AI in Asset Management Starts with Trusted Master Data

Reliability, predictive maintenance, and agentic asset operations all depend on one trusted digital identity for every physical asset.

Asset-intensive organizations are investing in sensors, IoT, Asset Performance Management (APM), digital twins, and AI. The return on all of it depends on a basic question: can the organization trust the digital identity and context of its physical assets?

The wider AI market is hitting the same wall. McKinsey reports that only about 7% of companies have fully scaled AI, and more than two-thirds of high performers name data as their primary obstacle (McKinsey & Company, 2026).

Abundant data, but no single trusted asset identity

Oil and gas, utilities, mining and manufacturing firms hold asset information across ERP and enterprise asset management (EAM) platforms, engineering systems, GIS, SCADA/DCS, historians, MES, OEM documents and condition monitoring. In SAP S/4HANA Asset Management, functional locations and equipment anchor this data (SAP SE, 2026b), but a single pump can also carry an engineering tag, an OEM serial number, a DCS tag, a historian point, and an APM ID. Each record is correct in its own system. Together, they are not one trusted asset.

Organizational silos compound the problem: in McKinsey’s master data survey, 80% of organizations said some divisions operate in silos (McKinsey & Company, 2024). Standards are catching up: ISO 55013:2024 now provides specific guidance on managing data to support asset management objectives (ISO, 2024).

EXHIBIT 1  |  ONE CENTRIFUGAL PUMP, NINE SYSTEM IDENTITIES

Illustrative identifiers. Highlighted rows show where the digital picture breaks.

Poor master data becomes an operational problem.

These gaps do not stay inside the database. An incorrect hierarchy hides the right technical object; an incomplete bill of materials (BOM) obscures the right spare part; duplicates split maintenance history; and unreliable criticality distorts priorities.

The chain runs from poor data to more search and verification, lower planning confidence, weaker reliability insight, and, ultimately, analytics and AI that people will not trust. The ERP stays the system of record, but stops being the system people trust.

AI can make governance continuous, with people in control

Cleansing campaigns temporarily restore quality, but decay returns as assets move and systems change. McKinsey notes that governance built for periodic review struggles when data is created and consumed continuously (McKinsey & Company, 2026).

An AI-enabled capability can instead continuously sense, detect, reason, and recommend: finding duplicates, inconsistent classes, missing attributes, hierarchy anomalies, and cross-system conflicts, and explaining the evidence behind each finding. Accountable engineers and data stewards approve significant changes, which then run through governed workflows. AI accelerates the analysis, while people retain control of critical decisions.

EXHIBIT 2  |  THE CLOSED LOOP, WITH A HUMAN APPROVAL GATE

The outcome: asset-data trust, continuously improved.

In practice, the same seven problems recur across EAM landscapes. In each case, AI does the investigating while people remain accountable for the decision.

EXHIBIT 3  |  SEVEN RECURRING EAM DATA PROBLEMS WHERE AI INVESTIGATES, AND PEOPLE STAY ACCOUNTABLE

What it looks like across four industries

Oil and gas: a common reliability language. AI maps COMP-101, K101, PROCESS GAS COMPRESSOR 1 and PG-COMP-A to one ISO 14224 equipment class (ISO, 2016) and flags duplicates and missing attributes; an engineer approves the result. McKinsey cites a manufacturer that used generative AI to clean thousands of failure mode and effects analysis (FMEA) lines across sites (McKinsey & Company, 2025).

Utilities: one identity across EAM, GIS and SCADA. Entity matching links SAP TR-44018, GIS XFMR-44018, and a serial-numbered condition record, flagging rating conflicts for the data steward, in line with IEC 61968-4 and the Common Information Model (CIM) (IEC, 2019, 2022).

Mining: protecting component genealogy. When one final drive (FD-3317, illustrative) appears on two haul trucks at once, AI weighs serials, install records, and work orders, then proposes the likely correct installation with a confidence level, so failures are analyzed against the component itself.

Manufacturing: bridging ERP, MES and operational technology (OT). Semantic matching proposes that FILLER_01 (SAP EAM), FL01 (MES), PLC04_LINE2_FILLER (SCADA), and PKG-L02-FL-001 (engineering) represent one machine, integrating maintenance, production, and signal data in line with ISA-95 (ISA, 2025).

SAP supplies building blocks; the assistant is a solution pattern

At Sapphire 2026, SAP introduced an Autonomous Asset Management scenario in which agents analyze past incidents, infer root causes, and prefill work orders (SAP SE, 2026c). Impressive as they are, agents like these inherit the quality of the master data they read.

“For the mission-critical processes of our customers, ‘almost right’ just isn’t good enough.”

Christian Klein, CEO, SAP SE (SAP SE, 2026c)

EXHIBIT 4  |  SAP BUILDING BLOCKS RELEVANT TO ASSET MASTER DATA, AS OF SEPTEMBER 2026

Autonomous must never mean uncontrolled; start small

Following the NIST AI Risk Management Framework (NIST, 2023), governed AI combines a recommendation with five controls (Exhibit 5). Changes to hierarchy, specifications, criticality, safety data, or asset identity stay under business control.

EXHIBIT 5  |  GOVERNED AI FOR ASSET MASTER DATA IS A RECOMMENDATION PLUS FIVE CONTROLS

AI recommendation + Evidence + Confidence level + Role-based authorization + Human approval where required + Audit trail = Governed AI

Start narrow and define “good enough” data per use case (McKinsey & Company, 2026):

One site → One asset class → One data problem → One measurable outcome  |  Prove, then scale

Before scaling, baseline completeness, duplicate rate, classification consistency, hierarchy accuracy, and cross-system match rate. The question then shifts from “How do we clean 500,000 records?” to “Which asset-data problems block better reliability and maintenance decisions?”

Five moves for leadership teams

1. Name an owner for asset identity across engineering, operations, and maintenance, not IT alone.
2. Choose the decision before the dataset, such as bad-actor ranking or work-order priority.
3. Define “good enough” and measure it per asset class, reported alongside operational KPIs.
4. Design approval gates and audit trails from day one.
5. Build on the SAP roadmap, with MDG for EAM as the system of control and Joule agents as consumers.

The destination is intelligent asset operations.

Every future EAM capability, from reliability risk to agent-driven work orders, depends on answering one question: which physical asset does this information belong to?

EXHIBIT 6  |  THE JOURNEY RUNS THROUGH TRUSTED DATA BEFORE IT REACHES INTELLIGENT OPERATIONS

Trusted asset data → Connected asset context → AI-assisted decisions → Intelligent asset operations

AI does not remove the need for EAM master data governance. It makes good governance more valuable.

• SAP
• SAP AI
• SAP S/4HANA