Key takeaways
- Accuracy asks whether a value correctly describes the intended entity at the relevant time.
- A measurement needs a defined field, population and reference standard.
- Report unresolved cases and sample limits rather than hiding them inside a headline percentage.
Overview
Accuracy requires a reference and a test method. A field can match its source perfectly while the source itself is wrong or outdated. Define the entity, observation date and acceptable tolerance before measuring. Estimated revenue ranges should not be tested or presented as if they were audited financial statements.
How it works
Define the fact, time period and acceptable tolerance for each field.
Compare a representative sample with suitable reference evidence.
Report errors and uncertainty by field and segment, then correct root causes.
Establish what the correct answer means
An employee count may refer to a corporate group, a subsidiary or a local office. Before checking accuracy, define the intended scope and date. Otherwise two reviewers can disagree while using different but defensible interpretations. The same applies to current employment, headquarters and industry classification.
Choose a reference appropriate to the field and document its limitations. A current company statement may be useful evidence for a product offering, while a profile page may be stale for employment. If the reference cannot settle the question, mark the case unresolved. Forcing an answer creates a falsely precise accuracy estimate.
| Element | Example definition | Why it matters |
|---|---|---|
| Field | Current employer of a named contact | Prevents combining unlike quality claims |
| Population | Contacts in the selected territory and roles | Limits where the result can be generalized |
| Reference | Dated evidence supporting employment | Explains how correctness was judged |
| Unknown policy | Report unresolved cases separately | Avoids making uncertainty disappear |
Show the denominator and uncertainty
In an illustrative review of 100 returned contacts, 90 are confirmed correct, five incorrect and five unresolved. Accuracy among resolved cases is 90 divided by 95, or about 94.7%. Confirmed-correct coverage of the entire reviewed sample is 90%. Both are useful, but reporting only the larger percentage can hide the unresolved group.
State how the sample was selected and whether it represents the target market. A small convenience sample cannot support a precise claim about millions of records. Review by segment when the workflow spans different geographies or company sizes, because an overall average can conceal where the source performs poorly.
Find the mechanism behind incorrect values
Separate wrong-entity matches, stale observations, source errors and transformation errors. Each requires a different repair. A newer source will not fix a matcher that confuses subsidiaries; better formatting will not fix an obsolete employer. Keep the original value and provenance so the team can trace the error to the stage that introduced it.
After a fix, evaluate new records and a representative regression sample. Do not judge improvement only on the exact examples used to design the correction. Also inspect whether a stricter rule reduced accepted coverage: improving correctness by returning almost nothing may be unsuitable for the workflow even when the accuracy percentage rises.
What this looks like in practice
A contact’s employer was correct when collected six months ago but is wrong today after a job change. The historical record may be accurate for its date while failing a current outreach requirement.
Examples explain the concept; they are not reported customer results.What to check
Inspect sampling, reference reliability, field definitions and confidence intervals when available. Avoid a single percentage that combines easy and difficult fields without explanation.
Common mistake
Treating format validation or agreement between copied sources as independent proof that a value is factually correct.
Data accuracy vs. Data completeness
Accuracy asks whether a value is correct. Completeness asks whether a required value is present. Filling every field with guesses improves neither trustworthy accuracy nor useful quality.
Read the Data completeness definition →Questions answered
What is Data accuracy?
Data accuracy describes how closely a recorded value reflects the real-world fact it is intended to represent at the relevant time and level of detail.
Can accuracy change without changing the stored value?
Yes. The world can change while the database stays the same. That is why observation dates and refresh rules matter.
What is a fair accuracy benchmark?
Specify the fields, sample population, reference method and date. Compare providers or processes using the same criteria and representative records.
Can a valid value be inaccurate?
Yes. An email can have valid syntax while belonging to the wrong person, and a properly formatted address can be obsolete. Validity checks format or rule conformity; accuracy checks the relationship to the real-world fact. A reliable workflow usually needs both.
Should unknown values count as inaccurate?
Define and disclose the policy. Unknown is not the same as confirmed wrong, but it still affects usability and coverage. Reporting correct, incorrect and unresolved groups separately is often clearer than forcing uncertainty into one category. Use the same policy when comparing sources or changes over time.
References and further reading
Primary documentation and source material for this topic. Sources checked September 14, 2026; provider requirements can change.
Continue reading on the blog
Explore all articles and guides →Put the concept to work.
Explore the relevant AstroFabric workflow and see how the pieces connect.
Help keep this guide useful. Suggest a correction or browse the full glossary.