A development engineer asks an AI a simple question: Which tests from the past two years showed anomalies similar to those found in the current measurement?
The data exists. Measurement files are stored on fileservers, test orders in databases, and vehicle and component information in separate systems. Test results appear in reports, while issues are documented in issue-tracking tools.
Yet the AI cannot provide a reliable answer.
The problem isn't necessarily the AI model. It's the missing connections between the data.
This raises a fundamental question: Does the AI actually understand what the available data means?
A Measurement Value Doesn't Explain Its Own Context
Consider a temperature reading of 118 degrees Celsius.
On its own, this value tells us very little. Does it come from a combustion engine, an electric motor, or a test bench? Under what load was it recorded? Which software version was installed? What sensor was used?
Experienced test engineers can often reconstruct these relationships. They understand naming conventions, recognize directory structures, and remember previous tests.
AI doesn't have this accumulated knowledge.
To interpret measurement data correctly, machines need metadata. More importantly, they need to understand how that information is connected.
A measurement dataset should reveal which test order it belongs to, what test object was used, which configuration was applied, and under what conditions the test took place.
These connections give the data its technical meaning.
The FAIR principles have emphasized this for years: Data should remain reusable and machine-readable through metadata, provenance information, and links to related datasets.
More Data Doesn't Automatically Mean Better Data
Vehicle development generates enormous amounts of data.
Test benches produce measurement series, prototype vehicles record sensor readings, and simulations generate additional results. Calibration data, diagnostics, reports, and issue records add to the volume.
The real challenge isn't the amount of data. It's how widely that information is scattered.
Data lives in different applications, on network drives, or simply in employees' heads. Some relationships can only be reconstructed through filenames or IDs.
For AI, this becomes a serious obstacle.
Language models can search documents and summarize their contents. Machine-learning models can identify patterns. But both depend on data whose meaning and origin can be understood.
Without that context, AI may compare measurements that aren't technically comparable or combine results from different development stages.
Adding more data won't necessarily solve the problem. It may even make it harder to identify.
Connections Turn Test Data Into Useful Knowledge
In Automotive Testing, every result is connected to numerous pieces of information.
A test order relates to a specific requirement. The test is performed on a vehicle or component at a defined development stage. A test specification describes the procedure. Sensors generate raw data, which is then analyzed and evaluated. Any anomalies may lead to issue reports.
Often, this information is spread across multiple systems.
Industrial AI therefore needs more than access to measurement files. It must also understand the technical relationships behind them.
This is where ASAM ODS comes in. The standard supports structured test and measurement data management by linking measurements to their descriptive information.
In a 2026 whitepaper, ASAM highlights how a systematic data and metadata foundation improves the quality, traceability, and explainability of data used by AI tools.
Rather than choosing an AI model first, companies should identify which technical relationships that model will need to understand.
Knowing Where a Result Comes From Matters
A plausible AI-generated answer is rarely enough for engineering decisions.
If a system reports that similar anomalies occurred in earlier tests, the engineer needs to know exactly which tests those were.
Vehicle development stages, test conditions, and supporting measurement data then become essential.
This makes data provenance a key requirement for industrial AI.
The National Institute of Standards and Technologies also emphasizes that information governance in Smart Manufacturing supports consistent, repeatable, and trustworthy data use.
A useful AI system should therefore guide engineers directly to the original information.
Instead of simply reporting similar anomalies, it could identify five comparable tests, highlight three performed with the same software version, and provide links to the relevant measurement files and issue reports.
That creates a traceable basis for technical decisions.
AI-Ready Test Data Starts Before AI Deployment
This is where modern test management platforms play an important role.
Cluu Assisted Testing brings information from different systems together within a shared technical context. Test orders, test objects, specifications, measurement data, results, and issues can be linked without replacing every existing application.
The underlying approach uses a virtual graph. Relationships become part of the data model rather than leaving information stored in isolation.
During data ingestion, measurement files can been riched with information from test planning, vehicle records, or asset data. Parsers extract information from different sources. Changes remain traceable, while interfaces enable programmatic access.
This foundation can support various AI applications: intelligent test searches, identification of recurring anomalies, technical question-answering systems, and models for pattern recognition or prediction.
All depend on one thing: reliable connections between measurement data and its technical context.
Being AI-Ready Starts With Being Test-Data-Ready
The AI boom has drawn attention to chatbots, copilots, and agents. In industrial environments, however, their value often depends on the less visible data foundation beneath them.
Data must be discoverable, terminology must be consistent, and development stages must remain distinguishable. Results need traceable origins, and measurements need their test context.
Companies that structure and connect their test data today are doing more than improving data management. They're building the foundation that will allow future AI systems to access the engineering knowledge accumulated throughout development.
Even the most powerful AI can only recognize patterns in data whose meaning it can understand.
The most valuable resource for industrial AI isn't simply more data. It's data with context.
