Introduction: AI is not an "add-on" to ERP, but the next-generation operations control panel
In the past few years, when companies talked about ERP, it was more like "deploying a system"; but the news and cases from the past week remind us that AI is pushing ERP from "bookkeeping and processes" toward "real-time operations." On one hand, vendors emphasize in their financial reports that cloud and AI are driving growth and accelerating product roadmaps; on the other hand, there are warnings about long-term loss of control in public sector ERP projects—technology is not scarce, what is scarce is data and governance, as well as the ability to integrate AI into a closed loop.
This article uses four observation points to sort out trends and provides a ready-to-implement checklist: from "selecting scenarios" to "controlling risks," turning ERP+AI into measurable efficiency and profit improvements.
Key Points (Observations from the Past Week)
-
Vendor level: Cloud + AI continues to be the main axis of ERP competition.
Recent interpretations of SAP's financial reports and industry analyses repeatedly mention that cloud business and AI narratives are influencing orders, product cadence, and customer expectation management. For enterprise users, this means that "AI capabilities" will increasingly enter the core ERP processes through subscriptions, platform capabilities, and embedded Copilot/Agent features. -
Risk level: The root cause of ERP failure is often governance and change, not technology selection.
The Register's ongoing coverage of a public sector Oracle ERP project once again shines a spotlight on typical risks such as "budget overruns, scope creep, master data chaos, inconsistent processes, and systems still unusable after go-live." After AI enters ERP, these issues will not disappear but will instead be amplified: if the data is wrong, AI will also "confidently be wrong." -
Assets and Manufacturing: EAM/Manufacturing scenarios are more likely to achieve quantifiable AI returns first.
Reports from Hitachi/Microsoft show that EAM (Enterprise Asset Management) and critical infrastructure operations are using AI for predictive maintenance, work order recommendations, knowledge retrieval, etc. The reason is practical: KPIs are clear (downtime, spare parts, MTTR/MTBF), the data chain is relatively closed-loop, and ROI is easier to measure. -
Application Modernization: AI Implementation is Bringing "Application Modernization/Data Platform" Back to Priority.
Vendors like NTT have recently emphasized that to scale AI within enterprises, the modernization of applications and data must first be addressed (observability, API-fication, data availability, and controllable permissions). For ERP, this often corresponds to: master data governance, interfaces and event buses, data domain models, and more controllable permissions and auditing.
Implementation Recommendations: An "ERP + AI" Roadmap (Ready to Use)
1) First select 3 types of high-win-rate scenarios (avoid idling)
- Query/Interpretation (Low Risk): "Natural language data query + indicator explanation" for finance, procurement, inventory, and projects. Goal: Reduce report back-and-forth and metric disputes.
- Advisory/Decision-support (Medium Risk): Procurement inquiry suggestions, inventory replenishment suggestions, accounts payable reconciliation anomaly alerts, expense compliance alerts. Goal: Improve first-pass rate and anomaly detection rate.
- Execution/Automation Type (High Risk): Automatically generate work orders, automatically generate accounting entry drafts, automatically create supplier communication emails, etc. It is recommended to first operate in a closed loop of "draft + approval".
2) Deliver data and process governance as part of the AI project
- Master Data Minimum Set: Customer/Supplier/Material/Account/Organization/Cost Center/Asset, etc., first define the scope, responsible person, and change process.
- Unified Process Standards: Standardize at least 80% of the "core processes" (procure-to-pay, order-to-cash, plan-to-produce, asset-to-work order).
- Data Quality KPI: Missing rate, Duplicate rate, Reconciliation discrepancy rate, Abnormal work order rate; without KPIs, there is no continuous improvement.
3) Design an "auditable" closed loop for AI (avoid black boxes)
- Prompts and Output Traceability: Record model version, prompt summary, and cited data sources (at minimum, table/field/document number).
- Permissions and Desensitization: AI can only "see" data that users are authorized to view; desensitize or partition personal information, salaries, contracts, etc.
- Human-machine collaboration default: Key actions (payment, shipment, journal entry posting) default to "AI-generated draft + human review + automatic verification".
4) Advance in a 30/60/90-day rhythm (with deliverables for each phase)
- 0-30 days: Select scenarios and lay the foundation: Determine 2 business domains, complete data scope and permission sorting, and create the first usable Demo (natural language query).
- 31-60 days: Closed-loop trial operation: Launch the "suggestion/draft" capability; introduce anomaly detection and audit logs; establish weekly reviews.
- 61-90 days: Scaling and risk control: Expand to more organizations/factories/branches; improve monitoring, cost accounting, canary release, and rollback mechanisms.
Summary: AI will make ERP "smarter" and governance more "rigid"
The essential value of ERP is to connect a company's capital flow, material flow, and information flow into a controllable operational closed loop. The addition of AI makes the process of "detecting anomalies—explaining causes—providing suggestions—generating actions" faster, but the prerequisite is: reliable data, unified processes, controllable permissions, and auditable outputs.
If you are planning ERP+AI: don't start with "buying models/plugins," but with "selecting scenarios + governance foundation + closed-loop design." This way, you can not only gain efficiency benefits but also avoid repeatedly stepping into pitfalls in complex organizations during ERP projects.
