Agile InsightsBriefing
A closed reference volume, a steel measuring caliper and a pair of headphones on a dark desk, labelled ISO/IEC 5259, Reference Series and Reference 10. A yellow bar along the bottom reads Reference series, number 10.

Reference 10 · Primer · About 5 minutes

ISO/IEC 5259: data quality for analytics and machine learning

The standard series that turns 'our data is good enough for AI' from a claim into a measurement.

#data

Download the MP3 · Leonard, an AI voice, reads a script written from the sources below.

In this episode

01A yardstick, not a certificateISO/IEC 5259 is a family of voluntary standards titled Data quality for analytics and machine learning. Nothing in it is mandatory and you do not certify against it. You use it as a shared yardstick.
02Catching weak data earlierA model learns from its data. If ten years of lending history leaves out a group of customers, a credit model will quietly fail them. The series gives common terms and tests to catch that earlier.
03Quality you can measureQuality is a set of named, measurable characteristics, not a feeling. Part 2 supplies the measures for machine learning. Quality also needs a management system with owners, requirements and records, and someone senior to oversee it.
04What is published, and the gap filledAs at October 2026 all five parts are published. ISO/IEC 42001 has controls for data quality in AI systems that say you must do it, but not how to measure it. This series fills that gap.
05What to ask your teams forList the datasets behind your AI systems, with a named owner for each. Measure each against a small set of quality characteristics from Part 2. Ask for the measurement to repeat, because data drifts after training.
Sources and links (8)
Part 1: overview and terminology · ISO catalogue page
ISO catalogue entry, published July 2024.
Part 2: data quality measures · ISO catalogue page
ISO catalogue entry, published November 2024.
Part 3: management requirements and guidelines · ISO catalogue page
ISO catalogue entry, published July 2024.
Part 5: data quality governance framework · ISO catalogue page
ISO catalogue entry, published February 2025.
Part 6: visualization framework (technical report) · IEC webstore
IEC webstore entry, published May 2026.
Australian Standards listings · Standards Australia store
Standards Australia store, where the adopted parts are sold.
ISO/IEC 25024: measurement of data quality · IEC webstore
The 2015 measures standard that sits beside the 25012 data quality model.
ISO 42001 Annex A.7: data for AI systems · Explainer
A practitioner explainer on the data controls in the AI management system standard.

Written from public sources and checked on 10 October 2026. The standard is paraphrased, not reproduced. Check the current text before relying on it.

How this maps to what we do

AI is downstream of data. These frameworks describe the data capability underneath trustworthy AI.

Standard or rule
What an organisation needs
Where we help
An honest read of data management capability against the use cases planned, with the gaps that matter first.
Data Foundation ReadinessJust-in-Time Data Governance
ISO/IEC 5259: data quality for machine learning
Data quality measured for the purpose it will serve in AI, with lineage and documentation.
Data Foundation ReadinessJust-in-Time Data Governance

These are Agile Insights AI Strategy & Governance offerings.

Want to know more?

Sign up for our podcast at agile-insights.com.au.

Read the transcript

Leonard Most organisations say their data is good enough for AI, and almost none can prove it. I S O, I E C, 5259 is the international standard series that turns that claim into something you can measure and govern. By the end of this episode, you will know what the series contains, which parts are published, and what to ask your teams for.

Leonard First, what it is. I S O, I E C, 5259 is a series published by the International Organization for Standardization and the International Electrotechnical Commission. It is titled Data quality for analytics and machine learning. It is a family of voluntary standards. Nothing in it is mandatory, and you do not certify against it the way you do against the AI management system standard. You use it as a shared yardstick.

Leonard Here is the problem it solves. A model learns from its data, so weak data means a weak model. Say a lender trains a credit model on ten years of approvals. If the history leaves out a group of customers, the model will quietly fail them. Nobody sees it until a complaint arrives. The series gives you common terms and tests to catch that earlier.

Leonard Who uses it? Data teams use the measures. A data quality lead uses the management requirements. Engineers use the process guidance. Boards and chief data officers use the governance part. It comes up in procurement, when a vendor says its training data is high quality. It comes up in audits, and when a regulator asks how you know your data suits a high risk system.

Leonard Now the core ideas. One: quality is a set of named, measurable characteristics, not a feeling. The series builds on I S O, I E C, 25012, the long standing data quality model, and adds characteristics that matter for machine learning. Two: measurement needs a method. I S O, I E C, 25024 supplies measures for the older model, and Part 2 of this series does that job for machine learning. Three: quality needs management, meaning a system with owners, requirements and records. Four: someone senior must oversee all of it, which is the governance part.

Leonard Here are the parts, and what is published as at October 2026. Part 1, overview and terminology, was published in July 2024. Part 2, data quality measures, in November 2024. Part 3, management requirements and guidance, in July 2024. Part 4, the process framework, in July 2024. Part 5, the governance framework, in February 2025. So all five parts are published. There is also a Part 6, a technical report on visualising data quality results. It is guidance, not a standard, and the I E C store lists it as published in May 2026.

Leonard How does it connect to the rest of the stack? At the data foundations layer, it sits beside the D A M A data management body of knowledge. That is the broad discipline, and this series sharpens its data quality area for machine learning. At the AI governance layer, the I S O, I E C, 42001 management system standard has controls for data in AI systems, including data quality and where data came from. Those controls say you must do it. They do not say how to measure it. That is the gap this series fills. For local rules, the same measures give you evidence when a regulator asks about your data.

Leonard Getting started takes three steps. First, list the datasets behind your AI systems, with a named owner for each. Second, choose a small set of quality characteristics from Part 2, and measure each dataset against them. Third, ask for the measurement to repeat. Data drifts after training, so a one off assessment is a warning sign, not a comfort. For learning more, read Part 1 first, then Part 5 if you sit on a board. Links are in the show notes.

Leonard A word on currency, as at October 2026. Parts 1 to 5 are published first editions. Standards Australia has adopted Part 4 as an Australian Standard in September 2024 and Part 5 in May 2025. I did not confirm Australian adoption of Parts 1 to 3, so check the store. The standards are paid documents, so read the current text before you rely on any detail I have given.

Leonard In short, this series makes data quality for AI something you can measure, manage and govern, and the question to ask is how often your teams re run the measures. If you'd like to know more, sign up for our podcast at agile-insights.com.au.

More on #data

Reference 15 · PrimerThe quiet obligations: records, data ethics, digital experience and cloudEpisode 30 · PrimerThe prompt that shows what your AI actually knows about you (and it’s less than you might think)Episode 28 · Primer and discussionWhen an AI keeps the wiki, checking has to keep pace with writing.Episode 24 · PrimerRemember, retrieve, recheck, forget: governing agent memory.Episode 18 · Primer and discussionWhen agents share one memory, memory is a data flow to govern.Episode 9 · PrimerYour company brain is a copy. Govern it like one.

Every episode tagged #data

All of the reference series (16) · Follow AI Governance Reference in Apple Podcasts or any podcast app

Everything from Agile Insights: briefings, Making You Think and the reference series