Decision infrastructure and enterprise AI agents

Importance of Data Quality: Business Cost and Controls

Written by Navdeep Singh Gill | 21 October 2024

Data quality matters because every decision, report and AI model inherits the condition of the data underneath it. Data quality is how well a dataset fits its intended use, judged on accuracy, completeness, consistency, validity, uniqueness and timeliness. When those slip, the cost shows up as rework, wrong decisions, compliance exposure and AI outputs nobody can trust.

This guide is for CDOs, CTOs and data and platform leads who need the business case and a practical order of work. For the fault types themselves, see common data quality issues and how to fix them; for measurement, see data quality metrics.

What does poor data quality cost a business?

Gartner states on its site that poor data quality costs organisations an average of $12.9 million every year (Gartner, "How to improve your data quality"; the page gives no year for the figure, so treat it as an indicative average and not as a forecast for your firm). The costs rarely arrive as one line item. They are spread across several places.

Cost areaHow it shows upWho feels it first
DecisionsPlans and forecasts built on wrong or stale figures, then corrected lateExecutives, finance, strategy
OperationsManual reconciliation, duplicate effort, exceptions that need a personOperations, shared services
Customer experienceWrong account or order details passed to service teamsService, sales, marketing
ComplianceRecords that cannot be shown to be accurate or traceableRisk, legal, audit
AI and analyticsModels and agents that act confidently on flawed inputsData science, platform, product

These are failure modes, not measured statistics. Put your own numbers against them, for example hours of reconciliation per month, returned mail, or the share of records failing a validity rule, before you ask for budget.

Which data quality characteristics should leaders track?

The ISO/IEC 25012 data quality model defines 15 characteristics from inherent and system-dependent views, as summarised by arc42 (secondary source; the standard itself was not read). Most business teams start with six plain ones.

CharacteristicThe question it answersExample failure
AccuracyDoes the value match the real-world fact?A wrong sales figure leads to over-estimated performance
CompletenessAre the required values present?Customer profiles with no contact or purchase history
TimelinessIs the data current enough for the decision?Stock levels based on last month's sales
ValidityDoes it follow the business rules and format?An email field that is not a valid address
UniquenessIs each entity held once?Duplicate customer profiles that receive double marketing
ConsistencyDo systems agree with each other?Different date formats or codes between departments

Each characteristic needs an owner and a rule that can be tested. The data quality metrics guide covers the formulas.

Why does data quality matter more with AI agents?

Dashboards show bad data to a person who may doubt it. Agents and models act on it. The EU AI Act, in the Article 10(3) text published on artificialintelligenceact.eu, says data sets for high-risk systems must be "relevant, sufficiently representative, and to the best extent possible, free of errors and complete" for the intended purpose. Article 10 applies to high-risk AI systems, so check whether your use cases fall under it. For personal data, GDPR Article 5(1)(d) requires data to be accurate and, where necessary, kept up to date, as published at gdpr-info.eu (unofficial text sites; confirm against EUR-Lex for legal use).

A practical consequence is that quality checks need to sit where agents read data, not only in the warehouse. ElixirData Context OS keeps governed context, versioned policy and decision traces in one decision layer above AI agents, and AgentAnalyst lets teams ask questions of enterprise data in plain language and get real-time insights, which makes the quality of the underlying data visible to business users quickly.

What do you gain when data quality is good?

Better quality brings the following benefits.

Better customer records also help repeat business, because service and sales teams work from the same correct view.

Which tools help find and fix data quality problems?

Tools fit different jobs. The table uses each vendor's own description; verify current features and pricing on each site.

ToolWhat it is, according to its siteWhen it fits
OpenRefineA free, open source tool for working with messy data: cleaning, converting between formats and extending data with web servicesOne-off or analyst-led clean-up of a dataset
Great Expectations (GX Core)A Python library for testing, checking and documenting data quality in pipelines; produces Data Docs and supports Actions that stop bad data or notify teams; Apache 2.0 licenceTests inside pipelines and ETL
DeequA library built on Apache Spark for defining unit tests for data, to measure data quality in large datasets (README at github.com/awslabs/deequ)Large Spark datasets
Monte CarloDescribes itself as a platform that monitors, troubleshoots and optimises agents and their underlying data at scaleMonitoring data and AI pipelines in production
Qlik Talend CloudOffers automated profiling and quality rules with data stewards, and AI agents that recommend fixes which people verifyManaged profiling and rule enforcement

Sources: openrefine.org, greatexpectations.io, github.com/awslabs/deequ, montecarlo.ai, qlik.com, read 12 Nov 2026.

Where does poor data quality hurt most, by industry?

These are typical exposure points, not case studies or measured results.

IndustryTypical exposureControl to put first
Retail and e-commerceDuplicate or incomplete customer profiles skew segments and marketing spendUniqueness and completeness rules on customer data
Sales operationsWrong contact details waste outreach effortValidity checks and scheduled cleansing of contact lists
Financial servicesCustomer records that differ between departments complicate complianceOne governed customer record and clear ownership
HealthcareDuplicate patient records add time and risk to care coordinationUnique identifiers and regular audits

How do you improve data quality in practice?

Gartner's guidance on the same page recommends linking data quality issues to business KPIs and revenue, agreeing with business stakeholders what good enough means, profiling data early and repeatedly, assigning data stewards clear responsibilities, and adding data quality checks as a stage gate in release management. A sequence that follows this:

1. Tie quality to a business outcome

Pick one decision or process, such as customer onboarding or monthly forecasting, and agree which characteristics matter for it.

2. Name owners and set policy

Give each critical dataset a steward and an escalation route. The open and unified metadata platform guide explains how a metadata layer shows lineage and ownership, and enterprise data catalog tools compares catalogues for finding and certifying data.

3. Test close to the source

Profile the data, then write rules as tests in pipelines so failures stop or flag bad data before it spreads. Audit regularly and cleanse duplicates and stale records.

4. Make ownership domain-based where it fits

In larger organisations, domain teams can own the quality of their data products. Read whether to adopt data mesh before moving to that model, because it needs team capacity and clear contracts.

5. Train people and close the feedback loop

Staff who enter data need to know how poor input affects downstream results, and users need a simple way to report issues, which then feed back into rules.

When is a data quality programme not the right first step?

If no one owns the data, fix ownership first, because rules without owners go unmaintained. If the data is used by few decisions, start with the highest-value process and not the whole estate. And if the source system keeps creating bad data, fix the source before adding more downstream checks.

For wider reading, see data quality management, master data management architecture and best practices, master data management in banking and data management with intelligent data agents.

To map one decision your agents make, and the data quality, context and policy it depends on, book a Context Workshop. See what happens in the workshop.

FAQ

Why does data quality matter for business?

Decisions, operations, customer service, compliance and AI all depend on the data they use. Gartner states that poor data quality costs organisations an average of $12.9 million a year.

What are the main characteristics of data quality?

Common ones are accuracy, completeness, timeliness, validity, uniqueness and consistency. ISO/IEC 25012 defines 15 characteristics, as summarised by arc42.

Does the EU AI Act mention data quality?

Article 10(3), as published on artificialintelligenceact.eu, says data sets for high-risk AI systems must be relevant, sufficiently representative, and to the best extent possible free of errors and complete.

Which tools can test data quality?

OpenRefine for cleaning, Great Expectations and Deequ for tests in pipelines, Monte Carlo for monitoring and Qlik Talend Cloud for profiling and quality rules, each according to its own site.

Where should a data quality programme start?

Start with one high-value decision, name owners, agree which characteristics matter and add tests close to the source.