What is Data Readiness Assessment?

An enterprise AI team plans to build a predictive maintenance model using operational records collected across several production systems. Before development begins, the team needs to understand whether the available data covers the relevant equipment, contains meaningful failure labels and supports the intended prediction task.

In his 2017 position paper on Data Readiness Levels, Neil D. Lawrence argues that project overruns can occur when teams fail to account for the time needed to curate and collate data. The paper outlines three rough stages of data preparedness and suggests they could form a common language for assessing a dataset’s readiness.

A readiness assessment should connect data conditions to the planned application. CUBIG’s AI readiness assessment article describes checking enterprise data on six axes, and notes that whether the data fits a specific task is checked separately. A data readiness assessment answers that separate, narrower question: can this particular data support a defined task?

A practical data readiness assessment commonly examines:

  • Availability: Determine whether the required source data can be accessed and collected.
  • Coverage: Check whether the dataset represents the relevant populations, events and operating conditions.
  • Quality: Examine missing values, duplicates, inconsistencies and potential label problems using data quality criteria.
  • Documentation: Review definitions, collection methods and intended uses. The Datasheets for Datasets paper proposes documenting dataset motivation, composition, collection process and recommended uses.
  • Task suitability: Evaluate whether available features and labels support the intended prediction or analysis.

How is a data readiness assessment different from an AI readiness assessment?

  • A data readiness assessment evaluates whether a particular dataset is suitable for a defined use.
  • An AI readiness assessment examines broader organizational conditions for adopting and implementing AI.
  • Data readiness is one part of an organization’s broader AI readiness, alongside infrastructure, people, processes and governance.

Where does data readiness assessment break in AI work?

A one-time assessment can become outdated when source systems, collection processes or business requirements change. A dataset that was suitable for one prediction task may require additional preparation when the intended outcome or prediction timing changes.

Assessment results also become misleading when general checks stand in for task-specific requirements. A CUBIG analysis of the synthetic UCI AI4I 2020 dataset, published in November 2026, examined 10,000 records with zero missing values and found 27 where the failure label disagreed with the five failure-mode flags. Two kinds of checks help catch issues a completeness check alone would miss: checks tied to the intended task, such as label consistency, and data validation repeated when data changes.

Frequently asked questions

What should a data readiness assessment include?

It commonly examines data availability, coverage, quality, documentation, labels and suitability for the intended AI task. The assessment should also identify preparation work and unresolved data issues.

Who should conduct a data readiness assessment?

Data engineers, data scientists, ML engineers and relevant business stakeholders can contribute. Technical teams evaluate data characteristics, while domain experts help interpret business rules, labels and intended use.

How often should a data readiness assessment be performed?

An assessment should be revisited when source data, collection processes, business requirements or AI tasks change. Repeating the checks shows whether data that was acceptable before is still acceptable.