Skip to content

MODULE 7 · LESSON 38 · AI-assisted analysis

Preparing data context for AI

Create a concise data brief that communicates definitions and limitations without exposing unnecessary records.

By Laurence Svekis · Self-paced · All worked data is synthetic unless you supply your own source.

Try the practice ↓

PredictInspectCalculateCheckExplainTransfer

Your starting point

Prerequisites: Use the ideas and vocabulary from lesson 37; revisit its worked examples if needed.

Definition: A data context brief describes the schema, row grain, units, collection period, missing-value conventions and permitted uses. It can support analytical assistance without sharing a complete dataset or sensitive individual records.

A schema is more than column names

A field named score needs its scale, direction and collection method. A date field needs its meaning: order date, delivery date or extraction date. An identifier needs a description of what it identifies. State whether a row is a person, event or aggregate. Without these definitions, an assistant can perform mathematically valid operations on the wrong conceptual quantities.

Minimize what you share

Use synthetic rows or aggregate summaries when they are sufficient for the question. Removing names alone may not protect a record if combinations of dates, locations or rare characteristics identify someone. Follow the data owner’s permissions and relevant organizational rules before using an external service. The local academy workbench does not upload your data; pasting into another tool is a separate decision.

State missingness and boundaries

Explain whether blanks mean unknown, not applicable or not yet recorded. Include important exclusions and known quality issues. Tell the assistant which conclusions the data cannot support, such as causal effects or individual diagnoses. A clear brief reduces ambiguity but does not eliminate model errors, so keep an independent verification plan beside it.

Two worked examples

A useful schema note

“One row is one date-channel pair. sessions counts recorded sessions; conversions counts sessions with at least one target action, so conversions cannot exceed sessions. Dates cover the same synthetic month. No unique-person or satisfaction fields are present.” This prevents several common denominator and interpretation mistakes.

Safer example rows

To ask how to calculate revenue, provide invented rows such as units 4 and price 15 rather than a real customer invoice with name and address. The formula can be explained from the synthetic structure. Keep the actual data local and verify the same operation on an authorized working copy.

Where this becomes useful

A teacher wants help designing a chart for course participation. A fictional cohort table and field definitions can support the discussion without sharing student names, grades or identifiable attendance histories.

Make an attempt

Open Website traffic in the workbench and draft a data context brief in the report notes. Include grain, units, time, conversion definition, missingness and two unsupported conclusions. Replace any actual identifying values with synthetic examples before considering an external AI tool.

Open the accompanying interactive lab →

Misconception check

Claim: AI can infer the correct meaning of every column from its name.

Inspect the correction

Short labels are often ambiguous. Supply a dictionary and verify the interpretation before calculating.

Independent application

Write a reusable data brief for a synthetic table with five fields. Include row grain, units, valid ranges, missingness, provenance and two inference limits. Ask a reader to restate the question using only the brief and revise any ambiguity.

Completion records practice and self-review. It is not certification of statistical competence.