Document Synthesis and Data-Driven Decisions

The last lesson had you asking plain-language questions of one messy spreadsheet, and a finding came back, no code needed. But real decisions rarely live in one file. You must choose by Friday (renew the vendor or switch) and the evidence sits in nine PDFs you have no time to read. So you skim, and decide on a hunch.

Why This Matters

This lesson closes the tools pillar by turning scattered evidence into a decision you can defend, the pAIRI Data Evaluation and Data Proficiency skill of reading data well enough to stake a decision on it. By the end, you will be able to:

• Synthesise findings across many documents at once using NotebookLM
• Challenge an AI-assisted recommendation by making it argue the opposing case
• Justify a business decision with evidence you can trace to its source

Synthesise, then decide

NotebookLM works differently from a general chatbot. It answers only from the documents you upload, and cites the exact passage behind every claim. That changes the trust question: you are no longer asking whether the AI invented something; you are checking whether the source really says what the answer claims. Upload the reports, the proposals and the market study, then ask one question across all of them at once (where do these proposals disagree on three-year cost?) and the answer returns with citations you can open.

Audio Overviews turn those same sources into a spoken discussion between two AI hosts. They are useful for hearing the shape of an argument on a commute, though you still check the written source before you cite anything from it.

But synthesis is not yet a decision. The evaluate skill covers the final two steps. Ask the tool to build the strongest possible case against the option you favour; a recommendation that survives its own counter-argument is one you can defend. And notice what the tool cannot see: what your documents leave out is yours to catch.

Key Takeaway

You have finished the tools pillar. You can now draft, analyse and synthesise across a full decision, and the call on what the data omits stays yours.