Data literacy: questions to ask before using a number
By Fabrice PAYETPublished Updated
Data literacy means being able to question a number before using it. You do not need to build every model yourself, but you should understand what was counted, what was left out and how strongly the evidence supports a decision.
Ask what the metric measures
For an illustrative example, a team says revenue rose after a campaign. Ask whether the measure is booked revenue, cash collected or recurring revenue. Check currency, refunds, time period and whether the comparison includes the same customers.
Write the definition where the number is used. If two dashboards disagree, compare definitions and data coverage before deciding that one system is broken. A familiar label such as “active user” can hide materially different populations.
Check coverage and denominators
A percentage needs a numerator and a denominator. “Conversion improved” means little until you know which visitors were eligible, which event counted as conversion and whether tracking changed. Small samples and missing records can make a confident chart misleading.
Check whether the dataset represents the people affected by the decision. Survey respondents, existing customers and the entire market are different groups. Do not extend a finding beyond the source population without an explicit reason.
Distinguish an observation from a cause
A change after a launch does not by itself prove the launch caused it. Seasonality, pricing, channel mix and simultaneous changes can also explain the difference. Ask whether an experiment or another suitable design supports the causal claim.
When someone presents a model score, ask what test data was used, how it was separated from training and which errors matter. Accuracy can conceal poor performance on a minority class. A useful explanation connects the metric to the decision and its error costs.
Read charts and AI summaries critically
Check axes, units, missing dates and aggregation. A cropped axis can magnify a small change; a cumulative series can rise while daily activity falls. Ask for the underlying values when a visual leaves the scale unclear.
An AI-generated summary is another claim to verify. Trace quoted numbers to the actual source, check calculations and ask whether the stated limitation appears in the data. Fluent prose does not establish that a file was read correctly.
Practice with a decision note
Choose a report your team already uses. Write a short note with its definition, data period, coverage, known limitation and decision. Ask a colleague to identify a question that could change the conclusion. This is a suggested exercise, not evidence of measured business improvement.
For hiring, assess whether a candidate can explain the same dataset to a technical peer and a decision-maker. Vocabulary recall is less useful than showing how they identify an unsupported conclusion and choose the next check.
Sources and next steps
Sources checked on October 2, 2026. Follow the linked documentation for current requirements.
PostgreSQL tutorial: queries, joins and transactions
