Most enterprise leaders I speak with today are asking the same question: when do we start using AI in our ERP? They mean Oracle. They mean SAP. And most of them are thinking about it entirely the wrong way.
The question should not be when to add AI on top of these platforms. The question is whether the platforms themselves — and more critically, the operating model around them — are architected to let AI actually do something meaningful. That is a very different conversation, and it starts long before a vendor demo or a proof of concept.
Let me be direct about what I am seeing in the field. Organisations that treat AI as a feature to be bolted onto their existing SAP or Oracle estate are getting the same result every time: modest automation gains, a handful of impressive dashboards, and no measurable change to how the business operates. The ones seeing real transformation are doing four things differently.
EVENT
Infosec Reimagined
Infosec Reimagined 2026 is the premier information security summit where top leaders—CISOs, CROs, CIOs, CTOs and risk executives—converge to redefine cyber resilience.
EVENT
CIO Prism
CIO Prism unites forward-thinking technology leaders to exchange transformative insights, shape digital strategies, and foster innovation, empowering enterprises to excel in an era of rapid technological change.
1. They Are Rethinking Architecture — Not Just Upgrading It
Moving SAP ECC to S/4HANA or Oracle EBS to Oracle Cloud Fusion is not modernisation. It is migration. And there is a crucial difference.
True modernisation means moving away from the assumption that a single platform can do everything. The AI era demands composable architectures — where ERP handles the core transactional record, and specialised AI models, data platforms, and domain-specific applications operate around it through clean API contracts.
CIOs who are still designing tightly coupled, monolithic ERP landscapes are building tomorrow’s legacy problem today.
The practical implication: before your next major ERP release upgrade, ask your architecture team to map every integration point and every customisation. If you cannot replace or evolve any component independently, your architecture is not AI-ready. It is AI-resistant.
2. Data Readiness Comes Before AI Readiness
AI models are only as intelligent as the data they reason over. And ERP data — in the vast majority of enterprises I have worked with — is a mess. Duplicate vendor records. Inconsistent cost centre hierarchies. Procurement data that does not reconcile with finance data. Master data that was set up during the original go-live and has never been meaningfully governed since.
You cannot build a reliable AI-powered demand forecasting model on top of inventory data with 30% record duplication. You cannot run an intelligent AP automation agent if your invoice matching rules have seventeen exceptions that only two people in the organisation understand.
Advertisement
The CIOs doing this well are investing in data governance before they invest in AI tooling. They are establishing data ownership at the business level, not just IT. They are cleaning master data as a strategic asset, not a technical task. And they are building data contracts between systems so that every downstream AI model knows exactly what it is working with.
3. Process Intelligence — Not Just Process Automation
There is an important distinction that many organisations miss when they talk about AI in ERP: the difference between automating a process and understanding one.
Traditional RPA-style automation executes a defined rule set. If X then Y. AI can do something different — it can reason about whether the process itself is producing the right outcome, identify where exceptions are clustering, and recommend process redesign, not just task completion.
SAP and Oracle both now embed AI capabilities — Oracle Fusion’s AI agents, SAP Joule — that go beyond automation. But to get value from these capabilities, enterprises need to redesign their process architecture, not simply automate the existing one. That means process mining before process automation. It means involving business operations in the design, not just IT. And it means measuring AI success by business outcome, not by tasks automated.
4. The Integration Layer Is Now a Strategic Asset
SAP and Oracle rarely exist in isolation. Most enterprises are running a combination of Oracle HCM, SAP supply chain, Salesforce CRM, a custom-built MES on the shop floor, and half a dozen SaaS platforms acquired through M&A. The AI layer needs to work across all of these — not just within one platform.
This makes the integration architecture one of the most strategically important decisions a CIO will make in the next 18 months. API management, event streaming, data virtualisation — these are no longer infrastructure concerns. They are the nervous system that allows AI agents to operate across your enterprise estate in real time.
CIOs who are still treating integration as a project cost rather than a platform capability will find that their AI ambitions stall precisely because the data flows do not exist to support them.
What This Means in Practice
The four shifts above are not sequential. They need to happen in parallel, with clear ownership and executive sponsorship. Based on what I have seen across large-scale transformation programmes, the organisations that are ahead have done the following:
- Separated the ERP upgrade conversation from the AI strategy conversation — and then deliberately reconnected them at the architecture level.
- Appointed a Chief Data Officer or equivalent who owns data quality as a business outcome, not an IT metric.
- Run a process intelligence exercise — using process mining tools — before committing to any AI automation roadmap.
- Invested in integration platforms that support event-driven architecture, not just point-to-point API calls.
The AI era is not a future state that enterprises need to prepare for. It is already determining which organisations can act on insight in hours rather than weeks, and which ones are still waiting for the month-end report.
SAP and Oracle have built genuinely powerful AI capabilities into their latest platforms. The question is whether your organisation’s data, architecture, and operating model are ready to unlock them — or whether AI will simply become another feature your ERP licence covers but your business never fully uses.
The CIOs who get this right will not be remembered for the platforms they chose. They will be remembered for the transformation they led.
The author is CTO & Innovation Officer, Mastek. Views are personal.