Legacy ERP modernization with AI: what the old system gets, and what to build beside it
Companies on an old ERP are asking whether it will ever take part in AI. The vendor’s answer comes with conditions. The alternative is to stop waiting for the ERP and give the company an AI and data layer of its own.
Contents
The search phrase behind this guide is some version of "does my old ERP get AI". It is a reasonable question from a company that has paid a support fee for fifteen years. The answer from the largest ERP vendor has changed twice in three years, and each version says something about who controls the pace.
What the vendor has said
In July 2023 SAP’s chief executive told investment analysts: "Our new innovations will not be available for on-premise or hosted on-premise ERP customers on hyperscalers." A spokesperson added that capabilities around generative AI would be available only in the cloud1. The German-speaking user group’s chairman replied that customers who had relied on S/4HANA on-premise "will be left behind", and the group reported a 30% additional cost for the premium package that carried the new functions1.
In May 2026 SAP changed course in part. At its Sapphire conference it said a significant share of its Joule assistants and agents would work on-premises, on ECC and S/4HANA, described by the chief executive as "an interim solution so that they can benefit from AI while they are modernizing"2. The conditions matter more than the headline:
| Condition | Source |
|---|---|
| The customer has started its migration on RISE with SAP | SAP’s board member for product and engineering2 |
| The customer holds the Max Success Plan, a commercial agreement | The Register2 |
| At least 50% of maintenance spending has shifted to the cloud | CIO.com3 |
| No extension of ECC mainstream maintenance beyond 2027 | SAP spokesperson3 |
SAP’s chief strategy officer summed up the position: "our Joule assistants and agents are designed for cloud. This is destiny and destiny is unchanged"3.
What customers are doing instead
They are not waiting. In the DSAG Investment Report 2026, a survey of 198 member companies in Germany, Austria and Switzerland, 43% had implemented AI use cases. Among those, the solutions in production were non-SAP in 77% of cases and SAP in 3%; in proof of concept the split was 62% to 26%4. Computerwoche reported the same figures for German readers5.
The French-speaking user club made a related point at its 2025 convention. Its president asked SAP to make the cloud simpler and cheaper, and the club considers a move of ECC into the cloud hard to envisage for many members6.
Why an added assistant is not the same as an AI-native system
An assistant inside an ERP sees what the ERP holds: transactions in the vendor’s tables. The work of a company is wider than that. An order begins as an email or a phone call, picks up a price exception in a chat, and becomes a transaction only at the end. An AI that can act on the whole chain needs the correspondence, the calls, the documents and the transactions in one place, under rules the company sets.
That is the difference this guide means by AI-native: one AI and data layer under every module, owned by the company, in which people and agents learn from each other. OpenLazarus describes the opposite pattern, in its list of what keeps companies behind, as adding AI tools without connecting them to existing systems, leaving knowledge locked in individual inboxes and documents, and continuing to subscribe to software platforms that hold the company’s data.
Four ways to modernise a legacy ERP
| Approach | What changes | What stays |
|---|---|---|
| Rehost | The old ERP moves to cloud infrastructure | The software, its structures and its limits |
| Convert | The old ERP becomes the vendor’s current product | The vendor, the rental model, the vendor’s timetable7 |
| Layer | An AI and data layer the company owns goes in beside the ERP and takes over processes one by one | The ERP as system of record, for now |
| Replace by module | Modules on the owned layer take over orders, stock, purchasing and finally the ledger | Nothing of the old system, at the end |
The third and fourth approaches are one path at two points in time. The layer comes first because it produces results while the old ERP still runs, and because it turns the eventual replacement into a series of small moves.
The plan
- Pick the process with the most hand-work around the ERP. Order entry from email and phone, supplier follow-up and customer service are the usual candidates.
- Put the data for that process in a layer the company owns. Transactions read from the ERP, plus the mail, calls and documents that never reached it.
- Give an AI agent a defined job in that process, with the actions that need human approval named in advance and a person accountable for the result.
- Measure, then widen. When the first process holds its numbers, add the next. The sequence is in the migration plan.
- Decide the ERP’s future from a position of strength. Once the daily work runs on the owned layer, the old ERP is a ledger. Whether to keep it on cheaper support, or replace it with the owned system’s ERP module, becomes a budget question. The tests are in upgrade or replace.
The end state is a company whose AI works on all of its information and whose core system no longer depends on a vendor’s roadmap to improve.
Questions buyers ask
Does SAP ECC get AI?
On conditions. In May 2026 SAP said a share of its Joule assistants and agents would work on ECC and on-premises S/4HANA as an interim measure, for customers that have started a RISE with SAP migration and hold its Max Success Plan. CIO.com reported that a customer must first shift at least 50% of its maintenance spending to the cloud.
What is legacy ERP modernization?
Bringing an old ERP up to date without necessarily replacing it in one step. The approaches are rehosting the old system in a cloud, converting it to the vendor’s current product, adding an AI layer beside it, or replacing it module by module. They differ in whether the company ends up renting a newer suite or owning its system.
Can AI agents work with a legacy ERP?
Yes, when they run on a layer beside it that reads the ERP’s data and writes results back. That is how most companies already do it. In the DSAG Investment Report 2026, 77% of AI use cases in production at member companies ran on non-SAP solutions and 3% on SAP solutions.
Is SAP AI-native?
Not in the sense used in this guide. Joule is an assistant added to an existing suite, which SAP’s chief strategy officer describes as designed for the cloud. An AI-native system has one AI and data layer under every module, so people and agents work from the same records and learn from each other.
Why does an old ERP block AI transformation?
Because of where the data sits and who controls it. An ERP stores records in the vendor’s rigid structures, exposes them on the vendor’s terms and adds AI on the vendor’s timetable. AI work that needs orders, emails, calls and documents together has to be built outside it.
Sources
- User groups unhappy over SAP push on cloud-only ‘innovation’. The Register, 3 Aug 2023. Accessed 11 Nov 2026.
- SAP U-turn brings AI features to ECC and on-prem S/4HANA. The Register, 13 May 2026. Accessed 11 Nov 2026.
- SAP’s AI offer to legacy customers comes with a catch. CIO.com, 13 May 2026. Accessed 11 Nov 2026.
- DSAG Investment Report 2026. DSAG (German-speaking SAP User Group), 26 Feb 2026. Accessed 11 Nov 2026.
- Wirtschaftlicher Druck macht SAP-Anwender pragmatischer. Computerwoche, 27 Feb 2026. Accessed 11 Nov 2026.
- L’USF encore inquiète sur le cloud et la maintenance SAP. Le Monde Informatique, 15 Oct 2025. Accessed 11 Nov 2026.
- SAP S/4HANA: commitment to innovation and maintenance until 2040. SAP News Center, 4 Feb 2020. Accessed 11 Nov 2026.
Updated 11 Nov 2026: First published.