ownership and process ownership are usually better served by a permanent position. As an IT recruitment consultancy, Paltron supports the search, while process accountability stays within the company.

Frequently Asked Questions

Can AI replace SAP?

No. AI does not replace an ERP, it builds on it. The ERP provides the transactions, master data and process logic that AI draws on. Without a clean ERP, AI lacks a reliable data foundation.

Do I have to move to the new ERP generation for AI functions?

Some AI functions are tied to current releases and cloud offerings. Whether a specific use case works with your existing system depends on the individual case and should be clarified with SAP or a partner before planning. Independently of that, ECC customers face the end of maintenance.

How can migration and AI be timed together?

A staged approach makes sense: first, plan data cleansing and a lean core, because both are needed either way. In parallel, choose one value stream where the benefit of AI can show early. Which functions are usable when depends on release and operating model and should be agreed with SAP or a partner.

When does SAP ECC maintenance end?

SAP has announced the end of standard maintenance for ECC 6.0 for the end of 2027. According to SAP, extended maintenance is possible for an additional fee. Ask SAP for current terms.

What does it cost to get started?

There is no flat answer. License and usage costs are only one part. The effort for data cleansing, data migration, integration, authorizations and operations varies widely depending on the starting point.

Which data does the first use case need?

That depends on the chosen value stream. For invoice verification it is the supplier master, purchase orders and goods receipts, for a demand forecast it is movement data and sales history. What matters is that the data is complete and consistent for exactly this process.

Are you planning your start with SAP AI and unsure who should carry it? We will talk through your starting point. The white paper Return on AI provides the methodological framework.

IT headhunter

Fill the missing SAP roles

Are SAP architects, data owners or process owners missing? We clarify with you whether a permanent hire, freelancer or interim fits.

Get in touchView profiles