How to clean Maximo asset data and failure codes for AI
The short answer
Start with the data AI depends on most: a consistent asset hierarchy, classifications on critical assets, and valid problem, cause and remedy codes on closed work orders. Measure the share of closed work orders with valid codes, fix the failure-code hierarchy, back-fill high-value history, and make the codes mandatory at close-out so the data stays clean.
Five signs your Maximo data is not AI-ready
- A large share of closed corrective work orders have no failure code, or only a problem code.
- The same failure is coded five different ways, or recorded only in free text.
- Assets sit at the wrong level of the hierarchy, or critical assets have no classification.
- Duplicate assets and locations exist after past migrations or site mergers.
- Meter readings are missing, irregular or recorded against the wrong asset.
Baseline your Maximo data before you fix it
Pick a handful of metrics you can calculate from Maximo today and track them monthly:
| Metric | How to calculate it |
|---|---|
| Failure data quality | Closed corrective work orders with valid problem, cause and remedy codes ÷ all closed corrective work orders |
| Classification coverage | Critical assets with a classification ÷ all critical assets |
| Hierarchy integrity | Assets with a valid parent and location ÷ all active assets |
| Duplicate rate | Suspected duplicate assets ÷ all active assets |
Fix the failure-code hierarchy and structures first
- Failure-code hierarchy. Rationalise it to a short, meaningful list per asset class. Industry taxonomies such as ISO 14224 are a useful starting point.
- Classifications. Apply them to critical assets first, with the attributes reliability analysis actually needs.
- Hierarchy. Fix parent-child relationships and merge duplicates, keeping history attached to the surviving record.
Back-fill the history that matters
You do not need to recode ten years of work orders. Focus on critical asset classes and the last two to three years of corrective work. Language models are now good at suggesting codes from free-text descriptions and technician notes, with a reliability engineer reviewing the suggestions in batches.
Keep failure data clean at the source
- Make problem, cause and remedy mandatory at close-out for corrective work.
- Give technicians short, asset-specific picklists on mobile instead of long generic lists.
- Review the metrics monthly and fix drift early.
This is exactly the job of the data steward and technician assistant in Perimo Asset Agents: fixing codes, duplicates and hierarchy gaps continuously, and capturing clean failure data at close-out.