← Back to Course Dashboard
Day 2 · Advanced Automation Track · Lesson 09

Data Extraction & Administration

Course Target: Use AI agents and agentic browsers to automate administrative data work — extracting data from documents, assembling reports and forms, and producing compliance reports — with verification, de-identification, and human sign-off so nothing wrong or private goes out.

Lessons 7–8 built workflows; Lesson 10 connects tools (MCP) and Lesson 11 governs data. This lesson is the practical payoff: automating the repetitive admin tasks — data extraction, form filling, document assembly, compliance reporting — that eat faculty time. The key is an assembly line with checkpoints: extract → verify → assemble → sign off → submit.

Start from your task

A concrete goal for this lessonYou file the same report every term — pulling numbers out of a dozen documents and retyping them into a form. By the end you’ll know how to let AI extract, assemble, and draft that report — safely, with a verification step and your sign-off before anything is submitted.

Learning Objectives

A note on what "able to" means hereThese activities are your first supervised attempts at each skill — a working draft and a repeatable method, not full mastery.

Key Terms

Select any card to flip it. Cards are keyboard-operable — press Enter or Space.

Course Outline & Lesson Modules

If you remember one thing: AI can pull data out of messy documents fast — but it can mis-read or invent values, so every extraction needs a quick check against the source before you use it.

The trick to reliable extraction is asking for structured output: tell the AI to return a table with named columns (or a fixed set of fields), not a paragraph. Structured results are consistent and easy to verify line by line.

A reusable extraction instruction

Paste and adapt this — the "not found, don’t infer" rule is what stops the AI from filling gaps with guesses:

Extract these fields into a table: [field 1], [field 2], [field 3].
One row per record. Keep values exactly as written — do not reformat.
If a field isn't present, write "not found" — do not infer or guess.
Return only the table, plus a column noting the source of each row.

How extraction goes wrong

  • a number is mis-read or transposed (23 becomes 32);
  • a value lands in the wrong column;
  • a field is invented that wasn’t in the source;
  • an OCR/scan error is copied straight through.

Verification is the skill: spot-check a few rows against the original. For high-stakes data, check every row.

Worked example: you ask AI to extract Course, Section, Instructor, and Enrolled from five section records into a table. Before using it, you open two of the source records and confirm the numbers match — and find one where Enrolled was copied from the wrong line.
Quick check: the AI returns a clean table of 40 rows pulled from a scanned PDF. Use it as-is?

No — scans invite OCR errors. Spot-check several rows against the PDF (and every row if the data is consequential) before trusting the table.

Cumulative Check

You want to auto-generate and file a term compliance report from course documents. Name one risk at each stage — extract, assemble, submit — and a control.

Reveal model answer

Extract: mis-read or invented values → verify against the source. Assemble: wrong template/merge or student PII included → use a fixed template and de-identify. Submit: filing an error or violation → require human sign-off and keep an audit trail before anything is submitted.

Hands-on Workspace Modules

Activity 01

Extract & Verify

Extract specified fields from a sample document into a table, then verify against the source — and catch the value that’s wrong.

  • fields extracted into a clean table;
  • rows checked against the source;
  • the bad value caught and corrected.
Open Activity 1
Activity 02

Build the Assembly Line

Lay out the steps for a recurring report, mark where the human checkpoint goes, and identify what must be de-identified.

  • steps in the right order;
  • sign-off before submit;
  • de-identification identified.
Open Activity 2
Activity 03

Compliance Routine Audit

Run the "Should I automate this?" aid on a flawed automation, find what’s missing, and redesign it safely.

  • decision aid applied;
  • flaws identified;
  • safe redesign with sign-off + audit trail.
Open Activity 3

Wrap-up

Reflect (2 minutes)Pick one recurring report or form you handle. Run the "Should I automate this?" aid on it — which steps could AI do, and where does your sign-off belong?
Model what you teach: disclose your AI useWhen AI helped extract or assemble a document, note it — briefly and plainly. A one-line note is enough.
RecapExtract: ask for structured output and verify against the source. Assemble: a small-step assembly line with a template and a sign-off. Compliance: de-identify, keep an audit trail, and never auto-submit consequential work.

Sources & Further Reading

  1. Primary FERPA, 20 U.S.C. § 1232g; U.S. Dept. of Education, Student Privacy Policy Office (studentprivacy.ed.gov) — de-identification and disclosure for any automated student-data work.

Tool features for extraction and form filling change quickly — verify current capabilities and your institution’s approved tools before relying on them.