Why do so many AI initiatives in asset management fail to deliver?

05.11.2026

You have invested in predictive maintenance, analytics or an AI pilot, but the results are slow to show, and trust is slow to build. The model is often not the main problem. This article shows why AI depends on the information foundation underneath it, and where to start if you want AI that your teams actually use.

In brief

An AI tool is only as good as the data behind it. Inconsistent asset data gives unreliable predictions, and people stop using a tool they do not trust.
AI-ready asset information is more than sensor readings. It needs context, a structured asset register, a common reference designation system (RDS) and clear ownership, across systems and the lifecycle.
Start with one use case, map the information it needs and fix that foundation first.

Why AI pilots in asset management stall

Boards and executive teams want to know what AI will do for the performance of their assets. Vendors promise predictive maintenance, risk-based inspections and digital twins. Many asset owners respond with a pilot on a single line, site or asset class.

Gartner, for example, predicted (Opens in new tab) in 2025 that organizations would abandon 60% of AI projects that are not supported by AI-ready data. The technology works, but the foundation it runs on often does not.

How inconsistent data turns into predictions nobody trusts

A predictive maintenance model learns from sensor data and maintenance history. If the sensor tags differ per site, if failure codes are used inconsistently, or if the maintenance history sits half in a maintenance management system (CMMS) and half in Excel, the model learns from a distorted picture. It then produces predictions that experienced technicians quickly recognize as wrong.

Each wrong prediction costs more than the model's accuracy score suggests. It costs trust. Once your maintenance planners stop believing the output, the model stays on the shelf, however advanced it is.

What AI-ready asset information looks like

A vibration reading is data. It becomes useful when you know which pump it came from, where that pump sits in the process, which specification defines the normal range, and when the bearing was last replaced. That context, the metadata, turns data into information. AI needs the information, not just the data.

A common identifier connects the pieces. A reference designation system (RDS) based on ISO/IEC 81346 gives every asset a unique, structured designation that follows it through design, operation and maintenance. When the sensor tag, the CMMS record and the drawing all carry the same designation, a model can link them. When each system names the same pump differently, someone has to match them by hand, and an AI model cannot do it reliably.

The same goes for documents. A drawing is just a file until you know which asset it describes, which revision is approved and who signed it off. A maintenance manual is just a file until it is linked to the right pump model, its spare parts and its inspection intervals. That metadata turns documents into information you can search, trust and act on.

This is why a data lake (a central store for raw data from many sources) alone does not make you AI-ready. What you need is an information foundation that brings together.

Dynamic data, static information and the connection between them

Dynamic data
Such as sensor readings, fault reports and condition scores, integrated continuously so the current picture is available when a decision is made.
Static information
Such as drawings, specifications, certificates and a structured asset register, under version control so everyone works from the approved version.
Connection
One reference designation per asset, so every record and reading points to the same object. That lets you interpret a reading against the specifications and design limits that give it meaning.

The data is already there, but nothing connects it across systems

Most asset owners already hold years of data. The issue is that it sits across PLM, ERP, EAM, CMMS, DMS and spreadsheets, defined differently per location and owned by nobody in particular. Fragmentation makes it impossible to turn the data you have into reliable information.

Adding more sensors or a more powerful model does not fix that. It adds more data to the same unstructured landscape. Structuring, connecting and governing what you already have comes first.

Start with one use case, not the whole landscape

You do not need to fix every system before you start. Start small. Pick a part of your operation that is easy to scope, for example a smaller plant, one process area or a single asset class. Then decide which use case you want to support there, such as predictive maintenance or risk-based inspection. Next, ask five questions:

Which information does this use case need, and in which phase of the asset lifecycle is it created?
Who owns it, and who approves changes?
Where does it live today, and can your teams trust it?
What has to be fixed or connected before they can rely on it?
Which process keeps it up to date, so it still matches reality in the future?

The answers give you a focused, realistic first step. They also show where AI can deliver value soon and where the foundation has to come first.

Cleaning up your information once is not enough. If nobody owns it and there is no routine for keeping it up to date, it will drift away from reality again. So set up the ownership and the process for changes at the same time. That keeps the information accurate and reliable after the project is over. And once it works for one area, you can repeat the same approach for the next.

Where do you start? Assess your asset information first

An AIM Assessment shows how ready your asset information and management of it is for the use cases you care about. We map your current information, review your processes and governance, and run a gap analysis. You get a recommended target state for your asset information, a business case and a roadmap.

Request an AIM Assessment to find out which use cases your data can support today, and what it takes to support the rest.

Get started with an AIM assessment

Know exactly where you stand with your asset information. Fixed price, defined scope, clear deliverables. We map your current state, review your processes and data, and run a gap analysis against ISO 55000/19650 and industry benchmarks. You get a recommended AIM model, a business case, and a roadmap. No obligation beyond that.

About the author

Lina Bäckman

Senior Service Solutions Manager, AIM & Data

About the author

Lina Bäckman

Senior Service Solutions Manager, AIM & Data

I help asset-intensive organizations such as those in energy, process industry and utilities turn information complexity into structured, usable knowledge across the full asset lifecycle, from investment projects through operations and maintenance. At Etteplan I've built and led teams and developed service offerings. My background combines technical engineering with leadership and commercial responsibility, which means I understand both what clients need to solve and how to design solutions that work at scale. Etteplan AIM is built on the insight that asset information only creates value when it serves a clear purpose. Our team brings operational and engineering context to every information management task. We don't just store and organize; we structure information and set up the governance and processes so it's genuinely usable for the decisions and actions your teams need to take.